@earendil-works/pi-coding-agent
1.6.0 · 20 Nov 2026MIT6.6M / weekRepositoryHomepagenpmAtom feed
65 stable releases, newest first, by release line. Notes are quoted as the maintainers published them.
- 1.x · latest line
1.6.0
20 Nov 2026- Breaking
Extension components:
render()now returnsreadonly string, andContainerreturns the same cached array while its children are unchanged. Extensions that modify render output in place, includingsuper.render()in aContainersubclass or an embedded component's output, must copy it first - Security
Updated
undicito 8.12.4 for security advisories GHSA-mqg3-3v4r-cm32, GHSA-7xg3-fphp-28hg, GHSA-5rm8-gv32-r954, GHSA-9v6h-8rgp-m69f, GHSA-3g9m-rv3r-w5w2, and GHSA-r4pc-4jcp-7pqc - Breaking
Removed the retired
deepseek-v3.2anddeepseek-v4-flashfrom the Qwen Token Plan Team providers; usedeepseek-v4-flash-0731ordeepseek-v4.1-flash
New Features
GPT-6.2 Astra and GPT-6.2 Sol: both models are available on the
openai,azure, andopenai-codexproviders, and GPT-6.2 Sol is now the default model for all three. See Providers.
Breaking Changes
- Breaking
Extension components:
render()now returnsreadonly string, andContainerreturns the same cached array while its children are unchanged. Extensions that modify render output in place, includingsuper.render()in aContainersubclass or an embedded component's output, must copy it first
Added
Added GPT-6.2 Astra and GPT-6.2 Sol to the
openai,azure, andopenai-codexproviders. Both reject reasoning effortnoneandminimal, so thinking leveloffis unavailable. Azure serves them with a 1.05M context window; Codex serves GPT-6.2 Sol with 500K and GPT-6.2 Astra with 272KAdded
qwen4-max-previewto the Qwen Token Plan Individual providersAdded
deepseek-v4.1-flashanddeepseek-v4.1-flash-1017to the Qwen Token Plan Individual providers, and per-plan key variablesQWEN_TOKEN_PLAN_TEAM_API_KEY,QWEN_TOKEN_PLAN_TEAM_CN_API_KEY,QWEN_TOKEN_PLAN_INDIVIDUAL_API_KEY, andQWEN_TOKEN_PLAN_INDIVIDUAL_CN_API_KEY; the sharedQWEN_TOKEN_PLAN_API_KEYandQWEN_TOKEN_PLAN_CN_API_KEYstill workAdded a transcript notice when OpenAI drops reasoning items from a request, such as
OpenAI dropped 2 reasoning items: prefix_mismatch (details in session)after compaction on GPT-6.2. Like the Anthropic thinking notice, it is shown whenshowCacheMissNoticesis on and only when more items are dropped than in the previous response. Both notices now name the reasons
Changed
Qwen Token Plan providers now use the current gateways (
token-plan.maas.qwencloudapi.cominternational,token-plan.maas.qianwenaiapi.comChina), andqwen-token-planandqwen-token-plan-cnare named "Qwen Token Plan Team" and "Qwen Token Plan Team CN". Provider IDs are unchanged, and the oldaliyuncs.comhosts still work- Security
Updated
undicito 8.12.4 for security advisories GHSA-mqg3-3v4r-cm32, GHSA-7xg3-fphp-28hg, GHSA-5rm8-gv32-r954, GHSA-9v6h-8rgp-m69f, GHSA-3g9m-rv3r-w5w2, and GHSA-r4pc-4jcp-7pqc The default model for the
openai,azure, andopenai-codexproviders is now GPT-6.2 Sol (gpt-6.2-sol) instead of GPT-6.1 Sol, at the same priceThe find tool's path patterns are now relative to the search directory, like grep's
glob:src/*.tsfindssrc/a.tsbut no longerpkg/src/a.ts; use**/src/*.tsto match at any depth. Patterns without/still match file names at any depthThe llama.cpp extension now connects to
http://127.0.0.1:9931by default, the current llama-server default port, instead of port 8080Anthropic message threads are now on by default for any provider whose baseUrl is on api.anthropic.com, not only the
anthropicproviderThe resume command printed on exit is now on its own line after
To resume this session:, so it can be selected with a triple-click
Fixed
Fixed
readstoring a note that the current model does not support images, which was replayed to the model as stale text after switching to a vision model. The image is still replaced with a placeholder per request for models without image input, and the read result now shows the warning in the TUI only, also when collapsedFixed MCP tools loaded by
tool_searchbeing dropped on resume and/reloadwhen an extension deactivates only its own tool atsession_start, as pi-web-access and pi-subagents do intoolActivation: "auto"modeFixed the standalone binary failing to resolve extension npm dependencies whose entry point is declared in
package.jsonmainorexports(for exampleshiki,yaml,graceful-fs), including imports made from inside other dependenciesFixed the find tool returning no results for patterns starting with
./, such as./src/*.tsFixed pi exiting with an uncaught
AbortErrorwhen areadtool call is cancelled while the file is being openedFixed pi exiting with an uncaught
AbortErrorwhen the session picker cancels loading after skipping a.jsonlfile that is not a session, and fixed such files staying openFixed
/sharecrashing pi when the GitHub CLI fails to startFixed
pi updateon Windows managed installs failing with "Could not verify managed Pi" when an older system Node.js comes before the installer's standalone Node.js on PATHFixed
/exportHTML showing raw terminal hyperlink escape codes around links in custom tool output; links are now clickableFixed the bash tool returning no output, with a footer like
[Showing lines 3-2 of 2], when the output ends in one line longer than about 100KB, such as agrepmatch in a minified file; it now shows the end of that lineFixed provider errors reading "Connection closed." not triggering automatic retries
Fixed
/resumesometimes selecting an unrelated session while the list reloads after deleting or renaming a sessionFixed Groq Qwen models failing with
400 Unexpected message role; OpenAI-compatible providers other than OpenAI now receive system messages (the system prompt and mid-conversation system messages) with thesystemroleFixed
enabledModels/--modelswarnings such asNo models match pattern "github-copilot/nope"being written over the interactive TUI on every/new,/resumeand/fork, which left stale lines in fullscreen mode. The warnings are now shown once at startup, in the chat in interactive mode and on stderr in other modes; RPC mode no longer repeats them on stderr for eachnew_session,switch_sessionorforkFixed the find tool matching path patterns through folders above the search path: with the project in
~/src/app,src/**/*.tsreturned every.tsfile in the projectFixed pi crashing with
RangeError: Maximum call stack size exceededwhen rendering a very large edit diff, for example after editing a file with a multi-megabyte line, and again on every resume of that sessionFixed pressing Esc during a batch of tool calls leaving the calls that had not started yet without a result in the session; they now get an "Operation aborted" result
Fixed an unreadable
CHANGELOG.md(pi's own or an extension package's) printingWarning: Could not parse changelogover the interactive TUI; it now shows no changelog notesFixed problems found when replacing the session with
/new,/resume,/forkor RPCnew_session,switch_sessionandforknever being shown, for example a project extension that fails to load after/resumeinto another project, or anenabledModelspattern that matches nothing after a settings change. Interactive mode now shows them in the chat and print, JSON and RPC modes write them to stderr; a problem already reported is not repeated. SDK hosts can call the newAgentSessionRuntime.takeUnreportedDiagnostics()at startup and after each replacement to get each diagnostic onceFixed RPC
new_session,switch_session,forkandclonerunning extensionsession_startandresources_discoverhandlers twice for the new session. RPC clients now also receive events emitted bysession_starthandlers, such asentry_appendedfor entries they appendFixed an exception thrown by an extension's
pi.eventshandler being printed withconsole.errorover the interactive TUI. It is now reported like other extension errors: in the chat in interactive mode, asextension_errorin RPC mode and asExtension error (...)in print mode. Errors from a handler that fails after its session was replaced are droppedFixed
AgentSession.reload()not emittingsession_startto extensions when the SDK host bound them withbindExtensions({}), after it had already emittedsession_shutdown. Extensions that start work insession_start, such as the MCP extension connecting its servers, stayed stopped after a reloadFixed
--mode jsondropping events emitted by extensionsession_starthandlers, such asentry_appendedfor entries they append, at startup and after an extension replaces the sessionFixed
Could not restore model <provider>/<model>. Using <fallback>being shown only at interactive startup. Resuming or forking into a session whose model no longer exists or has no login switched to another model without notice, and print, JSON and RPC modes never showed it. It is now reported in every mode, at startup and after/resume,/forkand RPCswitch_sessionorforkFixed git and npm output, such as
Cloning into ...oradded 12 packages, being drawn over the interactive TUI when pi installed or refreshed packages during/reload,/new,/resumeand/fork, including temporary-e git:and-e npm:packages. In interactive mode their output is now captured, and a failed install reports the end of it. Until the TUI starts, each install is announced asInstalling <source>...on stderr instead of showing git and npm output. git and ssh can no longer prompt in interactive mode: a package that needs a typed password or passphrase fails with a hint to set up a credential helper or ssh agent, or to runpi install <source>once from the shell.pi installandpi updatekeep live outputFixed fullscreen scrolling in long sessions saturating a CPU core by re-rendering the whole transcript on every wheel event
Fixed search in
/login,/logout,/scoped-modelsand/resumekeeping the highlight on the row scrolled to before typing, so Enter picked whichever entry landed there instead of the best match; a query now highlights the top match, as/modeldoes (#12245, #12246Fixed exiting pi,
pi configand the--resumeand other startup pickers dropping terminal output still queued in stdout, such as the resets that turn off mouse tracking and Kitty keyboard reporting; the shell could then print mouse and key reports as text. pi now waits up to one second for queued output before exiting or continuingFixed moving the cursor in a search field with Left, Right, Home or End moving the highlight back to the first entry in
/login,/logout,/model,/scoped-models,/resume,/settings, the per-model settings model picker andpi configFixed llama.cpp decision classifiers ignoring the image input that llama.cpp reports, such as Clef with its
mmproj;models.classify()now sends their images to/v1/systemoneFixed
kill -TSTPfrom outside pi stopping it without restoring the terminal, which left mouse tracking and the Kitty keyboard protocol on at the shell prompt and turned mouse movement into garbage input. pi now suspends as with Ctrl+Z and redraws onfgFixed changing the thinking level on the
openai-codexprovider invalidating the prompt cache on GPT-6 modelsFixed unscoped syntax-highlighted text inheriting the terminal foreground and becoming unreadable on contrasting tool cards
Fixed
/copysaying "No agent messages to copy yet." after Esc during a tool call or after a provider error; it now copies the previous answer. RPCget_last_assistant_textskips the same messages (#12119, #12121Fixed
pi.execreporting exit code 0 for commands ended by a signal, such as a timeout, an abort or a crash; it now reports 128 + the signal number, like a shell (143 for SIGTERM, 137 for SIGKILL) (#12088, #12089Fixed GitHub Copilot Gemini and Kimi models (
gemini-4-argon,gemini-4-carbon,gemini-3.8-flash,kimi-k3) ignoring the thinking level, and Copilot usage always reporting 0 reasoning tokensFixed provider errors that report only a Node socket error code, such as
read ECONNRESET,write EPIPE, orconnect ETIMEDOUT, not triggering automatic retriesFixed OpenAI rejecting requests with a 400 error when a tool's schema contains a node without a
type(such asitems: {}orType.Unknown()); such tools are now sent in non-strict modeFixed Mistral Large 4 tool calls stalling for minutes; Mistral models no longer receive strict tools unless
compat.supportsStrictModeis setFixed non-strict Anthropic tool schemas not telling the model that extra arguments are forbidden
Fixed GitHub Copilot models on the Responses API (GPT, Grok, MAI-Code) filling omitted optional tool arguments with empty strings or invented values
Fixed the OpenAI Codex WebSocket 60-minute connection limit failing the turn when it hit mid-stream, for example during long thinking; the request is now retried on a fresh connection
Fixed the
openai-codexprovider getting almost no prompt cache hits for the system prompt on GPT-6.2 modelsFixed the
openai-codexprovider ignoring a model'smaxTokensFixed fullscreen Kitty images disappearing after a full redraw such as a window resize
Fixed text after a multi-row inline image in Markdown drawing over the image
Fixed Kitty and iTerm2 images covering overlays drawn over them
Fixed Enter on a Tab path menu for an absolute path (
look at /tmp/fi) or a slash command argument menu (/open /tm) submitting the prompt instead of accepting the completion (#12309, #12312
Removed
- Breaking
Removed the retired
deepseek-v3.2anddeepseek-v4-flashfrom the Qwen Token Plan Team providers; usedeepseek-v4-flash-0731ordeepseek-v4.1-flash
1.5.2
10 Nov 2026Fork sessions whose folder is gone: when a session's original folder was moved or deleted, you can now fork it into the current folder instead of only continuing, so later resumes no longer ask. See Continue or switch sessions.
When a session's working directory no longer exists, for example after the project folder was renamed, interactive mode now offers
Continue,Fork, orCancelinstead of only continuing.Continueruns the session in the current directory without storing the choice, as before;Forkcopies it into the current directory as a new session with a new ID, like--fork, so the next resume no longer asks. This applies at startup (--session,--resume), to/resume, to/import, and toctx.switchSession()from extensionsFixed
/importof a session whose working directory no longer exists leaving a copy in the session directory, and continuing it leaving a second copy with the same ID (s.jsonlands-1.jsonl); the failed copy is now removed
New Features
Fork sessions whose folder is gone: when a session's original folder was moved or deleted, you can now fork it into the current folder instead of only continuing, so later resumes no longer ask. See Continue or switch sessions.
Changed
When a session's working directory no longer exists, for example after the project folder was renamed, interactive mode now offers
Continue,Fork, orCancelinstead of only continuing.Continueruns the session in the current directory without storing the choice, as before;Forkcopies it into the current directory as a new session with a new ID, like--fork, so the next resume no longer asks. This applies at startup (--session,--resume), to/resume, to/import, and toctx.switchSession()from extensions
Fixed
Fixed
/importof a session whose working directory no longer exists leaving a copy in the session directory, and continuing it leaving a second copy with the same ID (s.jsonlands-1.jsonl); the failed copy is now removedFixed Anthropic OAuth requests reporting an outdated Claude Code version
1.5.1
9 Nov 2026- Breaking
Custom
GrepOperations.readFilecallbacks are no longer called for context lines. The field remains accepted as optional and deprecated so existing integrations still compile. The grep tool now takes context lines from ripgrep and no longer reads matched files itself. To search a remote system, override grep'sexecute(); see Remote Execution - Breaking
OpenAI-compatible models no longer send
"tools":when all tools are disabled in a session that has used tools; the field is omitted. This fixes 400 errors (tools must not be an empty array) from vLLM 0.20+, also behind LiteLLM. If your proxy requirestoolsonce the history has tool calls, such as older LiteLLM versions in front of Anthropic, setcompat.requiresToolsWithToolHistory: truefor that model inmodels.json Shell login and logout: use
pi auth login <provider>andpi auth logout <provider>without starting a session, with API-key and no-browser sign-in options. See Credential commands.
New Features
Shell login and logout: use
pi auth login <provider>andpi auth logout <provider>without starting a session, with API-key and no-browser sign-in options. See Credential commands.Idle prompt cache warming: eligible caches now stay warm for up to 30 minutes between runs when a refresh is expected to pay for itself. See
cacheWarming.Anthropic message threads: requests send only new messages while the server continues the conversation, with full-request fallback when earlier context changes. See Anthropic compatibility.
Richer Mermaid diagrams: pie charts, author-defined styles, clickable labels on supported terminals, and diagrams in expanded branch and compaction summaries. See Display settings and Markdown transformers.
GitHub Copilot classifiers: call Microsoft-Decision-1 through codemode
models.classify()for typed decisions and probabilities. See Use classifier models.Package update summaries:
pi update --extensionsreports each package's npm version or git commit change and distinguishes unchanged, skipped, and warned updates. See Update Pi or packages.
Breaking Changes
- Breaking
Custom
GrepOperations.readFilecallbacks are no longer called for context lines. The field remains accepted as optional and deprecated so existing integrations still compile. The grep tool now takes context lines from ripgrep and no longer reads matched files itself. To search a remote system, override grep'sexecute(); see Remote Execution - Breaking
OpenAI-compatible models no longer send
"tools":when all tools are disabled in a session that has used tools; the field is omitted. This fixes 400 errors (tools must not be an empty array) from vLLM 0.20+, also behind LiteLLM. If your proxy requirestoolsonce the history has tool calls, such as older LiteLLM versions in front of Anthropic, setcompat.requiresToolsWithToolHistory: truefor that model inmodels.json
Added
pi updatenow lists each updated package with its npm version or git commit change, for examplenpm:foo 1.2.0 -> 1.3.0, and printsPackages are already up to datewhen nothing changed. When nothing changed but an update printed a warning, for example because npm left the old version installed after resolving stale registry metadata, it printsNo packages were updated, see the warnings aboveinstead; in offline mode it printsSkipped package updates (offline mode).PackageManager.update()reports each changed package as a newresultprogress event whosemessageholds that line (#11707, #11711)Anthropic sessions now use message threads: each request sends only the new messages and the server continues from the previous reply, falling back to the full request when the conversation changed (for example after compaction or a tool list change). On by default for the
anthropicprovider onapi.anthropic.com; setcompat.supportsMessageThreadsinmodels.jsonto enable it for gateways that forward thethreadfield, or to turn it off. Cache warming does not touch the session's threadpi.registerMarkdownTransformer()takes an optional{ messageTypes }that selects the Markdown surfaces a transformer runs on. Besides user, assistant, and thinking Markdown, transformers can now opt intocustom(custom messages and skill blocks),summary(branch and compaction summaries), andchangelog(What's New,/changelog, the Extension Updates notice, and the Update Available note). Without options a transformer runs on user, assistant, and thinking Markdown as before. The built-in Mermaid renderer now also draws diagrams in expanded branch and compaction summariespi auth login <provider>logs in to a provider from the shell without starting a session, for examplepi auth login radius. It signs in with an account in the browser and exits;--api-keysaves an API key instead, and--no-browseruses device code or copy-code sign-in instead. Logging in to Radius also configures the Radius MCP server.pi auth logout <provider>removes the stored credentials, like/logoutctx.ui.custom()overlays now honor theblockedoption: pi reports blocked program status (OSC 7501) from when the overlay is shown until it callsdone(), also while it is hidden withsetHidden(). Closing one overlay keeps other open questions blockedAdded the
qwen-token-plan-individual-cnprovider for Qwen Token Plan Individual on the China endpoint, usingQWEN_TOKEN_PLAN_CN_API_KEYAdded Microsoft-Decision-1 classifier support through GitHub Copilot, including codemode
models.classify()Added
compat.promptCacheKeyMode("auto","enabled", or"disabled") to enable prompt cache keys for compatible proxies or omit them for endpoints that reject them. Set it inmodels.jsonwhen using piAdded
compat.requiresNonNullAssistantContentfor OpenAI-compatible endpoints that require an empty string instead ofnullin assistant messages containing only tool calls. Set it inmodels.jsonwhen using piAdded syntax highlighting for Dart, Nix, Groovy, and Gradle files in read and write views
Added GLM 5.3 to the Qwen Token Plan Individual model allowlist
Added Claude Fable 5.2 (
claude-fable-5-2) to theanthropicprovider ($8/$40 per million input/output tokens, 1M context, 128K output). Like Fable 5.1 it uses per-message effort and mid-conversation system messages; requests its safety classifiers refuse before any output are answered by Claude Opus 5 through server-side fallback and billed at Opus 5's price. Claude Fable 5.2 and Mythos 5.2 on other providers, such as OpenRouter'santhropic/claude-fable-5.2, get the same per-message effort support
Changed
The default model for the
anthropicprovider is now Claude Opus 5.5 (claude-opus-5-5) instead of Claude Opus 4.8Default models moved to newer models at the same or a lower price:
amazon-bedrockClaude Opus 5.5,azureGPT-6.1 Sol,googleandgoogle-vertexGemini 4 Argon,vercel-ai-gatewayGLM 5.3,basetenGLM 5.3,minimaxandminimax-cnMiniMax-M3,qwen-token-planandqwen-token-plan-cnQwen3.8 Max,opencode-goKimi K3.1 Pro, and Kimi K2.7 Code instead of Kimi K2.6 formoonshotai,moonshotai-cn,huggingface,opencode,openrouter,cloudflare-workers-aiandcloudflare-ai-gatewaycacheWarmingnow defaults to"idle"instead of"streaming": pi also keeps eligible prompt caches warm for up to 30 minutes between agent runs, but only when a refresh is expected to pay for itself. Set"cacheWarming": "streaming"in global settings for the previous behaviorConsecutive cache warms now show as one
Cache warmed: $<total>line with the cumulative cost instead of one line per warmThe OpenAI Codex provider is shown as "OpenAI Codex" again instead of "OpenAI Codex (legacy)" in
/loginand the model selectorThe
/loginand/logoutprovider lists and/login <provider>completions now show the preferred partners Radius and OpenRouter first, then all other providers alphabetically. In the provider lists, partners are marked with✦, the selected Radius row shimmers in the Radius logo colors and the selected OpenRouter row glows in OpenRouter's brand color (lime on dark themes, violet on light themes)Upgraded terminal Mermaid rendering to support more diagram types, including pie charts, author-defined styles, and clickable labels on terminals with hyperlink support
Fixed
Fixed the grep tool's
globmatching from the folder pi was started in instead of frompath: withpath: "repo",glob: "migrations/main/*.js"returnedNo matches foundeven thoughrepo/migrations/main/a.jsexists. Globs are now relative topath, the same form grep prints its results in, also whenpathis a symlink. A leading./(./src/*.ts) is ignored instead of matching nothing (#11807, #10998)Fixed prompt template arguments dropping apostrophes and merging words: an apostrophe inside a word, as in
doesn't, no longer opens a quote; a quote with no closing match is kept as text; and\"inside"..."(or\'inside'...') gives a literal quoteFixed the bash tool reporting only
Command timed outwhen the command had already exited but a background process it started kept writing to its output until the timeout. The error now also reports the shell's exit code and when it exitedImages in steering and follow-up messages, including
prompt()while streaming, are now resized like images in idle prompts instead of being sent and saved at full sizeA failed token refresh of an MCP connection no longer registers an OAuth client or replaces the PKCE verifier of a sign-in in progress, which made that sign-in fail. MCP sign-in now refreshes the stored tokens with the stored client even when the loopback callback port changed, instead of registering a new client and opening the browser. Ctrl+C in
/mcpand the--timeoutofpi mcp loginnow also stop a sign-in while it waits up to 25 seconds for another pi process's token refresh, and a refresh lock that fails for another reason than that wait now fails the sign-in instead of opening the browser.Fixed
pi mcp listreporting MCP servers with"auth": { "provider": "<provider>" }, such as Radius, as needing sign-in although the provider was signed in: the shell command sent no token. It now sends the provider's/logintoken and refreshes it when it expired. For these servers the sign-in hint names/login <provider>instead ofpi mcp login, andpi mcp loginandpi mcp logoutpoint to/loginand/logoutinstead of failing with "does not use OAuth"Fixed RPC mode ignoring an extension's
ctx.shutdown()while idle, for example from a timer: the process kept running until the client sent another command. pi now shuts down right away when idle, and a shutdown requested by an extension command waits until the command'spromptresponse is writtenFixed changelog notes repeating on every fresh launch when the installed package version has a prerelease suffix, such as
0.1.5-20261020.7: the saved version was read as0.1.0, so the0.1.1to0.1.5notes counted as new each time. A saved version withoutmajor.minor.patchnow shows no notes and is replaced by the installed version, instead of showing the whole changelogFixed RPC mode cutting off shutdown cleanup, such as a pending MCP token refresh, when stdin closed and SIGTERM arrived together, which process managers like Paseo and systemd do. Pi lost the refreshed token and the MCP server revoked the sign-in. Once a shutdown has started, a stdin EOF or a single signal no longer interrupts it; a second signal still exits at once. Print mode likewise no longer dies from a signal that arrives during its final cleanup
Fixed the grep tool reading every matched file into memory when
contextwas set, which ran out of memory on large files; context now comes from ripgrep, which also fixes UTF-16 files and lone CRs in matched lines, and the match limit notice no longer appears when exactlylimitmatches exist (#9276, #11423)Fixed the find tool reporting the result limit when exactly
limitfiles match, andlimit: 0in find and ls returning every result or an empty listing; limits below 1 now mean 1Fixed grep match lines, codemode output, compaction summaries of tool results, and MCP error and server summaries cutting an emoji or other non-BMP character in half when truncating, which left an invalid lone surrogate in the text
Fixed the find tool on Windows returning no results for patterns with Windows separators such as
src\**\*.ts, and for every path pattern such assrc/**/*.tswith fd 10.5.1, which pi now downloads. Windows path patterns now match the same way with every fd version:src/**/*.tsalso finds files directly insrc/, andsrc/*.tsno longer finds files in subdirectoriesFixed the read tool loading whole files into memory, which ran out of memory on large files even with
offset/limit; it now streams the file and keeps only the shown lines. CustomReadOperationscan stream too by providing the new optionalcreateReadStreamFixed OAuth logins such as
openaiandanthropicfailing withinvalid_grantafter pi was stopped while it refreshed an expired token, for example an RPC client that stops pi right after startup while the background catalog refresh runs. The token endpoint had already replaced the refresh token, but pi exited before saving the new one. On exit, pi now waits for a token refresh in progress to be savedFixed Ctrl+P not switching to the only model in scope (or the only available model) when the active model is a different one; it showed "Only one model in scope" instead. Cycling forward from a model outside the scope now also selects the first scoped model instead of skipping it
Fixed
/reloadand session switches leaving extension overlays on screen when an extension had more than onectx.ui.custom({ overlay: true })open: only the topmost overlay was removed, and the others kept focus and input although the extension that opened them was gone. All extension overlays, including hidden ones, are now removed and disposed; pi's own overlays stay openFixed
/mcpsign-in staying on "Connecting…" for minutes after a successful browser sign-in, with Ctrl+C doing nothing: pi waited for the browser to drop an unused connection to the OAuth callback port. The callback server now closes remaining connections when it shuts downFixed pi aborting (
zsh: abort) after repeated image reads, reported on macOS arm64 with Node 24.16. Each image resize starts a worker thread, and pi did not wait for it to stop, so the next resize started a new worker while the previous one was still shutting down. The crash is most likely a Node bug that this overlap triggers and could not be reproduced on Linux; image resizes now wait for the previous worker to stop, which adds about 15 ms per imageFixed
tool_searchandsearchTools()not finding a tool by property names and descriptions behind a local$ref, as in MCP input schemas that put their models in$defs. Referenced definitions are read once, so recursive schemas do not loop, and definitions nothing references stay out of the search textFixed the codemode
resolves tosummary of a tool leaving out field names behind a local$ref, for examplestructuredContent?instead ofstructuredContent?: { results }for MCP output schemas that put their models in$defsFixed every codemode script failing with
URI malformedwhen any callable tool, such as an MCP tool, had a$refwith a malformed percent escape (#/$defs/%E0) in its schema, even scripts that do not call that tool. The parameter now renders asunknownin the tool's declarationFixed
ctrl+onot expanding or collapsing the output of!and!!commands run while the agent is working. New!/!!output now starts in the current expand state, the output no longer disappears when a steering or follow-up message is queued, and it moves into the chat when the agent finishes instead of on the next submitted messageFixed resuming a session restoring an old thinking level after
/treenavigated to before a/thinkingchange. Like the model, the live thinking level is now recorded on the new branch with the next promptFixed inline images in fullscreen mode showing only their first row in Orca, while regular mode showed them in full. Fullscreen now draws Kitty images after the rows they cover on every terminal, as it already did for WezTerm
Fixed the edit tool rewriting the line endings of untouched lines in files with mixed CRLF/LF or lone CR line endings, and running out of memory on large files after the file was already written; the diff now covers only the changed lines and their surroundings and is computed before writing
Fixed the Windows installer (
irm https://pi.dev/install.ps1 | iex) aborting withNativeCommandErrorin Windows PowerShell 5.1 when npm or Node.js printed any warning, for examplenpm warn Unknown user configfrom an unknown.npmrckey. It failed before the action menu when migrating an npm installation, and during managed installs. An invalidmanaged-install.jsonalso aborted the installer instead of being reportedFixed
pi update --modelsfailing withCould not refresh model catalogs: openai: OAuth refresh failed for openai: ... invalid_grantwhen a provider's login had gone stale. Built-in providers whose catalog comes from pi.dev, such asopenai, no longer refresh their OAuth token for a catalog refresh at all. For providers that list models with the login, such as Radius and OpenRouter, a failed token refresh now prints a warning naming the provider andpi auth login <provider>, and the other catalogs still refresh; the command fails only for other errors, such as a network failureFixed the powershell tool reporting
Command exited with code 1under Windows PowerShell 5.1 when a native command succeeded but its stderr was redirected, as ingit push 2>&1 | Select-Object -Last 3. Windows PowerShell 5.1 turns that stderr into error records; such commands now report the exit code PowerShell 7 reportsFixed deleting a paste marker with Backspace corrupting later short markers such as
[paste #2], causing the marker text to be submitted instead of its pasted contentFixed dropped provider streams reported as "cut short after the reply had begun" not triggering automatic retries
Fixed the subagent extension example passing Bun's Windows virtual entry path as part of the prompt when spawning a compiled pi binary
Fixed the model selector and llama.cpp Hugging Face search moving the selection when a background catalog refresh or search result arrives
Fixed internal cursor markers leaking into terminal output, including with multiple markers or horizontally clipped content
Fixed the subagent extension example showing running tasks as completed, losing subprocess failure diagnostics, and failing to stop child processes on abort. Agent discovery and trust checks now use each task's working directory, and single and chain results are truncated without losing the full tool details
Fixed large pasted text being replaced by literal paste markers when switching editors, opening custom UI, or restoring queued messages
Fixed prompts and session-tree navigation overlapping during asynchronous input hooks, authentication, or branch summaries, which could install the wrong conversation context
Fixed Claude Sonnet 5 requests sending an unsupported temperature parameter
Fixed extensions loaded through symlinked paths failing to resolve their dependencies by resolving the entry file's real path before importing it
Fixed the plan-mode extension example missing completion markers in thinking blocks and common completion phrases, dropping formatted or long plan steps, and sending an extra model turn to acknowledge completion
Fixed OpenAI Responses requests sending
session_idinstead of the documentedsession-idaffinity headerFixed delivered image-only steering and follow-up messages remaining in the pending queue
Fixed cancellation during lazy provider loading or authentication being reported as an error instead of an abort
Fixed session runtime disposal tearing down extensions before active tools finished aborting and their results were recorded
Fixed the session-tree search cursor not being shown or positioned for input methods
Fixed SDK and RPC reloads replacing extensions while a response or compaction was still running; reload now asks the caller to wait
Fixed rendered edit diffs losing shared indentation when a change inserted text at the start of a line
Fixed
!and!!commands omitting the truncation notice and full-output file link when earlier output chunks had been discardedFixed extension selection dialogs growing beyond the terminal height; they now show a scrolling window of up to ten options
Fixed
pi configignoring custom keybindings fromkeybindings.jsonFixed concurrent OAuth refreshes through symlinked paths to the same
auth.jsonby sharing the lock on the resolved fileFixed provider connection errors hiding their underlying cause, such as connection refusal or certificate errors, behind generic messages
Fixed Ctrl+V not pasting clipboard images or text on Windows and WSL; both Ctrl+V and Alt+V are now default paste shortcuts there
Fixed llama.cpp connection errors omitting the server URL and underlying network error
Fixed Alacritty hyperlinks not being enabled when
TERM_PROGRAMis unset; detection now also usesTERMandALACRITTY_WINDOW_IDFixed OpenAI Codex SSE retries retaining abort listeners after each backoff, causing listener buildup on repeated retries
Azure OpenAI models now keep OpenAI's long-context price tier, so requests above 272K input tokens are costed at Azure's long-context rates
Fixed Anthropic and Bedrock sending
strict: "prefer"tools strict again on every request after Claude rejected them for exceeding its strict tool limits, which cost a rejected request (400) per turn. Later requests with the samesessionIdnow send those tools non-strict right away, untilcleanupSessionResources(sessionId)Fixed tool arguments for
Type.BigInt()andType.Undefined()parameters being rejected: they convert as TypeBox converted them ("42"and42to42n,nulltoundefined), now that pi-schema supports both types.Type.Void()still rejects every JSON argument.Fixed Azure GPT-6 Astra, Luna, Sol, GPT-6.1 Sol, and GPT-6.1 Terra using a 272K context window instead of the 1,050,000 tokens Azure Foundry deploys them with
Fixed OpenRouter rejecting requests with a 400 (
This endpoint's maximum context length is ...) for models whose output limit is close to the context window, such as DeepSeek flash. OpenRouter counts the input with its own tokenizer, which can be several percent higher than the provider's count, and pi left only 4,096 tokens spare. pi then treated the 400 as a context overflow and compacted the session, even at around 20% of the window. The default output limit now leaves 15% of the estimated input unused, and near a full window at most 32,768 output tokens are requestedFixed Anthropic non-strict tool schemas dropping root
$defsanddefinitions, which left property$refs pointing at nothing. Claude then produced wrong arguments for such tools (common in MCP servers), for example a JSON string instead of an object, and some Anthropic-compatible gateways rejected the request outrightFixed OpenCode Go Claude Haiku 5.5 failing with a 400 (
Invalid signature in thinking block ... bound to a different conversation) after the tool list, system prompt, or earlier history changed, which left the session failing on every retry. OpenCode Go and Zen Claude 5.5 models, Opus 5, and Fable 5.1 now use per-message effort and sendthinking.block_binding: drop_block, like the Anthropic providerFixed Markdown displaying HTML comments (
<!-- ... -->) as visible textFixed codemode
console.*output and thrown non-Error values printingNaN,Infinity, and-InfinityasnullFixed large MCP stdio messages being repeatedly copied and scanned as chunks arrived, causing slow tool responses
Fixed MCP Streamable HTTP SSE replies with CR-only line endings failing with "stream ended without a response"
Fixed oversized MCP SSE responses keeping their HTTP connections open after the message-size limit rejected them
Fixed MCP stdio server subprocesses surviving disconnect or pi exit on POSIX when the server's main process had already exited
Fixed MCP Streamable HTTP JSON replies bypassing message-size limits and HTTP errors buffering entire response bodies. Oversized replies and error-body content beyond the first 8 KiB are now cancelled
Fixed valid tool arguments being rejected when their schema combined local
$refs with a relative root$id, such asschemas/tool.jsonFixed the find tool starting an untracked
fdprocess when cancelled during its parent-directory lookup
1.5.0
2 Nov 2026- Breaking
Tool argument validation runs on
@earendil-works/pi-schema. TypeBox features outside JSON Schema (Type.Refine(),Type.BigInt(),Type.Undefined(), codecs,Format.Set()formats) no longer apply to tool arguments; see Tool parameter schemas and TypeBox. Plain JSON Schema tools, including MCP tools, now get the same coercion as TypeBox tools, such as converting numeric strings and wrapping a single value into an array. Validation failures the model sees list pi-schema's messages (- path: missing,- limit: expected number, got string) instead of TypeBox's (must have required properties path,must be number). - Security
Updated
undicito 8.12.3 and the Gondolin extension example to 0.14.0 for security advisories - Breaking
Removed the native OS
sandbox/extension example and its dependency. Use the Gondolin example or the containerization guide.
New Features
Extension update changelogs: see what changed in installed extension packages on the next fresh launch after an update. Package authors can choose the changelog path, and
collapseChangelogcondenses the notices. See Show a changelog.Faster streaming: large tool calls are parsed incrementally, and code-heavy answers and tool rows reuse highlighting and rendered Markdown instead of repeating the work on every update. See Keep rendering responsive.
Custom self-update commands: use
PI_SELF_UPDATE_COMMANDorselfUpdateCommandto update pi through a wrapper or another package manager. See Custom update command.GPT-6.1 Terra: available through OpenAI, Azure OpenAI Responses, and OpenAI Codex, with tool search, mid-conversation thinking-level changes, and thinking
off. Codex offers a 500K context window. See Models.
Breaking Changes
- Breaking
Tool argument validation runs on
@earendil-works/pi-schema. TypeBox features outside JSON Schema (Type.Refine(),Type.BigInt(),Type.Undefined(), codecs,Format.Set()formats) no longer apply to tool arguments; see Tool parameter schemas and TypeBox. Plain JSON Schema tools, including MCP tools, now get the same coercion as TypeBox tools, such as converting numeric strings and wrapping a single value into an array. Validation failures the model sees list pi-schema's messages (- path: missing,- limit: expected number, got string) instead of TypeBox's (must have required properties path,must be number).
Added
Extension packages can now show their changelog after an update. Pi reads
pi.changelogPathfrom the package'spackage.json, orCHANGELOG.mdin the package root, and on the next fresh launch shows entries newer than the last seen version under "Extension Updates". ThecollapseChangelogsetting condenses this to one line per package (#5958, #8790)pi.exec()accepts amaxBufferoption that limits stdout and stderr combined, and its result has atruncatedfield that is true when output exceeded the limitDevice-code login (for example GitHub Copilot) now opens the verification page and copies the code when you press Enter. Providers can turn either action off with
openBrowserOnConfirm: falseorcopyCodeOnConfirm: false(#9282, #9301)pi updatecan update Pi through a wrapper or another package manager. SetPI_SELF_UPDATE_COMMAND(for examplemise use -g npm:{packageName}@{version}) or the globalselfUpdateCommandsetting;pi updateruns it when pi.dev reports a newer release and fills{packageName},{version}, and{installSpec}from that release. See Custom update command (#7756, #5607)Extensions can format the user's configured keybindings with
keyText()andkeyHint()from the tool render context and the message and entry renderer options. Importing these functions from@earendil-works/pi-coding-agentcan resolve the extension's own copy of the package, which does not see pi's keybindings, so hints such askeyHint("app.tools.expand", "to expand")rendered without the key. The docs now recommend the injected functionsAdded persisted session-entry
messageIdand ISOtimestamptoMarkdownTransformContext. They are available when rendering saved entries and undefined for live renderingAdded a
blockedoption to inlinectx.ui.custom()components so approval prompts and other custom dialogs can report that pi is waiting for user input. Overlays ignore the option. See Custom screens and overlaysAdded
Usage.cost.totalFromProviderto distinguish billed totals from catalog estimates in SDK and extension usage data; combined usage is marked provider-reported only when every part isAdded concrete
responseModelmetadata to messages from pi-messages providers such as Radius, including when a virtual model is requested. See AssistantMessageAdded GPT-6.1 Terra for OpenAI, Azure OpenAI Responses, and OpenAI Codex, with tool search and mid-conversation thinking-level changes. It supports thinking
off; Codex offers a 500K context window
Changed
Extensions keep importing
Type,StaticandTSchemafrom@earendil-works/pi-ai: Pi maps that import to pi-ai's compat entry plus the TypeBox it bundles, now that pi-ai itself no longer re-exports TypeBox, and itsStringEnumthere stays a TypeBoxTUnsafe<T>. The built-in tool factories (createReadTool()and the others) returnTypeBoxAgentTool<typeof schema>, which keepsparamstyped for TypeBox.Tightened system prompt
pi updatefor installs from the pi.dev installer no longer passes--min-release-age=0tonpm ci. The release lockfile pins every package, so the flag had no effect, and security scanners that block it stopped the update/modeland/scoped-modelsnow show a model's name next to its ID when the name adds words the ID lacks, such as a renamedmodels.jsonoverride (#5916, #11347)MCP OAuth sign-in now uses pi's Client ID Metadata Document by default when the authorization server supports it for public clients, and registers a client dynamically otherwise, like Codex, Claude Code, and opencode. Clients registered before stay in use while their redirect URI matches. The new default
oauth.clientRegistration: "auto"registers whenclientNameor a non-defaultcallbackUrlis set;"dcr"always registers.pi mcp add --oauth-client-registrationsets the modeChanged the default OpenAI model from
gpt-5.5togpt-6.1-sol, including after ChatGPT sign-in when no model is selectedUpdated the built-in model catalog to models.dev v2. Realtime and image-only models are no longer offered as chat models; newer OpenAI and Google chat models are included
Fixed
Updated glob matching to prevent denial of service from amplified glob patterns.
Fixed a saved "Do not trust" decision being ignored for projects without trust-requiring resources, for example a folder with only
AGENTS.md. pi auto-trusted such projects, soctx.isProjectTrusted()returned true and package commands treated them as trusted. A saved decision intrust.jsonnow takes precedence; auto-trust applies only when no decision is savedFixed
Shift+Enterqueueing a follow-up instead of inserting a newline in Apple Terminal setups that sendESC+ Return. pi now checks the local Shift key state for that sequence, as it already did for plain Return;Alt+Enterwithout Shift still queues a follow-upFixed the working status (OSC 7501) disappearing in Rex and iTerm2 during runs in regular TUI mode. Rendered messages wrote OSC 133 prompt markers to the terminal, which the terminal treats as a new shell prompt; regular mode no longer writes them
Fixed compaction, branch summaries, and bug-report summaries in SDK sessions passing the stored OAuth token back to the model runtime as an explicit API key. Provider auth that treats explicit keys differently from stored OAuth now sees the OAuth login, as normal prompts do
Fixed a redundant system prompt update (tools, rules) when an extension set the same active tools in a different order, for example a reversed
pi.setActiveTools()list. Tools that stay active now keep their order and newly activated tools are appendedFixed
pi.exec()crashing pi withRangeError: Invalid string lengthwhen a command printed more output than fits in a string. The default output cap is the smaller of 256 MiB and the runtime string limit; past the cap, the command is terminated and the result hastruncated: trueand a nonzero exit codeFixed
pi.exec()never returning when a command ignored SIGTERM after a timeout or abort. pi now sends SIGKILL 5 seconds later, as intendedFixed
pi installandpi removeadding ever deeper nested entries to the npmpackage-lock.jsonwhen the agent directory is reached through a symlink, such as~/.pilinked into a dotfiles repo. pi now passes the resolved real path to npmMCP sign-in now rejects authorization endpoints and other OAuth metadata URLs that are not
http:orhttps:. Before, a malicious MCP server could supply afile:or custom protocol URL, which pi passed to the platform URL handler.Automatic
fdandrgdownloads no longer fetch x86-64 binaries on ARM32 and other architectures without a prebuilt release. pi now warns and recommends installing the tool with the system package managerFixed
/modelsearch ranking letters scattered over an earlier part of an entry above a real match, for exampleopuslistinggpt-oss-120bandclaude-sonnet-4.5beforeclaude-opus-4.1(#11220, #11296)Fixed
/resumesearch matching scattered letters anywhere in session messages and IDs, for example1024matching inside a session UUID. Session names and working directories still match fuzzily; IDs and message text now match only as exact substrings (#11220, #11024, #11296)Fixed HTML export showing a blank page for sessions with thousands of entries (#11225, #11293)
Fixed HTML export not showing system messages: the system prompt and tools, and later section updates such as after
/reload.pi --exportnow includes the system prompt and tools too (#11261, #3464, #11290)Fixed aborting a bash command on macOS sometimes running the next command of a list such as
long_step; next_step. pi now stops the process group before killing it (#11252, #11289)Fixed large streamed tool calls taking excessive CPU time and memory because every delta re-parsed the accumulated arguments. Providers and agent proxy streams now parse each delta once, grammar tools no longer copy their entire input per delta, and interactive tool renderers run at most once per frame
Fixed the login dialog opening sign-in URLs that are not web URLs, and printing control characters from sign-in, device-code, and info URLs, which could end a terminal hyperlink and send escape sequences to the terminal
Fixed
tui.select.pageUpandtui.select.pageDownbeing ignored in slash command and file autocomplete,/settings, and the model, scoped models, login, fork,/mcp, and llama.cpp selectors. In fullscreen mode, PageUp/PageDown now move an open list or autocomplete menu instead of scrolling the transcript, and a page key bound to Ctrl+V pages autocomplete instead of pasting an image (#7629, based on #7680Fixed tool output and
!command output showing the parameters of control sequences with a<,=or>prefix as text, for example35;59;1Mfrom a mouse reportESC[<35;59;1Mor0;276;0cfrom a terminal replyESC[>0;276;0c. Text after the keypad-mode sequencesESC=andESC>is still keptFixed
pi -e npm:<pkg>staying on the version from the first download. On each start and/reload, pi now reinstalls an-enpm package without an exact version when the registry has a newer matching version.pi -e git:<repo>@<branch>likewise stayed on its first checkout and is now fetched and reset to the branch; a commit-hash ref that is already checked out is not fetched. Offline mode and failed checks keep the cached copy (#11375, #11590)Fixed Windows paths being compared case-sensitively. When the working directory and home directory differed only in drive-letter casing (
c:\Users\mevsC:\Users\me), skills in~/.agents/skillsloaded twice, as project and user skills, and were reported as name collisions. The same mismatch also re-asked for saved project trust, hid sessions in a custom session directory from--continueand/resume, and stoppedpi removefrom matching configured local packagesFixed paths next to the home directory being shown with a broken
~prefix, for example/home/me2/projectshown as~2/projectwhen home is/home/meFixed
pi mcp login -landpi mcp logout -lfailing withUnknown option --local.althoughpi mcp addandpi mcp removeaccept-l. With-l, they now require the server to be a project server. Unknown options are now reported as typed, and a mistyped server name gets a suggestion, for exampleDid you mean "gurker"?Fixed
pi updatenaming pnpm for installs that only live underPNPM_HOME, for example a wrapper under$PNPM_HOME/global-nub/. pi said "not managed by a global pnpm install"; it now names the package manager only when that manager owns the installFixed image paste, copy, and opening the browser for sign-in failing silently on WSL when the working directory is deep. Windows programs such as
powershell.exeandexplorer.execannot start from a directory whose\\wsl.localhost\...path exceeds 260 characters; pi now starts these helpers from/Fixed the external editor failing silently, for example when the editor command is wrong or a Windows editor cannot start from a deep WSL directory. pi now shows a status message such as
External editor exited with code 1; prompt unchangedFixed
tool_searchand codemodesearchTools()ranking a tool below tools that only mention its name in their descriptions, for examplepeekaboo apprankingmcp__peekaboo__app14th. A tool matching more of the query's words now ranks above one matching fewer, a query word in a tool's name scores more than any number of mentions in its description, a tool whose whole name is in the query gets a large bonus, and long server instructions no longer lower the scores of that server's toolsFixed
tool_searchand codemodesearchTools()finding nothing for Chinese, Japanese, Thai, and other non-ASCII queries, because the tokenizer kept only ASCII letters and digits. Words in any script now match, text without spaces between words such as搜索记录is split into words, and canonically equivalent text such ascaféandcafe\u0301matches (#10816, #11037Fixed
pi updatestopping withgit clean -fdx failed with code 1on Windows when another pi session had a git extension's native module or executable open. Windows cannot delete such files, so the update left the new code with the old dependencies. pi now warns, installs the dependencies anyway, and retries the clean on the next updateFixed tool rows and streamed answers highlighting the same code again on every redraw. A running codemode row highlighted its whole script on every progress update and timer tick, a complete
writerow its whole file on every update and expand toggle, a streamedwriteits first 50 lines on every delta, and a streamed answer every earlier code block on every delta. Rows now reuse their last highlighting until the code, theme, or loaded grammars change, and assistant messages keep their rendered Markdown blocks while streamingFixed resized photos being sent as multi-megabyte PNGs even when JPEG was much smaller, causing large requests to fail or time out. Image resizing now tries the smaller encoding first
Fixed
pi update --extensionsreporting success without warning when npm's stale registry metadata left a package at its old version. pi now reports the installed and expected versionsFixed skill validation warnings being labeled
[Skill conflicts]even when there was no name collision. They are now shown under[Skill warnings]Fixed codemode tool descriptions including entire union and multi-line output types in every request. Object unions now summarize their fields and discriminators; other multi-line types point to
describeTool()for the full declarationFixed the model-scope startup notice being left in the shell after exiting fullscreen mode. It now appears inside pi's fullscreen header
Fixed startup from a deleted working directory crashing with an uncaught
uv_cwdstack trace. pi now asks the user to change to an existing directoryFixed failed session writes leaving entries in memory that were absent from disk, so later entries pointed to missing parents and resumed sessions lost earlier history and tools
Fixed
/treebranch summaries ignoring the session's thinking level, which could make reasoning models reject summary requests with an unsupportednoneeffortFixed hidden thinking with an empty hidden-thinking label still adding blank rows before answers and tool calls
Fixed package installation and removal printing npm audit and funding advice, including
npm audit fixinstructions that could install a second copy of pi into the package directoryFixed replacing an extension editor while a dialog was open hiding the dialog and leaving it unanswerable. The new editor now appears after the dialog closes
Fixed
grepcontext lines being offset from ripgrep's match line numbers when a file contains a lone carriage returnFixed
--model <provider>/<id>failing for provider IDs containing slashes, such as URL-derived llama.cpp provider IDsFixed slow session forking on filesystems where opening files is expensive by writing the fork through one file handle
Fixed
pi.exec()corrupting UTF-8 characters split across stdout or stderr chunksFixed
pi.exec()returning empty stderr when a command could not be spawned, such as a missing executable. It now preserves the spawn error messageFixed a configured package failing to install crashing startup. pi now skips the package, reports an extension warning, and retries on the next start or
/reload; explicit install and update commands still fail with a nonzero exitFixed the todo extension example crashing the TUI when a tool call failed argument validation
- Security
Updated
undicito 8.12.3 and the Gondolin extension example to 0.14.0 for security advisories Fixed OpenRouter cost totals using catalog estimates instead of reported billing for Chat Completions and Anthropic Messages. BYOK totals include both OpenRouter charges and reported upstream costs; component costs remain catalog estimates
Fixed OpenCode Zen and OpenCode Go requests using shared endpoints instead of their protocol-specific OpenAI, Anthropic, Google, and System One endpoints
Fixed Zhipu/Z.ai usage-limit errors (code
1308) being treated as transient HTTP 429 errors, causing automatic retries that cannot succeed until the usage cap resetsFixed LiteLLM's
ContextWindowExceededErrornot being recognized as context overflowFixed Gemini tool schemas sending unsupported
constkeywords. They are now converted to single-valueenumschemas without rewriting literal data in defaults, examples, or enum values, or losing properties named__proto__Fixed the faux provider counting simulated cache-write tokens as both input and cache writes
Fixed Orca terminals falling back to text for inline images. Orca is now recognized as supporting true color and hyperlinks, and Kitty images when it advertises
ORCA_IMAGE_PROTOCOL=kittyFixed bare Markdown URLs swallowing following CJK punctuation and text; explicit Markdown links and angle autolinks are unchanged
Fixed Kitty-protocol F1–F12 shortcuts, including modified function keys, not being recognized, and F13+ and other functional keys inserting invisible private-use characters into the editor (#11419, #11345)
Fixed Kitty images inside padded or colored boxes disappearing when their first row scrolled above the fullscreen viewport. Partially visible images are now cropped correctly, including beside other text
Fixed deeply nested Markdown blockquotes producing exponentially many lines, hanging or crashing rendering once their borders exhausted the available width
Fixed Claude sessions failing with
role 'system' must precede an 'assistant' messagewhen a system update was followed by an empty or reasoning-only assistant replyFixed Claude mid-conversation thinking-level changes failing with
Invalid effort levelon accounts whose organization caps effort belowhighFixed Kimi K3.1 Pro models using Kimi K2 request settings on Kimi Coding, Moonshot, and OpenCode Go. They now send the correct reasoning effort, preserve earlier thinking, accept tools added mid-conversation, and use K3-family pricing
Fixed OpenCode DeepSeek V4.1 Flash requests failing when long cache retention sent the unsupported
prompt_cache_retentionparameterFixed GPT-5.4 and GPT-5.5 failing with
Missing namespace for function_callafter switching from a model that had called toolsFixed Bedrock Claude requests failing when preferred strict tools exceed the provider's strict-tool limits. Requests now retry once with preferred tools non-strict while keeping required strict tools strict
Fixed OpenRouter's
~anthropic/claude-*-latestaliases losing earlier thinking after tool calls by using the Anthropic Messages APIFixed terminal titles containing control characters being able to inject terminal escape sequences, and pi-messages streams accepting invalid content-block indices
Fixed fullscreen double-click selection splitting paths and tokens at
~,.,\, and+; word characters now match iTerm2's defaultsFixed display math wider than the content area being split apart by line wrapping; it now falls back to raw LaTeX
Fixed styled words ending exactly at the wrap width adding a leading space on the next line, and strikethrough leaking into wrapped-line padding
Fixed
/treebranch summaries including failed assistant replies and original content that context edits had hidden or replaced. Summaries now honor the same context edits as live requests and compaction
Removed
- Breaking
Removed the native OS
sandbox/extension example and its dependency. Use the Gondolin example or the containerization guide.
1.4.0
21 Oct 2026Image input for classifiers: codemode
models.classify()acceptsimagesfor TypeSafe's Jev and Cloudflare's Clef models ontypesafe,cloudflare-workers-ai, and OpenRouter. See Use classifier models and Codemode.Models Configuration reference: every provider, model, model override, and compatibility field of
models.json, generated from its schema. See Models Configuration.Summaries without a fixed output cap: compaction and branch summaries use the model's output limit, so long reasoning no longer makes them fail. See Compaction.
New Features
Image input for classifiers: codemode
models.classify()acceptsimagesfor TypeSafe's Jev and Cloudflare's Clef models ontypesafe,cloudflare-workers-ai, and OpenRouter. See Use classifier models and Codemode.Models Configuration reference: every provider, model, model override, and compatibility field of
models.json, generated from its schema. See Models Configuration.Summaries without a fixed output cap: compaction and branch summaries use the model's output limit, so long reasoning no longer makes them fail. See Compaction.
Codemode waits for a free worker: parallel scripts and in-process subagents queue instead of failing, and
PI_CODEMODE_MAX_WORKERSraises the limit. See Codemode limits.Rex terminal support: OSC 8 hyperlinks, truecolor, and inline images.
Added
Added a Models Configuration reference generated from the
models.jsonschema, listing all provider, model, model override, and compatibility fieldsCodemode
models.classify()now acceptsimagesfor TypeSafe's Jev and Cloudflare's Clef models ontypesafe,cloudflare-workers-ai, and OpenRouter, and lists Clef Omni and Jev 1.14 on Workers AI.Provider-level
compatinmodels.jsonnow also applies to classifier models, with the System One fieldsimageInputandsendImageOptInHeaderfor proxies that serve another service's classifiers. Extension classifier models acceptcompattoo.Added
MiniMax-M3.1-Flashto theminimaxandminimax-cnproviders, with thinking levels low through maxAdded support for the Rex terminal: OSC 8 hyperlinks, truecolor, and inline images (#11131, #11150
Changed
Provider attribution headers now come from pi-ai's client identity. Extensions'
before_provider_headershandlers no longer see the OpenRouter, NVIDIA NIM, and OpenCode attribution headers; returningnullfor one still removes it. Cloudflare requests send pi's detailedUser-Agentinstead ofpi-coding-agent, as do GitHub release checks for tool downloads, and OpenCode requests sendx-opencode-clientwhenever attribution is enabled, not only with a session.generateSummary()andgenerateSummaryWithUsage()now ignore theirreserveTokensparameter. Summary requests use the model's output limit instead.Changed the default GitHub Copilot model from
gpt-5.4, which Copilot retired, togpt-6.1-sol.Changed the default NVIDIA model to
nvidia/nemotron-3.5-super-vl-120b-a12bChanged the default Xiaomi model to
mimo-v2.6-profor thexiaomiand Xiaomi Token Plan providers, sincemimo-v2.5-proleft the catalog.Grok 4.3 and Grok 4.5 on the
xaiprovider now offer thexhighthinking level.Sign in with ChatGPT: ChatGPT Business workspace usage-limit errors are not retried and tell the user to ask their workspace admin to increase the limit.
Fixed
Fixed compaction and branch summaries failing with "generation hit the token cap" when reasoning used most of the fixed summary output limit. Summaries now use the model's output limit, and
compaction.reserveTokensonly controls when compaction starts (#9512, #9075).Fixed
models.schema.jsonmissing descriptions for provider and model fields, so editors showed no hints for themFixed
pi updatefailing with "Managed install marker is missing or invalid" whenmanaged-install.jsonstarts with a UTF-8 BOM, which Windows PowerShell 5.1 adds when it rewrites the file.Fixed HTML export showing literal asterisks for Markdown bold and italic next to CJK punctuation (#3353, #10154)
Fixed the terminal staying in mouse-tracking, focus-reporting, Kitty keyboard, and alternate-screen modes when pi exits without its normal terminal teardown, which flooded the shell prompt with escape sequences
Fixed concurrent extension
select/confirm/input/editordialogs and inlinecustom()components in interactive mode: a second dialog replaced the first and the first promise never settled; closing the visible dialog now shows the previous one again (#11166, #9909)Fixed pasting in the shell hanging after copying text in pi and suspending it with Ctrl+Z on Linux: the background
wl-copy/xclip/xselprocess that keeps serving the clipboard was suspended along with piFixed codemode scripts failing immediately when both codemode workers are busy, for example with parallel codemode calls or in-process subagents. Scripts now wait up to 60 seconds for a free worker, and
PI_CODEMODE_MAX_WORKERSraises the limit of two workersFixed extensions failing to load when importing
@earendil-works/pi-ai/models(e.g. forclampThinkingLevel) by exposing the subpath through the extension loader aliases and virtual modulesFixed
mcp.jsonfiles that start with a UTF-8 BOM failing to parse, which skipped all their MCP servers and madepi mcp add/pi mcp removefailFixed
/compact <instructions>ignoring the instructions when summarizing the start of a turn that compaction splits. When the split turn was the only new content since the last compaction, the instructions were ignored entirelyFixed resumed sessions and
/treenavigation leaving extension tools inactive when the branch recorded its tools before they were registered, such as tools from extensions installed after the session started. Extension tools the branch never added or removed are now active unless they are registered withdefaultActive: false(#11062, #11075)Fixed sessions breaking after recovery from a response cut off at the output limit: the truncated tool call was omitted but its tool results were kept, so providers rejected every later request (#11227, #11233
Fixed the
greptool followingRIPGREP_CONFIG_PATH, where options such as--fixed-stringssilently changed regex searches into literal ones. The tool now runsrg --no-config(#11203, #11230Fixed Windows shell detection rejecting runnable Microsoft Store shell aliases
Fixed Windows shell detection missing PowerShell outside
PATHand the legacy WSLbash.exelauncher when Windows is not installed inC:\Windows. pi now checks%ProgramFiles%\PowerShell\7,%SystemRoot%\System32\WindowsPowerShell\v1.0, and%SystemRoot%Fixed GitHub Copilot sessions with large request bodies, such as many images, failing with HTTP 413
failed to parse requestinstead of compacting and retryingFixed the
openai-codexprovider offeringgpt-5.5andgpt-5.3-codex-spark, which always fail for ChatGPT accountsFixed whitespace-only tool output (for example a bare
\r\nfrom bash) making OpenAI Completions, OpenAI Responses, and Google requests fail with a 400Fixed automatic retry not triggering for more gateway upstream failures,
stream disconnected before completionerrors relayed by Codex CLI proxies, and transient Google Vertex OAuth token refresh failures, and retrying errors because of digits inside request ids or byte counts (#10920, #11057, #10996)Fixed OpenAI Responses models losing the prompt cache on every deferred tool load after a tool was removed
Fixed Cloudflare Workers AI classifier pricing: Clef Flash costs $0.038 and Jev $0.042 per million input tokens instead of Jev being listed as free.
Fixed an unclosed
$$display-math block swallowing or corrupting every Markdown table after itFixed Markdown bold and italic not rendering next to CJK punctuation
Fixed
@file autocomplete returning no results for queries containing regex metacharacters such as[or(
1.3.0
15 Oct 2026Choose where
/shareuploads:sessionShareDestinationselectsauto,gist, orradius. See Sessions and context and Export or share a session.Cache-preserving thinking level changes on OpenAI GPT-6: changing the thinking level mid-session with an API key keeps the prompt cache.
Kimi Code over Chat Completions: tools added mid-conversation no longer invalidate the prompt cache.
models.jsonbase URL overrides must now include/v1. See Providers.
New Features
Choose where
/shareuploads:sessionShareDestinationselectsauto,gist, orradius. See Sessions and context and Export or share a session.Cache-preserving thinking level changes on OpenAI GPT-6: changing the thinking level mid-session with an API key keeps the prompt cache.
Kimi Code over Chat Completions: tools added mid-conversation no longer invalidate the prompt cache.
models.jsonbase URL overrides must now include/v1. See Providers.Clean model switching: thinking from a previous model is dropped instead of being shown to the new model as text.
Resumed sessions keep their active tools:
--tools +name/-nameapplies on top of them. See Tools.
Added
Changing the thinking level mid-session on OpenAI GPT-6 models (API key) now keeps the prompt cache
Added the
sessionShareDestinationsetting to choose automatic, GitHub gist, or Radius uploads for/share
Changed
Changed the
kimi-codingprovider to use OpenAI Chat Completions athttps://api.kimi.com/coding/v1, the protocol used by the official Kimi Code client. K3, K3-256K, and Kimi For Coding now accept tools added mid-conversation without invalidating the prompt cache.models.jsonoverrides of thekimi-codingbase URL must now include/v1Changed model switching to drop thinking blocks from earlier assistant messages produced by a different model, instead of sending them as plain text. The new model previously saw the old model's reasoning as part of its replies and could start writing its own reasoning into the visible answer
Fixed
Pinned internal Pi dependencies to exact release versions to prevent fresh installs from mixing incompatible packages
Fixed resumed sessions dropping tools activated during the session. Resume now restores the session's active tools, with
--tools +name/-nameentries applied to them, unless--toolswith tool names,--no-tools, or--no-builtin-toolsselects tools explicitlyFixed stopping a codemode script or reaching its deadline waiting for a non-interruptible built-in to return on Bun
Fixed repeated codemode timeouts or aborts accumulating surviving workers; admission now limits live workers to two across sessions, counting cancelled workers until actual exit
Fixed
--modelchoosing a provider by catalog order when several providers offer the match; Pi now prefers the provider with credentials and otherwise reports the ambiguity.--modelsandenabledModelsadd a model for each provider that offers it and warn about the ambiguityFixed
readline-range labels for numeric-string or null arguments in the TUI, session tree, HTML exports, and examplesFixed the Nix flake build failing when a provider switched APIs: the model catalog pin is now refreshed from the newest published catalog when its model API groups differ, and releases pin the newest catalog and build with it before tagging
Fixed the Nix package putting its Node.js first on
PATH, which made bash tool shells use Pi's Node instead of the user'snode/npm. Pi's Node is now only a fallbackFixed prompt templates dropping empty quoted arguments (
""or''), which shifted later positional arguments and prevented positional defaults from applyingFixed the Nix package build failing in its install check because it still expected chord and esbuild, which Pi no longer depends on
Fixed tool-call argument parsing dropping fields when a string contains an invalid
\uescape such as\usepackageFixed Anthropic and Bedrock Claude requests failing with a 400 when an error tool result contains images; the images are now sent after the tool results of the turn, so the model still sees them
Fixed OpenCode Go Gemini models failing with a 404 by moving OpenCode Zen and OpenCode Go to the
opencode.ai/inferenceendpoints, and added Google Generative AI support to the OpenCode Go providerFixed OpenCode Go MiniMax M2.7 failing with "Model does not support this protocol" by using the Anthropic Messages API
Fixed automatic retry not triggering for gateway
Unable to forward upstream streamerrorsFixed automatic retry not triggering for
The service is temporarily unavailableerrors, which also made compaction fail on the first attemptFixed OpenAI-compatible responses showing thinking after the answer when a proxy such as LiteLLM sends reasoning and content in the same stream chunk
Fixed Gemini 4 Argon requests on Google Vertex failing with "Thinking budget is not supported", and its thinking level being ignored on the Gemini API
Fixed Xiaomi MiMo models failing with a 400 at the
minimalthinking level; they now expose only theoffandhighlevelsFixed
openai-codex/gpt-6.1-solusing a 272K context window instead of the 500K window Codex now providesFixed tool calls whose final arguments are malformed JSON silently running with only the fields parsed before the error; they now fail validation and show the model the raw arguments it sent
Fixed Claude Haiku 5.5 and Claude Opus 5 not supporting the
offthinking levelFixed
done()of an extension'sctx.ui.custom({ overlay: true })closing the topmost overlay instead of its own, which left the finished overlay on screen and another extension's dialog unresolved (#10648, #10999
1.2.0
12 Oct 2026- Breaking
Published generated configuration schemas for models, settings, keybindings, and themes at
https://pi.dev/schemas/*.schema.json. The theme schema moved tohttps://pi.dev/schemas/theme.schema.json; themes loaded during startup now receive the same schema validation as later reloads. Update theme$schemareferences, define reusable colors undervars, and remove unsupported metadata (#9880, #10751 - Breaking
Changed
before_agent_startto also fire for runs started bypi.sendMessage(..., { triggerTurn: true })while idle.BeforeAgentStartEventis now a union onmessageType:"user"events carrypromptandimages, and"custom"events carry the startingmessageand noprompt. Messages returned by handlers are added to both kinds of run, and pendingdeliverAs: "nextTurn"messages are delivered with the custom-message run. Handlers that should act only on user prompts must checkevent.messageType === "user"(#10267, #5581, #10739 - Breaking
Changed the session to count as busy while a run starts, including during
before_agent_starthandlers:ctx.isIdle()is false,pi.sendMessage()queues into the starting run instead of starting a second run that failed with "Agent is already processing",pi.sendUserMessage()requiresdeliverAs, and an abort requested meanwhile aborts the run once it starts
New Features
Per-key OpenRouter model lists: with an OpenRouter key, Pi lists only the models that key may use, with the key's context size, max output, and prices, including regional endpoints set in
models.json. See OpenRouter.Published configuration schemas: add
$schematosettings.json,models.json, keybindings, and themes for editor validation and completion. See Settings, Models, Keybindings, and Themes.Clipboard copies over remote sessions:
PI_CLIPBOARD_OSC52=1sends copied text to a remote client's clipboard when Pi cannot detect the remote session, such as a multiplexer attached from another machine. See Pi Process Configuration.Codemode improvements: tool declarations show input constraints such as integer, minimum/maximum, and default, and codemode rows show elapsed and total time. See Codemode.
before_agent_startfor extension-triggered runs: runs started bypi.sendMessage(..., { triggerTurn: true })go through the same setup as prompts. See Events and concurrency.
Breaking Changes
- Breaking
Published generated configuration schemas for models, settings, keybindings, and themes at
https://pi.dev/schemas/*.schema.json. The theme schema moved tohttps://pi.dev/schemas/theme.schema.json; themes loaded during startup now receive the same schema validation as later reloads. Update theme$schemareferences, define reusable colors undervars, and remove unsupported metadata (#9880, #10751 - Breaking
Changed
before_agent_startto also fire for runs started bypi.sendMessage(..., { triggerTurn: true })while idle.BeforeAgentStartEventis now a union onmessageType:"user"events carrypromptandimages, and"custom"events carry the startingmessageand noprompt. Messages returned by handlers are added to both kinds of run, and pendingdeliverAs: "nextTurn"messages are delivered with the custom-message run. Handlers that should act only on user prompts must checkevent.messageType === "user"(#10267, #5581, #10739 - Breaking
Changed the session to count as busy while a run starts, including during
before_agent_starthandlers:ctx.isIdle()is false,pi.sendMessage()queues into the starting run instead of starting a second run that failed with "Agent is already processing",pi.sendUserMessage()requiresdeliverAs, and an abort requested meanwhile aborts the run once it starts - Breaking
Changed
pi mcp list --jsonto reportresources: truefor servers that offer resources instead ofresourcesandresourceTemplatescounts; Pi no longer lists resources when connecting
Added
Added
reachesToolsto tool definitions, so tools that reach other tools (such ascodemodeandtool_search) declare which tools a call may reach; the MCP extension uses it to wait for the servers a call needs and to decide whether a replacementcodemodeortool_searchtool can reach MCP toolsAdded
aliasesto tool namespaces; codemode'sdescribeNamespace()and thenamespaceoption ofsearchTools()match them, and MCP namespaces use the server nameAdded per-key OpenRouter model lists: with an OpenRouter key, Pi lists only the chat models that key may use, with the key's context size, max output, and prices, including guardrails and regional endpoints set as the base URL in
models.json. The list is saved and used offline (#10353, #10672Added
PI_CLIPBOARD_OSC52=1to send copied text to the terminal with OSC 52 after the local clipboard write, so copies reach a remote client's clipboard when Pi cannot detect the remote session, such as multiplexer attaches from another machine or remote IDE daemons (#10826, #10835, #10114)Added elapsed time while a codemode script runs and its total time once finished to the codemode tool row, like bash rows
Changed
Changed the
showHardwareCursorsetting to use only the terminal cursor instead of also drawing Pi's reverse-video cursorChanged codemode tool declarations to show input constraints such as integer, minimum/maximum, length, pattern, and default as
@tags on property comments, so scripts see them without the tool repeating them in descriptionsChanged codemode declarations of MCP tools to spell out the result type (
content,structuredContent,isError) instead of referring toCallToolResult<T>, whichdescribeTool(),searchTools(), andALL_TOOLSshowed without defining it; the codemode description no longer includes the shared MCP typesChanged
serializeConversation()(compaction, branch summaries, bug reports) to omit thinking blocks; replayed thinking is often not in the live context, so including it could make a summarization request larger than the context it summarizes and fail
Fixed
Fixed codemode ignoring a
// @options:line that is not the first line of the script, which ran the script without the requestedtimeout_ms; the call now fails with an error before the script runsFixed Mistral streams being aborted mid-generation (e.g. during long thinking) after the request timeout; the timeout now applies only to waiting for response headers
Fixed a symlinked
AGENTS.mdin a git worktree nested inside its main repo causing no project context file to loadFixed the code mode tool description claiming every failed tool call rejects; a tool that declares an output schema resolves with its value on failure too (e.g.
bashwith a nonzero exit)Fixed runs started by
pi.sendMessage(..., { triggerTurn: true })removing system prompt sections added inbefore_agent_startafter their first tool call, which changed the prompt mid-run and broke the provider prompt cache twice per run (#10267, #10739Fixed codemode blocking the process for seconds to minutes after scripts with many
text()calls; joining the output items was quadraticFixed print mode (
-p) exiting 0 without a warning when the final response was cut off at the model's output token limit; it now prints the partial text, writes a warning to stderr, and exits 1Fixed
pi mcp login --helpand the CLI docs describing--timeoutas only the wait for the browser; it limits the whole sign-inFixed
~not being expanded in!pattern,+path, and-pathentries of theextensions,skills,prompts, andthemessettings, so entries like!~/skills/pdfmatched nothingFixed MCP servers with resources registering no tools until every resource was listed; Pi no longer lists resources when connecting, so
/mcpandpi mcp listshow that a server offers resources instead of counting themFixed MCP tool calls failing when the server returns structured content that is not an object (an array,
null, or a primitive, as MCP 2026-07-28 allows); codemode declarations of MCP tools without an output schema now typestructuredContentasunknowninstead of an objectFixed reading or writing
auth.jsonon Windows failing withEPERMwhen the previous lock folder was still being deleted; Pi now retries briefly before reporting the errorFixed OpenRouter router models such as
openrouter/autoshowing a cost of -$1,000,000 per 1M tokens; OpenRouter's negative placeholder prices are now 0Fixed Anthropic requests failing when the provider rejects the strict tool set for grammar size, tool count, or union limits; Pi now retries once with preferred tools non-strict
Fixed "The upstream provider returned an error" not being retried as a transient provider error
Fixed focused fullscreen overlays blocking mouse-wheel scrolling of the transcript
Fixed fragmented terminal color replies leaking hex values into the prompt and opening the external editor (related reports: #10256, #10657)
Fixed inline images not rendering in Herdr since 1.1.0
Fixed key bindings that name a shifted symbol, such as
+orctrl++, not matching under the Kitty keyboard protocol or modifyOtherKeysFixed the compaction status indicator embedded in a custom editor using the default muted colors instead of the editor border or thinking-level color, as the working indicator does (#10809, #10810
Fixed MCP tool calls that were cancelled or timed out while waiting for an OAuth refresh still being sent to the server once the refresh finished
Fixed MCP OAuth sign-in accepting an authorization response with an
issparameter that could not be checked because no authorization server metadata was discovered; sign-in now fails with an issuer mismatch. ReportedFixed codemode failing with a broken bridge error when Pi runs under
node --watchon Node 24 or 26
Removed
- Breaking
Removed the unreleased experimental server, client, and plugin stack (
src/experimental,src/cli/experimental) and its source-only./clientand./experimental/pluginexports; none of it was part of the published package. The@earendil-works/pi-client,@earendil-works/pi-protocol, and@earendil-works/pi-serverpackages it used are removed and no longer released - Breaking
Removed the dependency on
@earendil-works/chord
1.1.0
7 Oct 2026Program status reporting: terminals and agent dashboards that support OSC 7501 see whether Pi is working, blocked on a dialog or login, done, or failed. See Program status.
Claude Haiku 5.5:
anthropic/claude-haiku-5-5, with adaptive thinking up toxhigh/maxeffort.Adjust default tools with
+name/-name:--toolsentries likepi -t +codemode,-writechange the default selection instead of replacing it. See Tools.
New Features
Program status reporting: terminals and agent dashboards that support OSC 7501 see whether Pi is working, blocked on a dialog or login, done, or failed. See Program status.
Claude Haiku 5.5:
anthropic/claude-haiku-5-5, with adaptive thinking up toxhigh/maxeffort.Adjust default tools with
+name/-name:--toolsentries likepi -t +codemode,-writechange the default selection instead of replacing it. See Tools.GPT-6 Luna and image classification: OpenAI's GPT-6 Luna is available as a classifier model through the Decisions API, and codemode's
models.classify()accepts images for classifiers that support them. See Use classifier models.Native llama.cpp decision models: Julia-1, Laya, Kev, lev, and OpenJev served by llama.cpp 0.6.0 or later run natively as classifiers through
/v1/systemone. See Classification.
Added
Added
+nameand-nameentries to--tools, which change the default tool selection instead of replacing it, for examplepi -t +codemodeAdded
durationMsto the tool render context and totool_execution_endextension events: the recorded execution time of a final tool resultAdded
outputPadto the tool render contextAdded OpenAI's GPT-6 Luna as a classifier model through the Decisions API, available with
OPENAI_API_KEY(see Use classifier models)Added
imagesto codemode'smodels.classify()context, so classifiers that accept images, such as GPT-6 Luna, can judge themAdded program status reporting with OSC 7501: terminals and agent dashboards that support it see whether Pi is working, blocked on a dialog or login, done, or failed.
PI_PROGRAM_STATUS=1|0overrides detection (see Terminal setup)Added
abortedtoagent_settledsession, extension, and JSON events, so integrations can tell a cancelled run from a finished oneAdded Claude Haiku 5.5 (
anthropic/claude-haiku-5-5), with adaptive thinking up toxhigh/maxeffort and prompt caching on BedrockAdded native llama.cpp decision models: Julia-1, Laya, Kev, lev, and OpenJev served by llama.cpp 0.6.0 or later are listed only as classifiers through
/v1/systemoneinstead of as chat models (see Classification)
Changed
Changed
outputPadto also apply to!command output, tool output, and summary blocks (#9946, #10557Changed
pi mcp login --timeoutto limit the whole sign-in, including requests to the authorization server, instead of only the wait for the browser
Fixed
Fixed bash and PowerShell results losing
Tookafter reloading a session, and the liveTookincluding wall-clock steps; both now show the recorded execution timeFixed managed installs keeping every old release;
pi updatenow keeps only the new release and the one it updated from (#10392, #10511Fixed standalone binaries loading
.env,.env.local, and.env.developmentfrom the launch directory into Pi's environmentFixed
!!command headers losing their dim color once output arrivesFixed the codemode description not marking
searchTools(),describeTool(), anddescribeNamespace()as async, which led models to serialize the unawaited promise as{}Fixed codemode output items running together, so models could not tell where one
text()orconsole.log()output ended and the next began. With several text items, each now starts with a==> text N/M <==line, andconsolecalls follow the other output in one<console_output>block with one line per callFixed
/mcpwaiting for all servers to connect before opening; the manager now updates live and remains usable while enabling, reconnecting, or disabling serversFixed images being dropped as "could not be resized" when running under
node --watchon Node 24.19+ and 26.x, where Node posts its own messages on the image resize worker channelFixed clipboard paste doing nothing in Termux, and failed copies there omitting the Termux:API install hint
Fixed
!and RPCbashoutput keeping fragments of color codes, such as a straym, when a code was split across output chunksFixed MCP OAuth sign-ins that could not be cancelled while waiting on the authorization server and kept running after the session ended. The sign-in screen now cancels with Esc at every step, session shutdown aborts a running sign-in, and each request to the authorization server times out after 15 seconds
Fixed shutdown waiting up to 15 seconds to refresh an MCP OAuth token that was about to expire, only to close the server's session
Fixed the fullscreen text selection surviving session switches and other transcript rebuilds, which highlighted unrelated text in the new transcript (#9311, #10567
Fixed OpenAI models on Bedrock ignoring the thinking level and always running at Bedrock's default reasoning effort (#9331, #10142
Fixed model
headersinmodels.jsonnot overriding theoriginatorandUser-Agentheaders of Codex requestsFixed
server_busyandservers are currently busyprovider errors ending the turn instead of being retriedFixed Mistral responses that end with
finish_reason: "error"not being retriedReduced context-limit request failures by estimating input at 3.5 characters per token instead of 4 when calculating output limits
Fixed Radius models disabled by an organization owner still being listed
Fixed Anthropic browser login failing with "localhost refused to connect" when port 53692 is reserved or in use, for example by Hyper-V/WSL port exclusions on Windows: login now falls back to a free loopback port
Fixed session costs undercounting long prompts on models with prompt-length pricing tiers, such as Claude Haiku 5.5, Gemini 3.1 Pro, and GPT-5.4, through OpenCode, OpenCode Go, OpenRouter, Vercel AI Gateway, Google, MiniMax, and other providers
Fixed Markdown links not being clickable in Herdr
1.0.4
5 Oct 2026Codemode persists images:
tools.read()on an image file now gives back an image block thatimage()can show. See Call tools.Added
*patterns to--toolsand--exclude-tools, for example--tools read,codemode,'mcp__radius__*'
New Features
Tool patterns and
--no-mcp:--toolsand--exclude-toolsaccept*patterns, for example--tools read,codemode,'mcp__radius__*'keeps only one MCP server's tools.--toolsnow keeps MCP tools unless an entry starts withmcp__, and--no-mcpturns off MCP for one run. See Tools and MCP tools.Codemode persists images:
tools.read()on an image file now gives back an image block thatimage()can show. See Call tools.
Added
Added
*patterns to--toolsand--exclude-tools, for example--tools read,codemode,'mcp__radius__*'Added
--no-mcpto disable the built-in MCP support for one run
Fixed
Fixed syntax highlighting losing colors after the first line of multiline strings and comments in fenced code blocks
Fixed codemode scripts not receiving images from
read:tools.read()now resolves to an image block for image files, whichimage()showsFixed MCP OAuth sign-in failing with
invalid_redirect_urion servers with OpenID Connect client registration, such asmcp.modem.dev: pi now registers as a native clientFixed
--toolsremoving MCP tools, which leftpi --tools codemodewithout any MCP servers.--toolsnow keeps MCP tools unless an entry starts withmcp__Fixed MCP session shutdown returning while a server was still connecting, leaving its transport open until the server answered or timed out
Fixed system prompt rules and the skills hint naming tools hidden by
prepareLoadout. Hidden tools are left out of the rules, the skills hint names no tool when the file reader is hidden, andcodemodeshows each tool's prompt guidelines with its declaration;ToolLoadoutgainsgetPromptGuidelines()Fixed Bedrock requests that fail with
The pending stream has been canceledafter a stalled HTTP/2 connection not being retried automaticallyFixed codemode scripts that patch built-ins (for example
Array.prototype.toJSON = ...) crashing pi and leaving the tool call unsettled. Built-ins are now frozen before the script runs, so such patches have no effect
1.0.3
5 Oct 2026- Breaking
Renamed the Azure provider from
azure-openai-responsestoazure. Rename the provider key inauth.json(or run/loginagain), inmodels.json, and insettings.json(defaultProvider,enabledModelspatterns, andmodelThinkingLevelskeys). Sessions that used the old provider fall back to another model when resumed, and their prompt cache is not reused. TheAZURE_OPENAI_*environment variables are unchanged Azure Foundry Chat Completions — The
azureprovider (renamed fromazure-openai-responses) now also serves Foundry Chat Completions deployments, starting withazure/deepseek-v4-pro. See Azure OpenAI.Codemode images saved to files —
image()also writes each image to a temp file and names the path in the result, so later turns can copy or move generated images. See Generate images.
New Features
Azure Foundry Chat Completions — The
azureprovider (renamed fromazure-openai-responses) now also serves Foundry Chat Completions deployments, starting withazure/deepseek-v4-pro. See Azure OpenAI.Codemode images saved to files —
image()also writes each image to a temp file and names the path in the result, so later turns can copy or move generated images. See Generate images.
Breaking Changes
- Breaking
Renamed the Azure provider from
azure-openai-responsestoazure. Rename the provider key inauth.json(or run/loginagain), inmodels.json, and insettings.json(defaultProvider,enabledModelspatterns, andmodelThinkingLevelskeys). Sessions that used the old provider fall back to another model when resumed, and their prompt cache is not reused. TheAZURE_OPENAI_*environment variables are unchanged
Added
Changed
Codemode
image()now also saves each image to a temp file and names the path in the result, so later turns can copy or move generated imagesOutput files (full text of truncated tool output, binary MCP resources, codemode images) are now readable only by the user
Home/Endnow always move the editor cursor to the line start/end; fullscreen transcript top/bottom moved toCtrl+Home/Ctrl+End, which no longer move the editor cursor
Fixed
Fixed subscription logins such as Sign in with ChatGPT failing with
refresh_token_invalidatedafter a request was cancelled during an OAuth token refreshFixed codemode failing for the rest of a session after a pnpm global update removed the running install, and added a restart hint when errors occur after pi was updated or removed on disk
Fixed interactive sessions reporting a
read EIOorsetRawMode EIOcrash (and asking to run /bug) when the terminal went away, e.g. after closing the window or resuming a suspended pi in a closed terminal
1.0.2
4 Oct 2026Sampling by thinking level —
samplingParamsByThinkingLevelinmodels.jsonsets sampling parameters such astemperatureandtop_pfor each thinking level on OpenAI-compatible APIs. See Configure sampling by thinking level.Added
samplingParamsByThinkingLeveltomodels.jsonfor per-thinking-level sampling parameter overrides on OpenAI-compatible APIs. See Configure sampling by thinking level
1.0.1
3 Oct 2026- Security
Fixed installations resolving vulnerable
brace-expansion5.0.9 by pinningbrace-expansion5.0.12 as a direct dependency (GHSA-q2hr-2g5m-vwhr, GHSA-qhr7-859c-m2p7, GHSA-6j4f-fj2g-mc7p) - Breaking
Removed
npm-shrinkwrap.jsonfrom the published package. npm installations no longer pin transitive dependencies, and library consumers can now override them. Use the pi.dev installer for pinned installations Nix flake —
nix run github:earendil-works/pi/stableruns the latest release, andnix profile add github:earendil-works/pi/stableinstalls it. See Install pi.
New Features
Nix flake —
nix run github:earendil-works/pi/stableruns the latest release, andnix profile add github:earendil-works/pi/stableinstalls it. See Install pi.Project overrides for MCP servers —
.pi/mcp.jsonand/mcpcan enable, disable, or change the exposure of a user-level server for one project. See Configure servers.MCP Client ID Metadata Documents —
oauth.clientRegistration: "cimd"lets authorization servers allow pi by its document URL instead of dynamic registration. See Authenticate with OAuth.Tool renderers for any tool —
pi.registerToolRenderer()draws calls to tools that are not registered yet, such as MCP tools in resumed sessions. See Tool rendering.Cloudflare Clef classifiers —
@cf/cloudflare/clefand@cf/cloudflare/clef-flashare usable from codemode scripts and extensions. See Use classifier models.
Added
Added a copy key (
app.message.copy, defaultctrl+x) to OAuth sign-in screens in/login,/mcp, and/mcp login, which copies the sign-in URL when the browser cannot be opened or the wrapped link cannot be selected.Added
oauth.clientRegistration: "cimd"for MCP servers, which identifies pi with its Client ID Metadata Document on pi.dev instead of dynamic client registration, so authorization servers can allow pi by URLAdded project overrides for user-level MCP servers: a
.pi/mcp.jsonentry withoutcommandorurlsets onlyenabled,exposure, andtoolExposureof the user-level server, and/mcpcan enable or disable a server for the current projectAdded Cloudflare's Clef and Clef Flash classifier models to
cloudflare-workers-ai, usable from codemode scripts and extensionsAdded
pi.registerToolRenderer(), which chooses how calls to a tool are drawn, including tools that are not registeredAdded a Nix flake for macOS and Linux:
nix run github:earendil-works/pi/stableruns the latest release, andnix profile add github:earendil-works/pi/stableinstalls it. See Install pi
Changed
pi updateon global npm installations now recommends migrating to the managed installation from the pi.dev installer, which pins all dependencies.Anthropic tools added or redefined mid-conversation are now defined inline in the conversation, so redefining a tool under the same name keeps the prompt cache instead of resending the full tool list.
Fixed
- Security
Fixed installations resolving vulnerable
brace-expansion5.0.9 by pinningbrace-expansion5.0.12 as a direct dependency (GHSA-q2hr-2g5m-vwhr, GHSA-qhr7-859c-m2p7, GHSA-6j4f-fj2g-mc7p) Fixed a trailing comma in
--modelsadding an extra model to the model cycleFixed a
codemodescript that prints in a loop crashing pi by running out of memory: a script fails once its output passes 16 Mi characters or 100000 itemsFixed JPEG, GIF, and WebP images rendered by extensions through
Imagenot appearing in Kitty, Ghostty, WezTerm, and WarpFixed MCP tool calls in resumed sessions and HTML exports rendering fully expanded until their server connected, or for good if it never did
Fixed fullscreen Kitty images collapsing to a one-row strip after scrolling in WezTerm
Fixed "Selected model is at capacity" provider errors ending the turn instead of being retried
Fixed Cloudflare AI Gateway Claude models failing with a 404 by using dashed model IDs (
claude-opus-5-5instead ofclaude-opus-5.5)Fixed Sign in with ChatGPT continuing when its callback port is taken by another login, which made the browser show "OAuth state mismatch"; it now fails with a port-in-use error
Fixed Amazon Bedrock OpenAI models costing requests above 272k input tokens at the short-context rate; Bedrock models now include the pricing tiers listed on models.dev
Fixed Amazon Bedrock Claude requests failing with "Invalid
signatureinthinkingblock" after the system prompt or tools changedFixed Together DeepSeek V4 Pro losing its thinking level controls after Together renamed it to
deepseek-ai/DeepSeek-V4-Pro-0813Fixed the default NVIDIA model pointing at
nvidia/nemotron-3-super-120b-a12b, which NVIDIA no longer serves; the default is nownvidia/nemotron-3-ultra-550b-a55b
Removed
- Breaking
Removed
npm-shrinkwrap.jsonfrom the published package. npm installations no longer pin transitive dependencies, and library consumers can now override them. Use the pi.dev installer for pinned installations
1.0.0
1 Oct 2026Fullscreen by default — The TUI now runs fullscreen. Set
tuiModeto"regular"to keep the terminal's normal scrollback. See Terminal and display.Leaner codemode — About 40% fewer prompt tokens, and errors that tell the model how to recover. See Codemode.
Image generation in codemode — Scripts call
models.generateImages()with the session's credentials. See Generate images and Use image models.
New Features
Fullscreen by default — The TUI now runs fullscreen. Set
tuiModeto"regular"to keep the terminal's normal scrollback. See Terminal and display.Leaner codemode — About 40% fewer prompt tokens, and errors that tell the model how to recover. See Codemode.
Image generation in codemode — Scripts call
models.generateImages()with the session's credentials. See Generate images and Use image models.Radius in
/login— Sign in with Radius and set up its MCP server in one step. See Radius.Anthropic copy code login — Sign in when the browser runs on another machine. See Authenticate interactively.
MCP OAuth hardening —
oauth.authServerMetadataUrl, RFC 9207isschecks, credentials per server, and step-up sign-in that keeps granted scopes. See Authenticate with OAuth.Header-only quiet startup —
quietStartup: "header"keeps the version and key hints and hides the rest. See Terminal and display.
Added
Added an
oauth.authServerMetadataUrlsetting for MCP servers that advertise a wrong OAuth authorization server or none. Pi uses the configured metadata document instead of discovery (#10172).Added
quietStartup: "header", which keeps the startup header with version and key hints but hides the model scope line and loaded-resource listing.Added
models.generateImages()to codemode scripts. It runs image models such as OpenRouter's with the session's credentials and returns base64 image blocks thatimage()attaches to the result; usage counts toward the session cost likemodels.classify(). Extensions can callctx.modelRegistry.generateImages(). See Use image models.Added a copy code login method to Anthropic
/loginfor headless setups where the browser runs on another machine
Changed
Changed the default TUI mode to fullscreen. Set
tuiModeto"regular"or pass--tui-mode regularto keep the terminal's normal scrollback./loginnow offers "Sign in with Radius" at the top level, as the last option, with its status. After a Radius sign-in,/loginoffers to configure the Radius MCP server in the globalmcp.jsonwith"auth": { "provider": "radius" }and reloads. Cancelling a login returns to the menu it was started from.The provider docs page is renamed to Providers, its "Cloud Providers" section is now "Provider Specific Config", and it documents Radius first.
MCP OAuth credentials are now stored per server name and URL, so MCP servers with the same URL can sign in with different accounts. Credentials stored by URL alone move to the first server that uses them (#10252).
Codemode costs far fewer prompt tokens: with the default tools and codemode active, a GPT-5.6 request shrinks from about 5,300 to 3,300 tokens. The
codemodedescription lists the script globals in one line each and points to the new Codemode reference for themodelsAPI, which the model reads when it needs it. Declared tools say in one line how scripts call them and what the call resolves to, instead of repeating their full declaration, and the system prompt's codemode guidance and MCP server section are shorter.Codemode errors now say how to recover: reading a tool or
modelsmember that does not exist names the close matches (tools.Bashsuggeststools.bash),models.classify()andmodels.generateImages()reject malformed arguments with the expected shape, an unknown model points tomodels.getAvailableOfType(), an oversizedstore()value explains what the store is for, and a script that generates images without showing them gets a note. Scripts that probed for a tool withtypeof tools.namemust use"name" in tools./loginand/logoutnow label providers without credentials as "not configured" instead of "unconfigured".OAuth browser pages now show the color Pi logo.
Fixed
Fixed MCP OAuth sign-in accepting an authorization response whose
issparameter names another authorization server; the code is now rejected before it is exchanged (RFC 9207).Fixed MCP OAuth sign-in failing with
Invalid scopewhen the token response contains"scope": "", and similar failures for other empty ornulloptional OAuth fields (#10266).Fixed the sign-in URL printed by
/mcp loginnot being clickable when it wraps (#10186).Fixed
--providerwithout--modelbeing silently ignored and running the default model from another provider; it now fails with an error (#10236).Fixed MCP servers that ask for more scope (
insufficient_scope) requesting sign-in over and over. The new sign-in requested only the missing scopes, so the new token lost access the previous one had; it now keeps the granted scopes.Fixed user messages in the transcript keeping two full-width copies of every rendered line; they keep one, with identical output.
Fixed
/loginand/logoutlabeling every OAuth sign-in, including Radius, as a subscription; only subscription-backed providers say "subscription", other OAuth sign-ins say "account".Fixed the startup header logo rendering with gaps in Apple Terminal; it now shows a colored "Pi" with the version instead.
Fixed deferred MCP tools that
tool_searchloaded being dropped on resume and/reloadeven when their server reconnected before the next prompt, because the session restored its tools before the MCP servers reconnected.Fixed the system theme making pastel palettes such as Catppuccin Frappe much more vivid; palette colors now keep their chroma (#10255, #10293
Fixed slash command autocompletion not triggering when the input starts with whitespace
Fixed color bleeding past mouse selections and search highlights in fullscreen mode when a styled token ends at the highlight boundary (#10169).
Fixed memory retained per rendered message in the transcript; a long assistant message keeps about a fifth of the heap it kept before.
- 0.99.x
0.99.2
30 Sep 2026MCP servers stay out of the way: servers with the default
codemodeexposure are no longer listed in thecodemodedescription and no longer block the first prompt. They appear in a short system prompt section, and scripts find their tools withsearchTools()anddescribeNamespace(). See Control tool exposure.More MCP authentication options:
oauth.clientNamefor servers that only accept known OAuth clients, and"auth": { "provider": "<provider>" }to authenticate HTTP servers with a provider's/logintoken. See Authenticate with OAuth.Anthropic workload identity federation from the Anthropic SDK environment variables. See Use an API key from the environment.
New Features
MCP servers stay out of the way: servers with the default
codemodeexposure are no longer listed in thecodemodedescription and no longer block the first prompt. They appear in a short system prompt section, and scripts find their tools withsearchTools()anddescribeNamespace(). See Control tool exposure.More MCP authentication options:
oauth.clientNamefor servers that only accept known OAuth clients, and"auth": { "provider": "<provider>" }to authenticate HTTP servers with a provider's/logintoken. See Authenticate with OAuth.Anthropic workload identity federation from the Anthropic SDK environment variables. See Use an API key from the environment.
/reloadenables tools newly added to thedefaultToolssetting. See Tools.
Added
Added a
descriptionfield for MCP servers (pi mcp add --description), shown with the server in the system prompt and used to rank its tools in tool search, and adescribeNamespace(name)codemode helper that returns a namespace's instructions and tool names.describeNamespace()andsearchTools()accept a namespace asmcp__dev-radius,mcp__dev_radius,dev-radius, ordev_radius.Added an
oauth.clientNamesetting for MCP servers (pi mcp add --oauth-client-name) to change the client name sent during OAuth client registration, for servers that only accept known clients (#10226).Added
"auth": { "provider": "<provider>" }for HTTP MCP servers to send a provider's current/logintoken as the bearer token instead of using MCP OAuth. The token is read on every request, so provider refreshes apply. Only allowed in the globalmcp.jsonand from extensions, and requires https except on loopback hosts.Added Anthropic workload identity federation from the
ANTHROPIC_FEDERATION_RULE_ID,ANTHROPIC_ORGANIZATION_ID, andANTHROPIC_IDENTITY_TOKEN_FILEenvironment variables (see Providers) (#10177, #10242/reloadnow enables tools newly added to thedefaultToolssetting. Tools removed from it stay enabled, tools turned off during the session stay off unless newly added, and--tools,--no-tools, and--no-builtin-toolsstill override the setting (#10245).
Changed
MCP servers with the default
codemodeexposure no longer appear in thecodemodedescription; scripts find them withsearchTools().codemode-deferredis now an alias forcodemode. Usedirectexposure for tools the model should see without searching (#10212).The
codemodedescription no longer includes deferred tools, tool counts, or MCP server instructions, so it no longer changes when MCP servers connect or change their tools. Thetool_searchdescription no longer lists the servers whose tools it can load, for the same reason. Servers are listed instead in anmcp_serverssystem prompt section with a one-line summary, updated at the start of each prompt; a changed section is appended to the conversation. Scripts read server instructions withdescribeNamespace()(#10212).The first prompt no longer waits for MCP servers without
directtools. They connect in the background and are waited for when a codemode script names them, a script searches tools, ortool_searchruns (#10212).
Fixed
Fixed new sessions intermittently ignoring the saved default model, or warning that no models are available, when it belongs to an extension-registered native provider with a stored credential (#9962, #10190
Fixed the
/mcpsign-in URL not being clickable when it wraps across lines, by emitting it as a terminal hyperlink with aCmd/Ctrl+click to openline like/login(#10186).Fixed codemode
image()accepting malformed base64 data or unsupported image types, which persisted an invalid image block that made every later provider request fail with HTTP 400 (#10215).Fixed codemode failing to start its script worker from the standalone Windows executable (#10204).
Fixed prompt submission slowing down with session length, because resolving the session's model selection looked up the model catalog once per assistant message (#10198).
Fixed model lookups slowing down for providers with a refreshed pi.dev catalog, because merging remote catalog models took quadratic time.
Fixed the
built-in-tool-renderer.tsandminimal-mode.tsextension examples removing the built-in tools' summaries and guidelines from the system prompt (#10072, #10193Fixed context overflow detection for Z.AI CN endpoint
Prompt exceeds max lengtherrors (#10208).Fixed Anthropic requests failing when a tool schema uses keywords Anthropic strict tool use rejects, such as
minimum/maximum; such tools are now sent non-strict (#9953).Fixed provider retries firing immediately when a
Retry-Afterheader contains an unparseable date; they now use exponential backoff (#9571).Fixed extension commands registered without a string name or handler crashing pi when typing
/; the extension now fails to load with an error instead (#10054).Fixed collapsed
codemodeand MCP tool results filling the screen when the output is one long line, such as minified JSON. Like bash output, the preview is now limited to wrapped lines instead of logical lines.Fixed
codemode.mode: "only"listingread,bash,edit, andwritein the system prompt's tool list although requests only declarecodemode(#10192).Fixed codemode scripts calling the wrong MCP tool when two tool names differ only in
-and_, such asread-fileandread_file. Like in Codex, MCP tool and namespace names now replace-with_(mcp__my-server__xis nowmcp__my_server__x), colliding tools of a server all get a hash suffix, and server names that differ only in-and_are rejected (#10239).
0.99.1
29 Sep 2026GPT-6.1 Sol — Available on OpenAI, Azure OpenAI, and OpenAI Codex, and now the default OpenAI Codex model. See Select a model.
Added GPT-6.1 Sol (
gpt-6.1-sol) to the OpenAI, Azure OpenAI Responses, and OpenAI Codex providers.Changed the default OpenAI Codex model to GPT-6.1 Sol (
gpt-6.1-sol).
New Features
GPT-6.1 Sol — Available on OpenAI, Azure OpenAI, and OpenAI Codex, and now the default OpenAI Codex model. See Select a model.
Added
Added GPT-6.1 Sol (
gpt-6.1-sol) to the OpenAI, Azure OpenAI Responses, and OpenAI Codex providers.
Changed
Changed the default OpenAI Codex model to GPT-6.1 Sol (
gpt-6.1-sol).
Fixed
Fixed
/loginwith OpenAI failing in the bundled release with a missingopenai-chatgpt.jsmodule error.
0.99.0
29 Sep 2026Codemode and MCP — Connect MCP servers and let models run JavaScript that calls tools in parallel. See MCP Servers and Enable codemode.
System theme — Pi's colors now come from your terminal's own palette by default. See Use your terminal's colors.
Sign in with ChatGPT — Use a ChatGPT subscription with the OpenAI provider through
/login openai. See Authenticate interactively.
New Features
Codemode and MCP — Connect MCP servers and let models run JavaScript that calls tools in parallel. See MCP Servers and Enable codemode.
System theme — Pi's colors now come from your terminal's own palette by default. See Use your terminal's colors.
Sign in with ChatGPT — Use a ChatGPT subscription with the OpenAI provider through
/login openai. See Authenticate interactively.Virtual models — Extensions can route each request to a different physical model. See Virtual Models.
Classifier models — Run Jev classifiers from codemode scripts, or use any llama.cpp model as a classifier. See How codemode works and Classification.
Added
Added codemode, tool search, and MCP support as built-in extensions. The
codemodetool runs model-written JavaScript in a QuickJS sandbox that calls pi's tools; enable it withdefaultToolsor--toolsand configure it withcodemode.modeandcodemode.inlineBudget.tool_searchfinds tools that are not declared to the model and declares them. MCP servers over stdio or streamable HTTP, with OAuth, come frommcp.json(global, or per project once trusted) orpi.registerMcpServer()and are managed with/mcpandpi mcp add|remove|list|login|logout. See MCP Servers and Enable codemode (#10040).Added extension tool APIs for orchestrating tools:
exposure(direct,model-only,codemode,deferred, orhidden),namespace,annotations,outputSchemawithstructuredContent,isErrorresults,prepareLoadout(), andctx.executeTool()for nested tool calls, which emit events withparentToolCallIdand are recorded as boundednestedCallson the calling tool's result. See Tool exposure.Added a warning when an extension that registers the same tool, command, or flag replaces a built-in extension
Added experimental virtual models: extensions register them with
pi.registerVirtualModel()and pick a physical model and thinking level for each request. The footer shows the routed model,/sessionlists cost per physical model, andexamples/extensions/jev-router.tsroutes with the Jev classifier. See Virtual Models.Added Sign in with ChatGPT to the OpenAI provider in
/login, which uses a ChatGPT subscription with the OpenAI API. Pi stores a stabledeviceIdin the global settings for this login and omits it from bug reports.Added the
systemtheme, now the default, which derives pi's colors from the terminal's reported foreground, background, and ANSI palette and rebuilds them when the terminal switches between light and dark. See Use your terminal's colors.Added
#rgb,oklch(), andokhsl()colors and an optionalappearancefield to theme files, andtheme.style(),theme.colors, andtheme.appearancefor extensions. See Themes and TUI.Added a classifier model for every llama.cpp chat model, answered from next-token label probabilities. See Classification.
Added inherited Jev classifier models on OpenRouter, Cloudflare Workers AI, Vercel AI Gateway, and OpenCode Zen.
Added the
fullscreenWheelScrollLinessetting and/settingsentry for fullscreen mouse-wheel scrolling. The default"auto"accelerates fast wheel spins outside local macOS terminals (#9758).Added per-input disposition to successful RPC
prompt,steer, andfollow_upresponses,AgentSession.steer()/followUp(), andRpcClient.prompt()/steer()/followUp();RpcClient.prompt()also acceptsstreamingBehavior(#9098, #9803).Added image generation to
ModelRuntime:generateImages()with runtime-resolved auth (stored credentials, OAuth, runtime API keys,models.jsonheaders), plusgetModelsOfType(),getModelOfType(),getAvailableOfType(),getAllModels(), andgetAllAvailable(). OpenRouter image models are listed under theopenrouterprovider and share its credential; an upstream ID can have separate chat and image entries.models.jsonproviders and extension registrations without a model list keep built-in image generation. Extension model lists can include discriminated chat, image, and classifier entries with operation implementations; when supplied, they replace the provider catalog across every operation. Chat-facing reads (getModels(),getAvailableSnapshot(), the model picker) are unchanged.Added classifier support to
ModelRuntime, includingclassify(), classifier model accessors, runtime-resolved authentication, and the built-in TypeSafejev-latestmodel.Added
types=chat,image,classifierto pi.dev model catalog requests so remote refreshes overlay every supported model type; entries of unknown model types are ignored.Added the
provider_stream_eventextension event for observing parsed provider events before normalization, with an opt-in/debug-providerexample viewer (#9784, #9901Added a show/hide toggle (
H) in HTML exports for custom messages markeddisplay: false. Messages remain hidden by default and can also be revealed from the sidebar (#8896, #10020Added inherited Claude Sonnet 5.5 support for Anthropic with adaptive thinking and a 1M context window.
Added a Built-in section in
pi configto disable the built-inmcp,llama.cpp,codemode, andtool-searchextensions globally or per project, stored as-builtin:<name>in theextensionssetting. SDK inline extensions opt in withbuiltin: true.Added
+nameand-nameentries to thedefaultToolssetting to add or remove tools without repeating the defaults, for example"defaultTools": ["+codemode"]. Project entries of this form apply on top of the user setting. Documented how to enablecodemodewithout MCP and how to use classifier models such as Jev from codemode scripts.Added the token usage and cost of codemode
models.classify()calls to the codemode tool result, so they count toward the session cost; the codemode result shows each call's cost.
Changed
Switched the build from the TypeScript native preview to TypeScript 7.0 with an ES2024 target, and replaced
tsxwith Node's built-in type stripping for running from source (#9965).Removed the
[Themes]section from the startup banner. Custom themes remain available in/settings, and theme conflicts are still reported.Changed the startup header to show the pi logo with the version instead of the app name.
Changed the built-in
darkandlightthemes to the revised pi colors, written in OKHSL.Changed light/dark terminal detection to use the reported background color first, then the terminal's light/dark report, then
COLORFGBG. The first-time setup no longer shows the detected appearance.Renamed the inherited OpenAI Codex provider to "OpenAI Codex (legacy)"; Sign in with ChatGPT on the OpenAI provider supersedes it.
Changed inherited terminal detection to treat
TERM=*-directas truecolor.Built-in extensions and tools are named
builtin:<name>(for examplebuiltin:mcpandbuiltin:read) in errors, diagnostics, RPC source info, and bug reports, instead of<inline:name>and<builtin:name>. Their slash commands no longer carry a[t]autocomplete tag.--no-extensionsalso disables the built-in extensions, including the llama.cpp provider. Load one explicitly with-e builtin:<name>, for examplepi -ne -e builtin:mcp.Tool calls without a custom call renderer, including direct MCP tool calls, now show their arguments: as
key=valuepairs on the title line when collapsed and onekey: valueline per argument when expanded. MCP calls are titledserver/tooland their results collapse to 5 lines.bashandpowershellstructured results, which codemode scripts receive, now hold up to 1 MiB of output instead of the model-facing 2000 lines or 50KB, and addtruncatedandfull_output_path. Longer output keeps its first and last 512 KiB. Empty output is""instead of(no output).
Fixed
Fixed X11 clipboard text being misidentified as an image when the clipboard owner accepts unadvertised image targets (#9786).
Prevented managed git packages from automatically installing Pi peer dependencies and added warnings for extension packages that list host-provided modules in
dependencies(#9863).Fixed pinned git extensions loaded with
-econtinuing to use the first downloaded commit after the ref changes (#9982).Fixed
RpcClientskipping the next event listener when a listener unsubscribes while handling an event, which could makewaitForIdle()time out aftercollectEvents()(#9990).Fixed full-file
readcalls rendering as:1when models sendnullfor omittedoffsetandlimit(#9996).Fixed new sessions being lost when pi exits before the first assistant response. The session file is now created when the first user message is sent (#10000).
Fixed unloaded llama.cpp autoload presets overwriting a cached runtime context window with the GGUF training context (#10077, #10158
Fixed custom themes ignoring
terminal.trueColorand other terminal capability overrides and rendering with 256 colors (#9973, #10039Fixed pasting files copied in Finder inserting the file icon image instead of the file paths; paths are quoted in bash mode (#9999, #10136
Fixed the startup header, loaded resources, and chat notices keeping their old colors after a theme change.
Fixed the Fireworks default model pointing at the removed Kimi K2.6 model; it now defaults to Kimi K3.
Fixed the OpenCode Go default model pointing at the removed Kimi K2.6 model; it now defaults to Kimi K3.
Fixed the Together default model pointing at the removed Kimi K2.6 model; it now defaults to Kimi K3.
Reduced CPU use while streaming in long sessions and when previewing themes: the footer caches session usage totals, collapsed bash results cache their preview, and
sanitizeBinaryOutput()no longer splits output into per-character arrays.Fixed the usage of tools called through
ctx.executeTool(), for example from codemode scripts, being dropped from the session cost; it is now added to the calling tool's result usage.Fixed inherited
/skillautocomplete appearing empty when loaded skill names did not contain the letters inskill(#9944).Fixed inherited path and
@autocomplete not working after opening wrappers such as(,[,{,<, or a backtick.Fixed inherited image stretching in terminals that use the Kitty graphics protocol (#8938, #9957
Fixed inherited shell cursor staying hidden after exit when an extension closed an overlay during shutdown (#10026).
Fixed inherited keyboard input being lost after a mouse click in a
/settingssubmenu closed it.Fixed inherited 1-hour Anthropic cache writes through Vercel AI Gateway being priced at the 5-minute rate (#9210).
Fixed inherited model-level
samplingParamsbeing dropped by directstream()/complete()calls on OpenAI-compatible APIs (#9506).Fixed inherited Mistral GLM requests failing with "Expected at most one leading ThinkChunk" after empty content deltas (#9674).
Fixed inherited OpenAI Fast mode requests being priced at the standard rate (#10034).
Fixed inherited Mistral reasoning models ignoring the requested thinking level (#9678).
Fixed inherited OpenCode Zen and OpenCode Go
qwen3.8-flashthinking being replayed as plain text on later turns (#10047).Fixed inherited OpenAI Responses streams from servers that omit
output_index, such as llama.cpp, running mixed-up tool calls; such streams now end with an error (#9974).Fixed inherited Anthropic and OpenAI Codex browser sign-in waiting indefinitely after the provider redirected with an authorization error, and Anthropic sign-in failing when its callback port is in use.
Fixed inherited GitHub Copilot Claude Opus 5.5 offering unsupported thinking levels when upstream model metadata is incomplete.
- 0.87.x
0.87.1
22 Sep 2026Latest frontier models — Use Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna through supported providers, including GitHub Copilot. See Choose a Model.
Grok 4.7 by default for xAI — New xAI sessions now default to Grok 4.7. See Provider Authentication.
Added inherited Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna support for GitHub Copilot.
New Features
Latest frontier models — Use Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna through supported providers, including GitHub Copilot. See Choose a Model.
Grok 4.7 by default for xAI — New xAI sessions now default to Grok 4.7. See Provider Authentication.
Added
Added inherited Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna support for GitHub Copilot.
Added inherited GPT-6 Sol and GPT-6 Luna support for OpenAI API keys and OpenAI Codex subscriptions.
Added inherited Claude Opus 5.5 support for Anthropic with adaptive thinking and a 1M context window.
Changed
Changed the default xAI model to Grok 4.7.
Fixed
Fixed split-turn compaction summaries being refused by Claude Fable 5.1 by clearly separating the conversation and using continuation-oriented instructions
Fixed missing or invalid
--modevalues being silently ignored instead of reporting an error and exiting with a nonzero status (#9045).Fixed inherited image-only user messages being rejected by some OpenAI-compatible providers because they included an empty text part (#9797).
Fixed inherited Anthropic OAuth requests reporting an outdated Claude Code version.
0.87.0
21 Sep 2026- Breaking
Removed the inherited
shouldStopAfterTurnagent option. UsefinishTurnand return{ action: "end" }instead.finishTurnruns beforeturn_endbut applies the decision afterward, and it also receives error and aborted responses; migrate normal-response predicates by returningundefinedfor those hard exits. See the@earendil-works/pi-agent-corechangelog for a complete before-and-after example. - Breaking
Added
ContextEditEntryto the exportedSessionEntryunion. TypeScript consumers with exhaustive entry switches must handlecontext_edit; usereplacement: nullfor omission and a content replacement otherwise. - Breaking
Made
SessionManagercanonical forAgentSessionprovider context. Assigningsession.agent.state.messagesno longer replaces future request history; restore withSessionManager.inMemory(cwd, { id }, entries), navigate withsession.navigateTree(), or append throughsession.sessionManagerand callsession.refreshContext().
New Features
Canonical session context and extension boundaries — Edit model context without rewriting history and add actionable lifecycle hooks. See ContextEditEntry and extension events.
Full-transcript context extensions — Use
context_with_systemfor per-request system-message transformations. Seecontext_with_system.Per-model image input limits — Configure cache-safe image resizing per model for attachments,
read, and tool-result images. See Image Input Limits.
Breaking Changes
- Breaking
Removed the inherited
shouldStopAfterTurnagent option. UsefinishTurnand return{ action: "end" }instead.finishTurnruns beforeturn_endbut applies the decision afterward, and it also receives error and aborted responses; migrate normal-response predicates by returningundefinedfor those hard exits. See the@earendil-works/pi-agent-corechangelog for a complete before-and-after example. - Breaking
Added
ContextEditEntryto the exportedSessionEntryunion. TypeScript consumers with exhaustive entry switches must handlecontext_edit; usereplacement: nullfor omission and a content replacement otherwise. - Breaking
Made
SessionManagercanonical forAgentSessionprovider context. Assigningsession.agent.state.messagesno longer replaces future request history; restore withSessionManager.inMemory(cwd, { id }, entries), navigate withsession.navigateTree(), or append throughsession.sessionManagerand callsession.refreshContext(). - Breaking
Expanded
TurnEndEventwith required boundary fields and addedAgentBeforeSettleEventto the exportedExtensionEventunion. Consumers constructing events or exhaustively switching onExtensionEventmust handle the new shapes.ExtensionRunner.emit()no longer acceptsturn_end; host integrations dispatch actionable boundaries withemitBoundary(baseEvent, buildContext). - Breaking
Deferred runs requested from
agent_settledhandlers until all settled handlers finish. Handlers still observectx.isIdle() === true, but no longer see a reentrantagent_startduring the same notification dispatch.
Added
Added append-only model-context edits. For example,
sessionManager.appendContextEdit(entryId, null)omits one message from future provider context without changing raw history, usage, or UI history.Added actionable
turn_endandagent_before_settleextension boundaries. Return{ entries: [...event.entries, draft], continue: true }to persist structural entries in order and ensure one next provider request without changing steering or follow-up scheduling.Added retain-none compaction input:
sessionManager.appendCompaction(summary, null, tokensBefore)stores the compaction's own ID as its kept boundary.Added the
context_with_systemextension event, which runs aftercontexthandlers on the full transcript including system messages and sends its result verbatim. Seecontext_with_system.Added per-model image resize profiles through
inputLimits.images.resizeinmodels.json, applied to file attachments, image reads, and tool-result images (#9631).
Fixed
Fixed string context-edit replacements producing invalid assistant and tool-result message content instead of text blocks.
Fixed context-invisible boundary metadata and replacement edits causing newly appended or replaced input to be summarized before its first provider request.
Fixed edited-context accounting both discarding valid assistant usage captured after the latest context edit and reusing that usage after a later compaction made it stale.
Fixed selected error retries and final length/overflow recovery retaining abandoned model attempts in future provider context; post-run recovery omissions are now persisted without hiding raw transcript history or changing queue scheduling.
Fixed
contexthandlers that filter or slice messages dropping the prompt and tool declarations, which after extension-driven compaction left requests without built-in tools or made Codex emit raw tool-call text. Handlers no longer see system messages; Pi restores the prompt and tool state after they run. Seecontext(#9789, #9822).Fixed
/bugallowing uploads in offline mode while preserving local zip exportsFixed idle prompt-cache warming rebuilding expired caches when its timer or an extension decision is delayed.
Improved crash diagnostics with hints identifying loaded extensions that appear in the stack trace.
Fixed text files beginning with
GIFbeing misclassified as images and omitted fromreadand CLI@fileinput (#9755).Fixed malformed prompt template frontmatter being silently ignored instead of reported as a resource warning
Fixed inherited unknown OpenAI-compatible Chat Completions endpoints receiving strict tool schemas unless they explicitly advertise support (#9816).
- 0.86.x
0.86.1
20 Sep 2026Meta Muse provider — Sign in with Meta using
/login metaor useMETA_API_KEYto access Muse Spark models. See Meta (Muse subscription).Added Meta (Muse subscription) login via
/login metawith automatic Model API key refresh, plusMETA_API_KEYsupportEnabled Node's persistent compile cache before loading the bundled CLI runtime, reducing repeat launch time.
New Features
Meta Muse provider — Sign in with Meta using
/login metaor useMETA_API_KEYto access Muse Spark models. See Meta (Muse subscription).
Added
Added Meta (Muse subscription) login via
/login metawith automatic Model API key refresh, plusMETA_API_KEYsupport
Changed
Enabled Node's persistent compile cache before loading the bundled CLI runtime, reducing repeat launch time.
Fixed
Fixed
/bugdescriptions dropping line breaks from pasted diagnostics.Fixed
/bughints appearing for user cancellations and retryable provider failures such as service unavailability.Fixed clipboard copy failing in containers and WSL without WSLg by restoring the OSC 52 fallback when no display is available, and added a verified Windows clipboard backend for WSL (#9688).
Fixed inherited z.ai
Prompt too longerrors not being recognized as context overflow (#9805).Fixed inherited Cerebras models advertising unsupported strict tool schemas, which caused HTTP 400 errors when strict and non-strict tools were mixed
0.86.0
19 Sep 2026- Breaking
Changed inherited pi-ai provider stream inputs from
Contextto normalizedTranscriptContextvalues. Custom providers must read system prompts and tool declarations fromcontext.messageswithgetCurrentSystemPrompt()andgetCurrentTools(). See Custom Streaming API. - Breaking
Restricted inherited
ToolCall.argumentsandToolResultMessage.detailsto JSON-compatible values, changedToolResultMessageinto a conditional type, and madeJsonValuearrays readonly. - Breaking
user_bashnow fails closed: errors or invalid defined results abort the command without invoking later handlers or executing locally. Returnundefinedto continue propagation; otherwise return{ operations }or{ result }(#9068).
New Features
Prompt cache warming — Keep valuable prompt caches alive during long tool runs and optionally while idle using cost-aware refreshes. See Cache Warming.
Bug reporting — Report problems with
/bugusing redacted diagnostics, optional transcripts, or exported ZIP archives. See Reporting Bugs.Transcript-aware prompt and tool updates — Preserve instruction and tool changes across resume and branch navigation while retaining cached prefixes. See
before_agent_start.Offline Radius model catalog — Select Radius models immediately, with cached and live catalogs overlaid when available. See Radius.
Per-model compaction budgets — Configure reserved and recent-token budgets by model. See Per-model overrides.
Breaking Changes
- Breaking
Changed inherited pi-ai provider stream inputs from
Contextto normalizedTranscriptContextvalues. Custom providers must read system prompts and tool declarations fromcontext.messageswithgetCurrentSystemPrompt()andgetCurrentTools(). See Custom Streaming API. - Breaking
Restricted inherited
ToolCall.argumentsandToolResultMessage.detailsto JSON-compatible values, changedToolResultMessageinto a conditional type, and madeJsonValuearrays readonly. - Breaking
user_bashnow fails closed: errors or invalid defined results abort the command without invoking later handlers or executing locally. Returnundefinedto continue propagation; otherwise return{ operations }or{ result }(#9068).
Added
Added transcript-backed mid-conversation system prompt and tool changes so instruction and tool updates survive resume and branch navigation while preserving cached prefixes on supported models. See
before_agent_startand Entry Types (#9548).Added inherited native deferred tool loading for Fireworks Messages models. Use
ToolSearchortool_searchas the loader name for prompt-prefix deferral (#9323).Added click toggling for branch summaries, compaction summaries, and skill invocation entries.
Added the public Radius model catalog for immediate and offline model selection, with cached and live gateway catalogs overlaid when available.
Added
ctx.modelRegistry.stream()andstreamSimple()for extension model calls through configured providers with resolved authentication (#8964).Added per-model
reserveTokensandkeepRecentTokenssettings throughcompaction.modelOverrides, with ordinary compaction settings as fallback (#8133).Added
compat.allowedFallbackModelsconfiguration for overriding or disabling Anthropic server-side fallback models (#9294).Added an unsubscribe function from
pi.on()so extensions can drop event handlers. Handlers added or removed during a dispatch apply to later dispatches, not the current one (#8967).Exported extension hook event and result types that were previously omitted from the package entry points (#9642).
Added
/bug [description]to report a bug to the Pi developers. The report bundles environment, model, provider, extension, and settings metadata (secrets redacted), assistant message diagnostics from the session, optionally the session transcript, or a model-written summary of what went wrong instead. It is uploaded to Radius (no login required; attributed when logged in) or exported as a zip archive, and the report id is recorded in the session as api.bug-reportentry. Crashes are recorded in~/.pi/agent/crashes.json, announced once on the next start, and attached to the next report; unexplained errors and exhausted retries point at/bugonce per session.Added cost-aware prompt-cache warming during long tool runs and optionally while idle, with configurable modes, model cache-lifetime metadata,
/sessiondiagnostics, transcript notices, and thecache_warming_decisionextension event. See Cache Warming (#9668).
Changed
Made
--resumesession results appear progressively, using file modification times to prioritize all-folder loading and cancelling outstanding transcript reads after selection.Reduced
--continuestartup time by checking candidate session headers in modification-time order and stopping after the newest matching session.Replaced the external native clipboard dependency with bundled asynchronous macOS, Windows, and X11 helpers while preserving platform command and OSC 52 fallbacks (#9163).
Reduced inherited fuzzy search latency for long texts by using native substring search instead of scanning each character in JavaScript (#9267).
Moved compaction, branch summarization, and retry spinners into the editor border alongside the working indicator. Custom editors use the same embedding opt-in for all status spinners.
Enabled strict-prefer JSON-schema sampling by default for built-in
read,bash,powershell,edit, andwritetools, without requiringPI_EXPERIMENTAL. Extensions can re-register tool definitions withconstrainedSampling: false.Formatted Bash and PowerShell tool durations of at least one minute as minutes and seconds, with hours when needed (#9628).
Deferred the extension compiler and bundled virtual modules until a filesystem extension is loaded, reducing the baseline SDK import cost (#9540).
Fixed
Fixed GitHub Copilot GPT models, including GPT-6 Astra, using the Chat Completions adapter instead of the required Responses adapter
Fixed inherited DeepSeek V4.1 thinking levels on OpenRouter and OpenCode Go preserving provider effort metadata (#9485).
Fixed inherited bodyless HTTP 400/413 errors from non-Cerebras providers being misclassified as context overflow (#9482).
Fixed inherited Vercel AI Gateway replaying unsigned thinking as assistant text (#9676).
Fixed inherited Google Generative AI and Vertex AI using unsupported thinking levels when reasoning is omitted or when model capabilities differ within a Gemini family (#9455).
Fixed inherited Anthropic-compatible relays breaking signed thinking replay when they report a different response model, while preserving fallback pricing (#9188).
Fixed inherited Amazon Bedrock one-hour cache writes being priced at the five-minute rate (#9457).
Fixed inherited quadratic CPU usage when draining buffered
EventStreamevents (#9055).Fixed inherited Mistral Medium reasoning requests to use
reasoning_effortfor all reasoning-capablemistral-medium-*model IDs instead of the unsupportedprompt_mode(#8700).Fixed inherited OpenCode and OpenCode Go requests to send
x-opencode-sessionfromsessionIdacross all supported API adapters (#9326).Fixed inherited OpenAI Codex requests to send the model's Off reasoning effort instead of omitting it, while respecting unsupported Off mappings (#9191).
Fixed inherited Fireworks unsigned thinking replay and reasoning effort selection using catalog metadata, with verified DeepSeek V4 and Qwen3.8 fallbacks and removal of redundant GLM 5.2 and Kimi K3 effort aliases (#9323).
Fixed inherited OpenRouter requests to send
x-session-idfromsessionIdfor Chat Completions and Anthropic Messages models when prompt caching is enabled (#9102).Fixed the inherited DeepSeek catalog to advertise
deepseek-flashfor DeepSeek V4.1 Flash instead of retired Flash aliases, and refreshed DeepSeek pricing metadata (#9423).Fixed inherited Mistral-hosted GLM-5.2 reasoning requests to use
reasoning_effortinstead of the ignoredprompt_mode(#9375).Fixed inherited OpenAI-compatible Responses errors to identify the actual provider instead of always labeling them as OpenAI errors (#9298).
Fixed inherited Baseten requests to send session-affinity headers from
sessionIdfor automatic prompt-cache routing (#9629).Fixed inherited retry classification for Cloudflare 520 responses (#9627).
Fixed inherited retry classification for transient Azure peak-load capacity errors (#9669).
Fixed session tree navigation racing with active compaction and replacing its progress UI
Fixed exact session ID lookup scanning complete transcript bodies instead of reading session headers
Fixed repeated Anthropic thinking-drop notices being shown for the same dropped blocks, and shortened notices while retaining details in the session (#9391).
Fixed mid-run threshold compaction silently skipping oversized trailing tool results (#9740).
Fixed signal-terminated local shell commands being reported as successful with partial output
Fixed local clipboard failures reporting success when the terminal ignored the fallback OSC 52 write, and added platform-specific setup guidance when no clipboard backend works (#9618).
Capped agent-level retry backoff at
retry.maxAgentDelayMs(60s by default) so long retry runs stay responsive during prolonged transient outages (#8826).Fixed direct RPC
steerandfollow_upcommands bypassing extensioninputhandlers (#8718).Fixed premature missing-model errors after login by waiting for catalog discovery. Radius now defaults to
balanced, falling back to the first available Radius model when needed.Fixed fullscreen mode reserving a blank row for custom footers that render zero rows (#8919).
Fixed extension tools without parameter schemas to be rejected during registration instead of breaking provider requests (#9300).
Fixed
before_agent_starthandlers returningsystemPrompt(andforceSystemPrompt) on models with mid-conversation system messages: the forced prompt is now sent as the provider's leading system prompt instead of being appended as a section patch after the original prompt.Fixed loaded llama.cpp models with
enable_thinkingchat templates ignoring Pi's thinking level (#9528).Fixed cancellation races that could start automatic compaction, leave stale retry state, or miss cancellation while waiting for summarization authentication (#9340, #9777).
Fixed asynchronous Kitty image conversion replacing newer partial tool output images
Fixed inherited skill slash-command autocomplete ranking the
skill:prefix instead of the bare skill nameFixed inherited file autocomplete boundaries and path quoting around CJK punctuation
Fixed inherited LaTeX legacy font switches falling back to raw source, centered
caseslayouts around surrounding equations, and vertically laid out unsupported and nested display scripts (#8827, #9564, #7929).Fixed inherited fullscreen Kitty images being erased by later row clears in WezTerm (#9169).
Removed
- Breaking
Removed unavailable inherited GPT-5.4 and GPT-5.4 mini models from OpenAI Codex selection (#9394).
- 0.85.x
0.85.1
5 Sep 2026GPT-6 Astra — Available through OpenAI API keys and OpenAI Codex subscriptions. See API Keys and OpenAI Codex.
Added GPT-6 Astra for OpenAI API keys and OpenAI Codex subscriptions.
Added five-times-faster mouse wheel scrolling while holding Alt in fullscreen mode
New Features
GPT-6 Astra — Available through OpenAI API keys and OpenAI Codex subscriptions. See API Keys and OpenAI Codex.
Added
Added GPT-6 Astra for OpenAI API keys and OpenAI Codex subscriptions.
Added five-times-faster mouse wheel scrolling while holding Alt in fullscreen mode
Fixed
Fixed configurable save keybindings in the model and thinking selectors
Fixed SDK import failures caused by unintentionally publishing internal experimental code and dependencies in 0.85.0. The experimental
clientandexperimental/pluginsubpaths and server/client commands are now source-only throughpi-test.sh; the supported local SDK and stdio RPC API are unchanged (#9132).Fixed mouse hover changing selection and recentering autocomplete and settings lists, causing clicks to target a different item.
Fixed long prompt-cache requests for GPT-5.6+ Responses models to use
prompt_cache_options.ttl: "30m"instead ofprompt_cache_retention: "24h".
0.85.0
4 Sep 2026Persistent Claude thinking effort — Supported Anthropic transports preserve per-turn effort and recover safely from signed-thinking mismatches. See Model Configuration.
Fullscreen transcript controls — Jump to the latest message from a scrolled transcript and use the embedded working indicator. See TUI Fullscreen Viewport.
Restorable in-memory sessions — Resume externally stored session entries through the SDK. See Session Management.
New Features
Persistent Claude thinking effort — Supported Anthropic transports preserve per-turn effort and recover safely from signed-thinking mismatches. See Model Configuration.
Fullscreen transcript controls — Jump to the latest message from a scrolled transcript and use the embedded working indicator. See TUI Fullscreen Viewport.
Restorable in-memory sessions — Resume externally stored session entries through the SDK. See Session Management.
Added
Added
SessionManager.inMemory()support for restoring externally managed session entriesAdded inherited OpenAI-compatible
vllmPriorityandsupportsMaxOutputTokensmodel settings for vLLM scheduler priority and OpenAI Responses output-token limitsAdded inherited LaTeX rendering for relational algebra join symbols
Added a clickable "Jump to latest message" label with the
tui.altScreen.bottomshortcut to the fullscreen transcript while it is scrolled up
Changed
Moved the streaming working indicator into the default editor border and matched its default spinner and label to the thinking-level border color. Custom editors retain the standalone indicator unless they opt in to embedding it
Reduced inherited fullscreen transcript search latency on large transcripts by caching unchanged search results, indexing ASCII runs, and limiting highlight work to visible matches
Fixed
Fixed managed
fdand ripgrep downloads on Linux musl systemsRemoved the unavailable inherited Grok Build 0.1 model from
/modelFixed inherited provider streams emitting incompatible event sequences and custom tool-call deltas.
Restored the
@earendil-works/pi-coding-agent/clientcompatibility entry point.Fixed the inherited Qwen Token Plan Individual catalog to include Qwen3.8 Flash (#9021).
Fixed inherited OpenAI Codex SSE parsing to process terminal events that are not followed by a blank line (#9047).
Fixed inherited GitHub Copilot Claude Fable 5 requests so selected reasoning levels are sent (#8961).
Fixed inherited Baseten GLM-5.2 models incorrectly advertising image input support
Fixed skills being unavailable when Bash is the only enabled tool
Fixed concurrent session shares overwriting one another
Fixed image orientation detection skipping EXIF data after non-EXIF APP1 segments
Fixed imported sessions overwriting an existing session with the same filename
Fixed session forks losing their compaction boundary
Fixed in-memory session forks before an active turn settled
Fixed inherited Fireworks GLM models using the wrong API adapter.
Fixed inherited
NO_PROXYmatching for root domains and subdomainsFixed
bash,edit,find,grep,ls,read, andwritetools ignoringctx.cwdFixed inherited terminal startup under restricted seccomp policies that reject the
SIGWINCHself-signalFixed inherited Zed terminal image capability detection
Fixed drag selection continuing over the fullscreen editor.
Fixed managed
fdand ripgrep downloads requiring the GitHub Releases APIFixed branch summaries failing when reasoning consumes the previous 2048-token output cap (#8845).
Fixed the write tool reporting UTF-16 code-unit counts as byte counts by removing the misleading count (#8979).
Fixed proxied plain-HTTP provider requests hanging after a tool call by tunneling them with CONNECT (#8134).
Fixed RPC
abortreporting success without cancelling an in-progress manual compaction (#8920).
- 0.84.x
0.84.4
28 Aug 2026Terminal capability overrides — Override detected terminal hyperlink, image, and truecolor support. See Capability Overrides.
Extension UI prompt events — Integrations can distinguish active agent work from time spent waiting for
ctx.uiprompts. See Extension UI prompt events.RPC queue clearing — Retrieve and clear queued steering and follow-up messages with
clear_queue. See RPCclear_queue.
New Features
Terminal capability overrides — Override detected terminal hyperlink, image, and truecolor support. See Capability Overrides.
Extension UI prompt events — Integrations can distinguish active agent work from time spent waiting for
ctx.uiprompts. See Extension UI prompt events.RPC queue clearing — Retrieve and clear queued steering and follow-up messages with
clear_queue. See RPCclear_queue.Fullscreen selection copy controls — Disable automatic selection copying in fullscreen mode and use Ctrl+X to copy the active selection. See UI & Display.
DeepSeek V4 Flash Vision (experimental) — Use the vision-capable model through the built-in DeepSeek provider. See API Keys.
Added
Added
ui_prompt_startandui_prompt_endextension events so host integrations can distinguish active agent work from waiting on user-facingctx.uipromptsAdded
detectSupportedImageMimeTypeFromFile()to the public library exportsAdded inherited experimental vision-capable
deepseek-v4-flash-vision-expmodel support.Added transcript usage notices for compaction and branch summaries when cache miss notices are enabled.
Added RPC
clear_queueto retrieve and remove queued steering and follow-up messages (#8432).Added environment variables and advanced settings for overriding auto-detected terminal hyperlink, image, and truecolor capabilities (#8665).
Added
fullscreenCopyOnSelectto disable automatic fullscreen selection copy; when disabled,Ctrl+Xcopies the active text selection before falling back to the last assistant message, while/treestill copies the selected message (#7720).
Fixed
Fixed toggling thinking visibility clearing partial output from running Bash tools (#8611).
Fixed Windows shell aborts crashing Pi when
taskkill.exeis unavailable onPATH(#6596).Fixed resumed sessions corrupting the next appended entry when their JSONL file lacks a trailing newline (#8345).
Fixed extension messages sent with
triggerTurn: falsewhile the agent is running being inserted between a tool call and its result, which made providers that validate message order reject the replayed history. They are now appended once the turn's tool results are in (#8537).Fixed compaction and branch summaries forcing
toolChoice: "none"(#8649, #8638).Fixed large tool results crossing the auto-compaction threshold being sent to the provider before compaction. Pi now compacts between tool execution and the next assistant response in the same run, and restores interactive progress when that run resumes (#6879).
Fixed Google Vertex requests failing with
HttpsProxyAgent is not a constructorwhen the bundled Node.js runtime uses an HTTP(S) proxy (#8610).Fixed saving a default model from a non-empty model scope so it remains available in that scope.
Fixed inherited
@file autocomplete ranking to prefer direct and shallower matches over similarly ranked nested paths (#8669).Fixed inherited OpenAI-compatible streams serializing thinking signatures repeatedly during streaming (#8671).
Fixed inherited main-screen rendering crashing when image-heavy output exceeded V8's string length limit (#8028).
Fixed inherited fullscreen double-click word selection splitting paths and kebab-case tokens on
/and-(#8676).Fixed inherited Cloudflare AI Gateway catalogs omitting supported
workers-ai/*passthrough models.Fixed inherited OpenAI-compatible reasoning replay to merge consecutive streamed text and summary
reasoning_detailsdeltas.Fixed inherited OpenRouter reasoning controls so reasoning-mandatory models do not receive
effort: "none"Fixed inherited OpenAI-compatible Chat Completions ignoring an explicitly requested
toolChoicewhen no tools are defined.Fixed inherited fragmented Mistral tool calls splitting when continuation chunks omit the tool-call ID (#8387).
0.84.3
24 Aug 2026- Breaking
Renamed the inherited
GoogleThinkingLeveltype toGoogleApiThinkingLeveland addedResolvedGoogleThinkingLevelfor normalized adapter levels. PowerShell tool — Use optional native PowerShell command execution on Windows. See PowerShell Tool.
Safer managed updates — Stage, verify, and atomically activate updates for installer-managed installations. See Install and Manage.
New Features
PowerShell tool — Use optional native PowerShell command execution on Windows. See PowerShell Tool.
Safer managed updates — Stage, verify, and atomically activate updates for installer-managed installations. See Install and Manage.
Model and thinking controls — Select thinking levels with
/thinking, search defaults, keep selections session-scoped, and persist them explicitly with Ctrl+S. See Models and Thinking.
Breaking Changes
- Breaking
Renamed the inherited
GoogleThinkingLeveltype toGoogleApiThinkingLeveland addedResolvedGoogleThinkingLevelfor normalized adapter levels.
Added
Added an optional
powershelltool for Windows, configurable throughdefaultToolsand the SDK. See PowerShell Tool.Added a
/thinkingselector and searchable default choices to the model and thinking selectors; Ctrl+S saves the selected model as the global default. See Models and Thinking.Added optional routing session IDs to exported compaction summary helpers so callers can preserve provider routing without enabling prompt cache writes.
Added transcript usage notices for compaction and branch summaries when cache miss notices are enabled.
Added
session_compact_failedextension events so compaction failures and aborts expose their reason, retry state, source, and error message to handlers (#8175).Added inherited provider-neutral
toolChoicesupport to simple stream requests.Added inherited automatic Anthropic server-side refusal fallback for supported first-party models, including returned-model usage pricing (#8017).
Added inherited configurable OpenAI-compatible thinking-token budget fields for vLLM, Qwen/SGLang, and llama.cpp servers. See OpenAI Compatibility
Added inherited China-specific ZAI Coding Plan models, including GLM-4.6V vision support and API-equivalent usage cost estimates (#8220).
Added inherited
deepseek-v4-pro-0813support to the Qwen Token Plan Individual catalog (#8194).
Changed
Changed experimental installer-managed installations so
pi updatestages, verifies, and atomically activates the selected release in place. See Install and Manage.Changed inherited built-in xAI models to use the Responses API with encrypted reasoning replay and made Grok 4.6 the default xAI model
Changed inherited Anthropic, Azure OpenAI, Google, Mistral, and OpenAI adapters to send Pi's default
User-Agentunless overridden (#8305).Changed Windows and WSL keybinding defaults to avoid terminal-reserved shortcuts for image paste, model cycling, editor undo, fullscreen transcript navigation and search, and message queueing (#8372).
Changed Bun release archives to ship the native clipboard binary only inside the wrapper package, removing a duplicate platform package from each archive.
Changed package resource glob expansion to use Node.js's built-in implementation with deterministic visible-path matching, reducing the installed runtime dependency tree.
Changed the bundled Node.js runtime to load jiti only when importing an extension and Babel only when uncached source needs transformation, reducing CLI startup time and bundle size.
Changed syntax highlighting to initialize only twenty common languages eagerly and defer the remaining grammars until after the initial TUI render, reducing CLI startup time.
Changed the Node.js CLI and RPC entrypoints to load a bundled runtime, reducing startup filesystem reads while keeping the public library and legacy module paths on the modular runtime for normal dependency identity.
Changed session sharing to render clickable terminal links, display only the canonical Radius artifact URL, and include the current system prompt and active tool definitions in Radius session shares.
Fixed
Fixed failed extension factories leaving event subscriptions, provider registrations, and default flag state active
Fixed
models.jsontypings omitting the documented OpenAI-compatiblecompat.supportsFinishReasonprovider and model overrideFixed
/modeland/thinkingselections being persisted globally unless explicitly saved with Ctrl+S (#5263).Fixed JSON and RPC
toolcall_startevents omitting the tool call id and nameFixed extensions failing to load when the Node.js CLI runs as a single-executable application (#8237).
Fixed nested Markdown skills inside
.agents/skills/grouping directories not being discovered.Fixed compaction and branch summarization requests exposing tools to providers.
Fixed single-object
edittool inputs failing validation by accepting them as one-edit arrays in both coding-agent and harness edit tools (#7835).Fixed root Markdown files such as
README.mdandAGENTS.mdin skill directories being reported as broken skills unless they declare valid skill frontmatter (#7805).Fixed the default Cerebras model referencing an unavailable Z.AI model.
Fixed inherited OpenAI-compatible Chat Completions reasoning replay to preserve and resend assistant-level
reasoning_detailsverbatim and in order (#7994).Fixed inherited Anthropic server-side fallback responses being priced with the requested model instead of the returned fallback model (#8285).
Fixed inherited GitHub Copilot login triggering model-policy rate limits by limiting policy updates, retrying model discovery once, and honoring server retry delays (#7850).
Fixed inherited Amazon Bedrock dropping and failing to replay opaque redacted reasoning from non-Anthropic models
Fixed inherited Z.AI Coding Plan models deriving incomplete reasoning-effort metadata, including missing GLM-5.3 low, high, and max levels (#8336).
Fixed inherited DeepSeek V4 Flash on OpenCode and OpenCode Go omitting its supported low thinking level
Fixed inherited Azure OpenAI Responses ignoring
toolChoicein provider-specific stream requests.Fixed inherited Amazon Bedrock response hooks receiving only a synthesized request id instead of the raw response headers (#8234).
Fixed inherited Kimi usage reporting so top-level
cached_tokenscount as cache reads instead of normal input tokens (#8075).Fixed inherited Google custom models ignoring
thinkingLevelMap, which dropped extended thinking controls (#8135).Fixed writes to
auth.jsonandmodels-store.jsonoverriding administrator-managed file permissions and ACLs (#7779).Fixed UTF-8 BOM markers preventing frontmatter and user configuration files from loading (#8337).
Fixed invalid settings files being easy to miss during interactive startup by rendering warnings with the file path inside the TUI (#7829).
Fixed the subagent example repeatedly prompting before running project-local agents in trusted repositories (#8261).
Fixed npm package update checks treating older registry versions as available updates, preventing
pi updatefrom downgrading already-newer installed packages (#8226).Fixed built-in llama.cpp models disappearing from
/modelwhen/llamarefreshed a configured server underPI_OFFLINE, and included idle-sleptsleepingrouter models in the selectable catalog (#8167).Fixed
pi.registerFlag()accepting default values that do not match the declared flag type (#8064).Fixed Z.AI Coding Plan defaults referencing the removed GLM-5.1 model (#8096).
Fixed repeated ambiguous truncated-response recovery being mislabeled as context overflow (#8130).
Fixed duplicate fullscreen right-click paste in VS Code-based terminals on Windows (#8186).
Fixed inherited padded text exceeding narrow terminal widths (#8252).
Fixed inherited wrapped Markdown table links leaking color into borders and neighboring cells, including tables inside blockquotes (#8335).
Fixed llama.cpp login guidance to direct users to
/llamabefore/modelwhen no local models are loaded (#8203).Fixed hung pi.dev model catalog requests consuming the entire refresh deadline without retrying (#8198).
Fixed inherited Xiaomi model catalogs listing shut-down MiMo V2 models in
/modeland--list-models(#8187).Fixed branch summary entries recording the navigation destination in
fromIdinstead of the pre-navigation source leaf.Fixed threshold auto-compaction being skipped when providers omit streaming usage data (#8328).
Fixed dash-prefixed prompts being parsed as options by supporting
--as an end-of-options delimiter (#7269).
0.84.2
14 Aug 2026- Security
Updated the transitive
nanoiddevelopment dependency to address a denial-of-service vulnerability. Fullscreen transcript search — Search and navigate matches in fullscreen mode. See TUI Fullscreen Viewport.
Configurable default tools — Choose startup built-in tools globally or per project. See Tools.
New Features
Fullscreen transcript search — Search and navigate matches in fullscreen mode. See TUI Fullscreen Viewport.
Configurable default tools — Choose startup built-in tools globally or per project. See Tools.
Configurable fullscreen exit output — Print the transcript or only a resume hint on exit. See Interactive Mode.
Added
Added fullscreen transcript search with
Ctrl+Shift+F, incremental match highlighting, configurable search match theme colors, and next/previous navigation withEnter/Ctrl+GandShift+Enter/Ctrl+Shift+G.Added experimental strict JSON-schema constrained sampling for the default
read,bash,edit, andwritetools underPI_EXPERIMENTAL=1.Added a fullscreen exit output setting to choose between printing the final transcript and only a session resume hint.
Added the
defaultToolssetting for configuring the initial built-in tool selection globally or per project.Added
--use-theme <name[/name]>to choose an initial per-run interactive theme without changing saved settingsAdded
expandPromptTemplatesto extensionpi.sendUserMessage()options for explicitly dispatching commands and expanding skills and prompt templates. Seepi.sendUserMessage()Added inherited
createGatewayBindingFetch()for routing Cloudflare AI Gateway requests through a Workers AI binding without an API tokenAdded inherited
AssistantMessage.endTurnto preserve OpenAI Codex's terminalend_turnsignal for diagnostics (#7766).Added inherited unbound single-line transcript scrolling actions for fullscreen mode. See TUI Fullscreen Viewport
Changed
Changed inherited Kimi Coding requests to use pi's runtime
User-Agentheader.Replaced the inherited Mistral SDK transport with a native Chat Completions HTTP stream, eliminating its generated client and schema runtime overhead.
Documented the generic
AI_AGENT=piprocess marker and how it differs fromPI_CODING_AGENT=true(#7747).Changed inherited OpenAI Responses deferred tool loading to prefer message-anchored
additional_toolswhere supported while retaining tool-search and top-level fallbacks (#7709).Reduced inherited fullscreen rendering allocation churn by painting full-width layout rows directly instead of recompositing them on every frame.
Fixed
Fixed managed-tool downloads delaying TUI startup and hiding diagnostics in fullscreen mode by mounting the TUI first and showing download progress and warnings inside it.
Fixed opening a model selector immediately after startup cancelling and restarting the in-progress model catalog refresh.
Fixed inherited GitHub Copilot login triggering API rate limits while enabling model policies by limiting concurrent policy updates (#6187).
Fixed fullscreen transcript search snapping back to the current match during manual scrolling and fragmented mouse input leaking into the search query.
Fixed inherited required LaTeX arguments starting on a new line being parsed as empty (#7760).
- Security
Updated the transitive
nanoiddevelopment dependency to address a denial-of-service vulnerability. Fixed fallback rendering for extension tool results to collapse long output and honor tool expansion (#7979).
Fixed JSON and RPC
message_updateevents dropping cumulative usage during streaming. See JSON Event Mode and RPCmessage_updateFixed
pi.sendMessage(..., { triggerTurn: false })steering an active run instead of only recording the custom messageFixed the
defaultToolssetting dropping extension and SDK custom tools when selecting built-in defaults.Fixed the subagent example rejecting YAML array syntax for the
toolsfrontmatter fieldFixed the subagent example dropping parent session model, thinking, and tool configuration
Fixed custom system prompts concatenating the current working directory with later appended prompt content
Fixed inherited OpenAI Responses function and custom tool calls losing namespaces during streaming, proxying, and replay (#7709).
Fixed inherited upstream request buffer failures not triggering automatic assistant retries.
Fixed inherited built-in and custom DeepSeek API models sending output limits through an unsupported field.
Fixed inherited Amazon Bedrock replay rejecting tool arguments that contain empty object keys while preserving all valid nested values
Fixed inherited DeepSeek compatibility detection for base URLs whose hostname contains uppercase letters
Fixed inherited Google Generative AI and Vertex AI responses with tool calls incorrectly treating output-limit or provider-error stops as normal tool use (#8059).
Fixed inherited fullscreen mouse drag selection and OSC 8 link activation in terminals that report generic SGR mouse release button codes (#7963).
Fixed inherited focused fullscreen overlays not receiving mouse wheel or viewport scroll keys such as PageUp and PageDown (#7894).
Fixed inherited LaTeX control spaces split across line endings causing complete expressions to fall back to raw source.
Fixed split
Alt+Enterinput over SSH being misread as Escape, addedPI_TUI_ESC_TIMEOUTfor high-latency terminals, and limited that timeout to lone Escape inputFixed inherited idle fullscreen sessions repainting and clearing text selection when the terminal loses focus
Fixed fullscreen selection copy to use the host clipboard and report failure instead of claiming success when OSC 52 is unsupported
0.84.1
7 Aug 2026Qwen Token Plan Individual — Use the built-in provider for models documented for Individual subscriptions. See API Keys.
Authentication readiness checks — Use
pi auth checkto verify provider or model credentials, optionally emitting the resolved credential.Improved fullscreen interaction — Select words and paragraphs with multiple clicks and configure half-page transcript scrolling. See TUI Fullscreen Viewport.
New Features
Qwen Token Plan Individual — Use the built-in provider for models documented for Individual subscriptions. See API Keys.
Authentication readiness checks — Use
pi auth checkto verify provider or model credentials, optionally emitting the resolved credential.Improved fullscreen interaction — Select words and paragraphs with multiple clicks and configure half-page transcript scrolling. See TUI Fullscreen Viewport.
Terminating blocked tool calls — Extension
tool_callhandlers can stop all-terminating batches without another model call. See Tool Events.
Added
Added Qwen Token Plan Individual as a built-in provider with its documented subscription model catalog and the shared international
QWEN_TOKEN_PLAN_API_KEY. See API KeysAdded
pi auth checkprovider/model auth preflight with optional credential output (#7152).Added
terminatesupport to blocked extensiontool_callevents so all-terminating batches can skip the automatic follow-up model call. See Tool EventsAdded inherited double-click word and whitespace selection, granularity-aware drag selection, and triple-click paragraph selection in fullscreen mode (#7725, #7733
Added inherited unbound half-page transcript scrolling actions for fullscreen mode. See TUI Fullscreen Viewport (#7735).
Changed
Softened the bash tool's
PI_*environment guideline in an attempt to reduce unnecessary inspection commands (#7128).Reduced worst-case automatic terminal theme detection delay from 200 ms to 100 ms by probing color-scheme and background support concurrently.
Fixed
Fixed Bun standalone binaries crashing on startup when the cwd contains a
bunfig.tomlwithpreloadby compiling with--no-compile-autoload-bunfigFixed extension TUI method wrappers recursing indefinitely when delegating to the original method (#7731).
Fixed right-click not pasting clipboard text in fullscreen mode on Windows.
Fixed inherited
Agent.reset()clearing transcript and runtime state during active runs; it now rejects until the agent is idleFixed inherited LaTeX relation, multiplication, and named-operator spacing, and matrix composition with stacked fractions, operator limits, and adjacent matrices.
Reduced inherited fullscreen mouse event volume under tmux, Zellij, and GNU Screen by using button-motion tracking instead of all-motion tracking.
0.84.0
6 Aug 2026- Breaking
Renamed the inherited pi-ai
ModelsStreamTransformsinterface toModelsRequestTransformsbecause its header transformation now applies to all authenticated provider requests. - Breaking
Changed JSON and RPC
message_updateevents to emit onlyassistantMessageEventdeltas, removing the cumulativemessageandassistantMessageEvent.partialfields that caused quadratic output growth. Clients that need partial messages must assemble deltas betweenmessage_startandmessage_end; the latter remains authoritative (#7290). - Breaking
ModelRegistry.getApiKeyAndHeaders()now returnsProviderHeaderswithstring | nullvalues and preservesnullheader-deletion markers. Extensions that inspect returned headers must handlenull; extensions forwarding them to pi-ai streams should pass them through unchanged. This prevents placeholder OpenAI credentials from being sent through Cloudflare AI Gateway (#7030).
New Features
Fullscreen TUI mode — Switch between regular and fullscreen modes at runtime, with a sticky editor and footer, independently scrollable transcript, and draggable scrollbars. See UI & Display.
Mermaid and LaTeX rendering — Render Mermaid diagrams and terminal-friendly Unicode math in interactive transcripts. See Markdown settings and TUI Markdown.
Per-directory context overrides — Use
AGENTS.override.mdto replace context files for a specific directory. See Context Files.Advanced custom model sampling — Configure arbitrary OpenAI-compatible
samplingParamsand opt-in vLLMthinking_token_budgetvalues. See Sampling Parameters.Baseten provider — Use built-in Baseten authentication and model support. See API Keys.
Breaking Changes
- Breaking
Renamed the inherited pi-ai
ModelsStreamTransformsinterface toModelsRequestTransformsbecause its header transformation now applies to all authenticated provider requests. - Breaking
Changed JSON and RPC
message_updateevents to emit onlyassistantMessageEventdeltas, removing the cumulativemessageandassistantMessageEvent.partialfields that caused quadratic output growth. Clients that need partial messages must assemble deltas betweenmessage_startandmessage_end; the latter remains authoritative (#7290). - Breaking
ModelRegistry.getApiKeyAndHeaders()now returnsProviderHeaderswithstring | nullvalues and preservesnullheader-deletion markers. Extensions that inspect returned headers must handlenull; extensions forwarding them to pi-ai streams should pass them through unchanged. This prevents placeholder OpenAI credentials from being sent through Cloudflare AI Gateway (#7030). - Breaking
Changed
ModelRegistry.refresh()to acceptModelsRefreshOptionsand returnModelsRefreshResultinstead of discarding cancellation and provider errors. - Breaking
Changed
ModelRuntime.setRuntimeApiKey()to accept auth cancellation options rather than catalog refresh options. Callrefresh({ providers: [providerId], signal })separately when remote freshness is required. - Breaking
Required config-form extension OAuth
refreshToken(credentials, signal)callbacks to accept and honor a concrete abort signal. - Breaking
Replaced dynamic provider refresh context store access with the read-only
context.storedsnapshot and generation-checkedcontext.publish()transaction.Providers built with
createProvider({ fetchModels }): no catalog-publication migration is required. Before and after, return the fetched models and register the resulting provider;createProvider()owns restoration, persistence, and in-memory publication.// Before const beforeProvider = createProvider({ // ... fetchModels: async ({ signal }) => { const response = await fetch(catalogUrl, { signal }); return parseModels(await response.json()); }, }); pi.registerProvider(beforeProvider); // After: unchanged const afterProvider = createProvider({ // ... fetchModels: async ({ signal }) => { const response = await fetch(catalogUrl, { signal }); return parseModels(await response.json()); }, }); pi.registerProvider(afterProvider);Handwritten native
Provider.refreshModels(): replace direct store access and pre-publication mutation with generation-guarded publications.// Before refreshModels: async (context) => { const stored = await context.store.read(); if (stored) currentModels = stored.models; if (!context.allowNetwork) return; const refreshed = await fetchModels(context.signal); currentModels = refreshed; await context.store.write({ models: refreshed, checkedAt: Date.now() }); }, // After refreshModels: async (context) => { if (context.stored) { const restored = context.stored.models; if (!(await context.publish({ update: () => { currentModels = restored; }, }))) return; } if (!context.allowNetwork) return; const refreshed = await fetchModels(context.signal); if (context.signal.aborted) return; await context.publish({ persist: { models: refreshed, checkedAt: Date.now() }, update: () => { currentModels = refreshed; }, }); },For the config-form
pi.registerProvider(name, { refreshModels }), callbacks that only return models remain unchanged; pi publishes the returned list. If such a callback previously usedcontext.storefor custom persistence, readcontext.storedand callcontext.publish({ persist: entry }). Inpublish(), omitpersistto leave storage unchanged, pass aModelsStoreEntryto write it, or passpersist: nullto delete it. - Breaking
Replaced the inherited pi-agent-core harness session model with the v4 lane-based
Session,SessionStorage, andSessionRepoAPIs, including durable operation records, global facts, shared sequence numbers, and tree-scoped lane views. - Breaking
Promoted the inherited v2 session and
AgentHarnessAPI from pi-agent-core's experimental entrypoint to its default export and removed the experimental subpaths. - Breaking
Removed the inherited legacy JSONL and in-memory repository APIs. Use pi-agent-core's v4
JsonlSessionRepoorInMemorySessionRepo, both implementing the newSessionRepocontract. - Breaking
Added the inherited required pi-agent-core
FileSystem.renameFile()operation for atomic JSONL publication; custom harness file-system implementations must provide same-filesystem replacement semantics - Breaking
Replaced experimental remote-session list summaries with durable
SessionMetadata;RemoteSession.sessionsno longer exposes runtime phase, model, thinking, attachment, or lock state, which remains available from acquiredSessionSnapshotvalues (#7708).
Added
Added built-in Baseten provider support with
BASETEN_API_KEYauthentication andzai-org/GLM-5.2as the default model.Added experimental remote-session client APIs: the transport-neutral
PiClient, CBOR protocol, Unix-socket transport, and@earendil-works/pi-coding-agent/clientRemoteSessioncontroller with transcript reducers. See Pi Client and Remote Protocol (#7344, #7348, #7371, #7409).Added
CredentialSynchronizationErrorfor credential changes that commit successfully but fail to synchronize local model state.Added chainable
pi.registerMarkdownTransformer()hooks for display-only transformation of user and assistant Markdown. Seepi.registerMarkdownTransformer()Added an experimental fullscreen TUI mode, selectable through
--tui-mode fullscreenor/settings(#7304).Added runtime switching between regular and fullscreen TUI modes through
/settings.Added a sticky editor, status, widget, and footer dock to fullscreen mode while keeping the transcript independently scrollable.
Added a draggable transcript scrollbar to fullscreen mode with configurable
auto,always, andhiddenmodes through/settings;alwaysreserves the rightmost column.Added page scrolling and marked-message navigation shortcuts to fullscreen mode.
Added an optional
scrollbarThumbtheme color for fullscreen scrollbar thumbs, falling back toselectedBg.Added configurable themed Unicode rendering for supported Mermaid diagrams in interactive messages, including optional rendering while streaming. See Markdown settings
Added opt-in
Ctrl+P/Ctrl+Nprompt history navigation, with explicit history bindings taking precedence over application shortcuts while the editor is focused.Added per-directory
AGENTS.override.mdcontext files, which replaceAGENTS.mdorCLAUDE.mdin the same directory while preserving context from other directories. See Context FilesAdded
AI_AGENT=pito CLI and RPC child-process environments for generic agent attribution. See Environment VariablesAdded inherited terminal-friendly Unicode rendering for LaTeX expressions in Markdown. See TUI Markdown.
Added stacked transient notifications in fullscreen mode.
Added arbitrary OpenAI-compatible model sampling parameters through
samplingParamsinmodels.json, model overrides, extension providers, and stream options. See Sampling ParametersAdded inherited opt-in vLLM
thinking_token_budgetsupport for OpenAI-compatible models, reserving output tokens for the final answerAdded inherited support for OpenAI-compatible streams that omit
finish_reason, usingcompat.supportsFinishReasonto infer normal and tool-use stops when the stream ends. See OpenAI Compatibility.Added inherited deferred provider request contracts, durable response handles, authenticated fetch/cancel dispatch, and faux-provider support for pending, ready, failed, and cancelled responses
Added inherited vendor-neutral telemetry contracts plus agent-owned typed AI-request and harness schemas, composed span starters, and callback helpers. See the agent telemetry schema reference.
Added inherited structured Amazon Bedrock failure diagnostics with HTTP status, modeled error code, and AWS request id when available
Added inherited
AgentOptions.shouldStopAfterTurnfor gracefully stopping after a completed turn before queued messages or another model call are processed. See Agent OptionsAdded inherited v4
JsonlSessionReposupport for append-only JSONL harness sessionsAdded inherited bounded branch-entry and indexed open-operation recovery queries to the v4 session API (#7448, #7646).
Added the inherited compile-complete
AgentHarnessv2 scaffold; unfinished operation paths reject withHarnessNotImplementedwhile durable execution is implemented.
Changed
Added inherited optional cancellation to pi-ai
ModelsStorereads, writes, and deletions; catalog orchestration binds these waits to the provider refresh signal.Reduced the inherited default fullscreen mouse wheel step from three lines to one for finer scrolling.
Fixed
Fixed the footer showing
(sub)for generic OAuth/OpenID sign-ins without a known subscription; extension OAuth providers can opt in withisSubscription.Fixed inherited OAuth token refreshes so stalled requests release the credential-store lock (#7508).
Fixed inherited tool argument validation to preserve values that already match an
anyOf/oneOfunion arm before coercion, avoiding nullable unions convertingnullto another primitive value (#7328).Fixed inherited Fireworks GLM 5.2 requests sending the unsupported
prompt_cache_retentionfield when long cache retention is enabled, and enabled session affinity for automatic prompt caching (#7676).Fixed inherited
JsonlSessionRepoenforcing session IDs globally across working directories; IDs are now unique within each working directory.Fixed inherited JSONL session forks and torn-tail repairs to publish atomically, avoiding partially written or corrupted sessions after interrupted writes
Fixed path-containing
findglobs returning no results on Windows (#6817).Fixed messages queued during manual
/compactfailing instead of being sent after compaction completes.Fixed Git Bash, MSYS, Cygwin, and WSL drive paths passed to built-in file tools resolving against the current Windows drive instead of their native drive (#7064, #7547).
Fixed project-level nested provider retry settings replacing unmodified global provider retry settings (#7572).
Fixed inherited GitHub Copilot Grok 4.5 requests to use the supported Responses API (#7560).
Fixed fullscreen shutdown leaking terminal capability-query replies into the parent shell prompt.
Fixed bare exact
--modelIDs shared by multiple providers choosing the first catalog entry instead of the sole authenticated provider or a clear ambiguity error (#7327).Fixed standalone x64 binaries requiring Haswell-era AVX2/BMI2 instructions by compiling release executables against Bun's baseline runtime
Fixed
Ctrl+Xcopy confirmations in fullscreen mode adding a transcript status line instead of showing the transientCopied!marker.Fixed Kitty image previews in fullscreen mode overlapping the sticky editor and footer dock while scrolling.
Fixed image-heavy fullscreen sessions lagging when layout changes retransmitted visible Kitty image payloads and rendered the transcript twice per frame.
Fixed spaces in
/settingssearches toggling the highlighted setting while typing multi-word queries such as TUI mode or Quiet startup.Fixed custom editors not inheriting the default editor's autocomplete dropdown item limit (#7333).
Fixed malformed resource arrays in package manifests crashing session startup (#7187).
Fixed the DOOM overlay example downloading its shareware WAD from a dead URL.
Fixed
setToolsExpanded(false)to be a no-op when tool output is already collapsed, avoiding redundantTool output: collapsedstartup notices from extensions (#7292).Fixed extension-driven model calls in custom compaction, handoff, and Q&A examples to dispatch through the coding-agent model runtime so custom providers and resolved auth options are preserved (#7325).
Fixed long-running sessions using stale credentials after another process updates
auth.jsonwithout serializing concurrent credential reads and delaying startup (#7319).Fixed concurrent
models-store.jsonreads forming a file-lock convoy and delaying startup.- Security
Updated the packaged
brace-expansiondependency to 5.0.8 to address GHSA-mh99-v99m-4gvg (#7316). Fixed forced model availability refreshes remaining blocked behind a stalled earlier refresh (#7301, #7421
Fixed
/modelcatalog refresh failures to identify every catalog that failed.Fixed provider login remaining stuck after saving credentials when a model catalog refresh stalls by separating local credential consistency from bounded background freshness (#7027, #7113, #7418).
Fixed
/scoped-modelswaiting for remote catalogs before rendering instead of showing cached models and cancelling refresh on close (#7153).Fixed
/model <name>waiting for catalog refresh before checking cached model matches (#7443).Fixed stale availability snapshots and errors publishing after a newer availability pass.
Fixed stale pi.dev, Radius, llama.cpp, and extension catalog refreshes publishing after a newer provider refresh.
Fixed cancellation while waiting for file-backed credential or model-catalog locks, preventing cancelled mutations from running or committing later.
Fixed concurrent in-memory credential mutations losing unrelated provider updates by serializing their read-modify-write sections.
- Security
Updated
undicito 8.9.0 and the packagedbrace-expansionto 5.0.9 to address GHSA-8xcm-r25x-g524, GHSA-4cwx-7wf7-3272, GHSA-m8rv-5g2x-5cg5, GHSA-jr45-8vmc-qm54, GHSA-v3r7-h72x-cjcm, and GHSA-rgw5-rvv9-x895. Fixed GitHub Copilot compaction and branch summaries using the Individual endpoint instead of the credential-resolved Business or Enterprise endpoint (#6768).
Fixed extension model calls dropping credential-resolved endpoints when forwarding request authentication, including custom compaction with GitHub Copilot Business and Enterprise accounts (#7579).
Fixed fullscreen transcript navigation leaving no editor-accessible
Home,End,PageUp, orPageDownvariants by adding Ctrl-modified editor bindings (#7574).Fixed extension event-bus listeners surviving session reloads and disposal
Fixed
/copyfailing to read clipboard text on Wayland when no X11 clipboard is available (#7387).Fixed slow connections failing during the initial connection attempt by increasing the connect timeout
Fixed oversized images returned by extension and built-in tools bypassing automatic image resizing. See Image settings
Fixed session discovery missing sessions stored through symlinked directories
Fixed manual compaction racing with threshold auto-compaction
Fixed responses truncated below their intended output limit ending the run instead of compacting and retrying once
Fixed Git package updates leaving dependencies missing when
git cleancannot remove an ignored dependency directoryFixed
findresults from POSIX and Windows filesystem roots losing the first path segment or gaining duplicate trailing separatorsFixed transient version-check, catalog, managed-tool, and package-management HTTP failures not being retried
Fixed interactive errors ignoring the configured output padding.
Fixed the inherited OpenCode Go provider display name.
Fixed inherited provider error normalization treating arrays and class instances as structured response bodies instead of preserving their original errors
Fixed inherited Anthropic streams dropping text or thinking included in the initial content-block event
Fixed inherited Google history conversion dropping signed empty text and thinking blocks required for replay
Fixed inherited OpenAI Codex cached WebSocket sessions being shared across different account credentials (#7364).
Fixed inherited transient Google Generative AI and Vertex AI provider errors bypassing automatic retries
Fixed inherited Gemini 3 tool call ids being discarded during history conversion, breaking signed multi-turn replay
Restored inherited GitHub Copilot models returned through account-specific policy responses
Replaced the inherited retired Qwen Token Plan
qwen3.8-max-previewmodel withqwen3.8-maxFixed inherited terminal width accounting for Indic conjunct grapheme clusters
Fixed inherited nested fullscreen stack layouts ignoring child minimum sizes.
Fixed inherited batched terminal color-scheme reports being parsed as one malformed response (#7550).
Fixed inherited terminal progress clearing to emit the complete OSC 9;4 sequence (#7581).
Fixed inherited iTerm2 image payloads omitting the size metadata required by the xterm.js image addon (#7612).
Fixed inherited width truncation leaving OSC 8 hyperlinks unterminated
Updated inherited GPT-5.6 Terra and Luna pricing across OpenAI and passthrough model catalogs.
Fixed inherited Fireworks Kimi K3 models to use the OpenAI-compatible API with native reasoning-effort levels and deferred tools (#7199, #7230
Updated the inherited Groq Qwen reasoning override for the replacement
qwen/qwen3.6-27bmodel.Fixed inherited Windows Shift+Enter detection by reading modifier state from the native Win32 helper.
Fixed the inherited pi-tui npm package omitting the source and build scripts needed to rebuild its Windows and Darwin native addons.
Fixed inherited Windows console truecolor detection when Windows Terminal does not provide
WT_SESSIONto child shells.Fixed inherited phantom fullscreen text selection from unmatched mouse events when changing terminal pane focus.
Fixed inherited keyboard input rendering latency on Windows by letting input preempt the throttled render timer.
Fixed inherited agent harness path handling on Windows for file basenames, recursive skill loading, and prompt template names.
- 0.83.x
0.83.0
29 Jul 2026- Breaking
Upgraded bundled TypeBox aliases to 1.3.7, removing deprecated APIs including
Type.Base,Type.Awaited,Type.Promise,Type.AsyncIterator,Type.Iterator,Type.Options, andValue.Mutate, while fixing compiled validation of nullable array tool arguments. Extensions using removed APIs must migrate to supported TypeBox APIs. See Package Dependencies Credential export for external clients —
pi auth print-api-keyandpi auth print-bearer-tokenexport configured credentials with automatic OAuth refresh and minimum-validity enforcement.Headless OpenRouter sign-in — Complete
/loginover SSH by pasting the redirect URL or authorization code when the loopback callback is unavailable. See OpenRouter.
New Features
Credential export for external clients —
pi auth print-api-keyandpi auth print-bearer-tokenexport configured credentials with automatic OAuth refresh and minimum-validity enforcement.Headless OpenRouter sign-in — Complete
/loginover SSH by pasting the redirect URL or authorization code when the loopback callback is unavailable. See OpenRouter.Claude Opus 5 on GitHub Copilot — Use Claude Opus 5 through GitHub Copilot with adaptive thinking and a 1M context window. See GitHub Copilot.
Breaking Changes
- Breaking
Upgraded bundled TypeBox aliases to 1.3.7, removing deprecated APIs including
Type.Base,Type.Awaited,Type.Promise,Type.AsyncIterator,Type.Iterator,Type.Options, andValue.Mutate, while fixing compiled validation of nullable array tool arguments. Extensions using removed APIs must migrate to supported TypeBox APIs. See Package Dependencies
Added
Added
pi auth print-api-keyandpi auth print-bearer-tokencommands for exporting configured credentials to external clients, including automatic OAuth refresh and configurable minimum token validity (#7168).Exposed the session's resolved model scope as
ctx.scopedModelsto extensions. See Extension ContextAdded inherited per-request
fetchinjection for supported text and image provider transports.Added the inherited
"pending"stop reason for partial streaming messages. See Custom Provider Stream PatternAdded inherited raw provider stop reasons across Google, Anthropic, Amazon Bedrock, Mistral, and OpenAI streams; unmapped terminal reasons now surface as provider errors instead of successful stops (#7272).
Added manual redirect URL and authorization-code entry to OpenRouter login for remote and headless environments. See OpenRouter
Added inherited Claude Opus 5 support for GitHub Copilot with adaptive thinking and a 1M context window. See GitHub Copilot
Changed
Changed inherited OAuth credential resolution to refresh tokens with less than five minutes of validity remaining instead of waiting until expiration (#7168).
Fixed
Added a status line when the tool output expansion is toggled (#7180).
Fixed file-backed
SYSTEM.mdandAPPEND_SYSTEM.mdprompts being omitted from the interactive startup context listing. See System Prompt Files (#7096).Fixed context files loading twice when a linked Git worktree is nested under its main repository. See Context Files
Fixed llama.cpp streamed responses reporting zero token usage and leaving session context accounting empty. See llama.cpp
Fixed session replacement and committed tree navigation during an active response to abort and persist the outgoing turn instead of leaving dangling tool calls. See Sessions
Fixed failed Git package installs leaving partial directories that blocked clean retries. See Install and Manage
Fixed the
/modelselector retaining a stale selection while filtering instead of highlighting the top matchFixed direct RPC bash commands bypassing extension
user_bashhandlers. See User Bash Events (#7214).Fixed skills, prompts, and themes losing package source metadata after extensions reload resources. See Resource Events (#6968).
Fixed cancellation of concurrently running user bash commands so every active command is aborted
Fixed duplicate messages appearing when extensions switch sessions during interactive startup
Fixed inherited Qwen Token Plan reasoning models to send their service-specific thinking controls and supported reasoning-effort levels (#6951, #6998).
Fixed inherited Z.AI output limits being sent through an unsupported parameter. See Providers
Fixed explicitly configured Amazon Bedrock profiles being overridden by ambient AWS access keys. See Amazon Bedrock
Fixed inherited image fallback paths overflowing narrow terminals, shortened home-directory paths, and made absolute paths clickable when terminal hyperlinks are available (#7262).
Fixed inherited OpenAI-compatible tool calls losing their function arguments when malformed deltas also contain an empty
customobject
- 0.82.x
0.82.1
25 Jul 2026Claude Opus 5 — Available on Anthropic and Amazon Bedrock with adaptive thinking (including
xhigh), inference profiles, and prompt caching. See Providers.Anthropic gateway bearer auth —
ANTHROPIC_AUTH_TOKENauthenticates against Anthropic-compatible gateways that requireAuthorization: Bearer, including compaction and branch summaries. See Environment Variables or Auth File.Faster, more resilient model catalogs — pi.dev catalogs revalidate with
If-None-Matchso unchanged providers answer with an empty304, and llama.cpp models stay listed across restarts. See llama.cpp.
New Features
Claude Opus 5 — Available on Anthropic and Amazon Bedrock with adaptive thinking (including
xhigh), inference profiles, and prompt caching. See Providers.Anthropic gateway bearer auth —
ANTHROPIC_AUTH_TOKENauthenticates against Anthropic-compatible gateways that requireAuthorization: Bearer, including compaction and branch summaries. See Environment Variables or Auth File.Faster, more resilient model catalogs — pi.dev catalogs revalidate with
If-None-Matchso unchanged providers answer with an empty304, and llama.cpp models stay listed across restarts. See llama.cpp.
Added
Exposed the
outputPadsetting to custom message renderers. See ExtensionsAdded inherited
ANTHROPIC_AUTH_TOKENbearer authentication for Anthropic-compatible gateways. See Providers (#5871).Added inherited Claude Opus 5 support for Anthropic and Amazon Bedrock with adaptive thinking, inference profiles, prompt caching, and preserved AWS validation messages
Changed
Changed pi.dev model catalog refreshes to revalidate with
If-None-Match, so unchanged provider catalogs answer with an empty304instead of a full download.Changed inherited Radius OAuth device authorization, token exchange, and refresh requests to use the configured gateway directly.
Changed inherited model loading errors to append the underlying cause, so auth failures such as
OAuth refresh failed for openai-codexreport the provider response instead of a bare wrapper message.
Fixed
Fixed compaction and branch summaries for providers whose authentication resolves entirely to request headers
Fixed unavailable scoped models being hidden from
/models, allowing them to be removed without editing settings manually (#6949, #7032Fixed startup context file discovery to skip directories that match context file names such as
AGENTS.md, which producedEISDIRwarningsFixed the llama.cpp extension to persist its model catalog, so llama.cpp models stay listed before the first successful refresh. See llama.cpp
0.82.0
24 Jul 2026- Security
Updated the packaged
protobufjsdependency to 7.6.5 to address GHSA-j3f2-48v5-ccww (#7005). Constrained tool sampling — Tools can prefer or require strict JSON Schema sampling or use OpenAI Lark/regex grammars, with model capability metadata preventing unsupported requests. See Constrained Sampling for Tools.
OpenRouter and Kimi Code sign-in — Use
/loginto authorize OpenRouter or a Kimi Code subscription without manually configuring API keys. See OpenRouter.
New Features
Constrained tool sampling — Tools can prefer or require strict JSON Schema sampling or use OpenAI Lark/regex grammars, with model capability metadata preventing unsupported requests. See Constrained Sampling for Tools.
OpenRouter and Kimi Code sign-in — Use
/loginto authorize OpenRouter or a Kimi Code subscription without manually configuring API keys. See OpenRouter.Session-aware, streaming bash integrations — Bash tools receive current session/model metadata, while direct RPC bash commands stream correlated output. See Bash Tool Session Environment and RPC bash events.
Added
Added inherited
Tool.constrainedSamplingwith strict JSON Schema (prefer/require) and OpenAI Lark/regex grammar variants across OpenAI, Anthropic, Amazon Bedrock, Google Gemini, and Mistral. See Constrained Sampling for Tools.Added inherited
supportsGrammarToolsandsupportsStrictToolscompatibility flags, expandedsupportsStrictModecoverage, and generated model capability metadata to gate constrained sampling.Added inherited Kimi Code subscription OAuth login for the Kimi For Coding provider, including device authorization and automatic token refresh
Added inherited OpenRouter OAuth PKCE login through
/login, minting a user-controlled API key. See OpenRouterExposed
PI_SESSION_ID,PI_SESSION_FILE,PI_PROVIDER,PI_MODEL, andPI_REASONING_LEVELto commands run by built-in and factory-created bash tools. See Bash Tool Session Environment.Added streaming
bash_execution_updateevents for direct RPC bash commands, correlated with request IDs. See RPC bash events
Changed
Changed inherited generated model catalogs to expose only provider-verified reasoning effort levels from models.dev
Fixed
Fixed inherited DNS lookup failures such as
getaddrinfo,ENOTFOUND, andEAI_AGAINto trigger automatic assistant retriesFixed inherited OpenRouter Anthropic cache breakpoints to advance through tool results and enabled cache control for
~anthropic/*-latestaliasesFixed inherited OpenAI Codex WebSocket sessions to retry once without a missing previous-response continuation after
previous_response_not_founderrorsFixed TUI debug and crash logs to respect custom agent directories instead of always writing under
~/.pi/agentFixed slow Ctrl+G external-editor startup when the system temporary directory contains many entries
Fixed startup resource display to preserve relative paths for sibling npm extensions loaded by a package
Fixed compaction and branch-summary requests to use fresh routing session IDs with prompt caching disabled where supported
Fixed explicit self-updates when
PI_SKIP_VERSION_CHECKis set (#6977).Fixed scoped model IDs containing brackets to resolve as literal exact matches before glob matching (#6210).
Fixed inherited OpenAI and Anthropic provider retry waits to honor abort signals and configured delay limits
Fixed fresh installs from preferring bundled model catalogs over newer remote catalogs because package file mtimes were newer
Fixed inherited editor scroll indicators overflowing narrow terminals
Fixed llama.cpp models to use the loaded context window as their output token limit instead of capping it at 16K
Fixed release source archives to include the generated provider model data used to build standalone binaries.
- Security
Updated the packaged
protobufjsdependency to 7.6.5 to address GHSA-j3f2-48v5-ccww (#7005). Fixed
/copyon Wayland to fall back to X11 or OSC 52 whenwl-copyfailsFixed
/modelto reload updatedmodels.jsonconfiguration when opening the model picker (#6999).
- 0.81.x
0.81.1
21 Jul 2026Verifiable release source archives — GitHub releases now include deterministic, checksummed source archives with instructions for rebuilding standalone binaries. See Building standalone binaries from release source.
Resilient compaction and branch summaries — Transient provider failures now follow the configured retry policy, with retry lifecycle events available to interactive, JSON, RPC, and SDK consumers. See Compaction & Branch Summarization and RPC retry events.
Added deterministic, checksummed source archives to GitHub releases with documented standalone binary rebuild instructions
New Features
Verifiable release source archives — GitHub releases now include deterministic, checksummed source archives with instructions for rebuilding standalone binaries. See Building standalone binaries from release source.
Resilient compaction and branch summaries — Transient provider failures now follow the configured retry policy, with retry lifecycle events available to interactive, JSON, RPC, and SDK consumers. See Compaction & Branch Summarization and RPC retry events.
Added
Added deterministic, checksummed source archives to GitHub releases with documented standalone binary rebuild instructions
Fixed
Fixed compaction and branch summarization to retry transient provider failures using the configured retry policy, with retry lifecycle events exposed to interactive, JSON, RPC, and SDK consumers
Fixed interactive startup waiting for background model catalog refresh while computing the footer provider count.
Restored the default stream fallback for extensions using the pre-0.81 agent-core API (#6915).
Fixed inherited Kimi K3 models from Moonshot AI and Moonshot AI China to use the OpenAI thinking format and expose reasoning effort support.
0.81.0
21 Jul 2026Local llama.cpp model management — Connect to a llama.cpp router, search and download Hugging Face models, and explicitly load or unload models with live progress. See llama.cpp.
Full provider extensions — Extensions can register complete pi-ai providers with authentication, model refresh, filtering, and custom streaming. See Register New Provider.
Qwen Token Plan providers — Use the built-in international and China subscription providers with regional endpoints and API-key authentication. See API Keys.
New Features
Local llama.cpp model management — Connect to a llama.cpp router, search and download Hugging Face models, and explicitly load or unload models with live progress. See llama.cpp.
Full provider extensions — Extensions can register complete pi-ai providers with authentication, model refresh, filtering, and custom streaming. See Register New Provider.
Qwen Token Plan providers — Use the built-in international and China subscription providers with regional endpoints and API-key authentication. See API Keys.
Expanded usage accounting — Tool, compaction, and branch-summary usage is persisted and included in session totals. See Compaction & Branch Summarization.
Added
Added Qwen Token Plan and Qwen Token Plan China to built-in provider setup, default model resolution, and provider documentation
Added the
get_available_thinking_levelsRPC command andRpcClient.getAvailableThinkingLevels()methodExported message and tool execution lifecycle event types from the package root
Added built-in llama.cpp router support with
/loginconnection setup and/llamaHugging Face model search and downloads, explicit loading, unloading, and live progress. See llama.cpp.Added extension registration for complete pi-ai providers, including native authentication, model refresh, filtering, and streaming behavior.
Added usage accounting for tools, compaction, and branch summaries in persisted sessions, footer totals, and session statistics
Fixed
Updated the packaged
brace-expansiondependency to 5.0.7Fixed persisted remote model catalogs from overriding newer bundled catalogs after an upgrade.
Fixed inherited stored API-key credentials to apply their provider-scoped
envvalues, including Amazon Bedrock profilesFixed inherited OpenAI-compatible cross-provider replay to keep tool call IDs unique when multiple calls share a provider call ID
Fixed inherited Kimi K3 thinking levels to expose low, high, and max, and normalized the
k2p7alias tokimi-for-coding.Fixed inherited OpenCode Go models routed through the OpenAI Responses API.
Fixed inherited
pi-aipackage metadata to avoid repeated consumer lockfile changesFixed inherited terminal shutdown to clear the editor's inverted software cursor before restoring the hardware cursor
Fixed inherited ANSI-aware text wrapping to recognize CRLF and CR line endings while preserving styles
Fixed inherited editor paste registry corruption after deleting and undoing paste markers, preventing literal or mismatched paste markers in submitted prompts (#6844).
Fixed sessionless OpenAI Codex WebSocket requests to use UUIDv7 request IDs
Fixed inherited GPT-5.6 Codex models to default to the 272K context window, avoiding automatic long-context pricing
Fixed messages queued during compaction to preserve steering and follow-up delivery behavior
Fixed read tool errors being syntax-highlighted as if they were file contents
Fixed llama.cpp router download progress updates and removed redundant wording from model action confirmations.
Moved automatic model catalog network refresh out of startup initialization and into the running interactive and RPC modes.
Fixed persisted sessions being read and parsed twice when opened, reducing startup latency for large sessions (#6793).
Fixed prompt-template defaults for all arguments (
${@:-default}and${ARGUMENTS:-default}) (#6695).Fixed obsolete custom UI, custom tool, and custom editor examples in the extension documentation (#6735).
Fixed Kimi Coding sessions to show API-equivalent implied costs with the subscription indicator.
Fixed OpenAI Responses early stream endings to trigger automatic retry instead of ending the agent run (#6727).
- 0.80.x
0.80.10
16 Jul 2026Kimi Coding thinking compatibility — Kimi Coding models now use adaptive thinking correctly; K3 exposes its supported
maxlevel and supports replaying empty-signature thinking blocks. See Kimi For Coding setup and Model Options.Fixed inherited Kimi Coding requests to use Anthropic adaptive thinking effort without token budgets, and enabled empty thinking signatures for K3 and
kimi-for-coding.Fixed inherited Kimi K3 pricing metadata for Moonshot AI and Moonshot AI China.
New Features
Kimi Coding thinking compatibility — Kimi Coding models now use adaptive thinking correctly; K3 exposes its supported
maxlevel and supports replaying empty-signature thinking blocks. See Kimi For Coding setup and Model Options.
Fixed
Fixed inherited Kimi Coding requests to use Anthropic adaptive thinking effort without token budgets, and enabled empty thinking signatures for K3 and
kimi-for-coding.Fixed inherited Kimi K3 pricing metadata for Moonshot AI and Moonshot AI China.
Fixed inherited Kimi Coding K3 thinking-level metadata to expose only the supported
maxlevel (#6737).Fixed inherited catalog generation restoring xAI models removed in 0.80.9 (#6736).
0.80.9
16 Jul 2026- Breaking
Removed Grok 3, Grok 3 Fast, Grok 4.20 variants, and Grok Code Fast 1 from the built-in xAI model catalog
Kimi K3 and deferred tool loading — Use Kimi K3 across built-in providers, including progressive extension tool activation through Kimi’s native protocol. See Dynamic Tool Loading, OpenAI Compatibility, and the
kimi-deferred-tools.tsexample.Added inherited Kimi K3 support for Kimi Coding, Moonshot AI, Moonshot AI China, OpenRouter, and Vercel AI Gateway.
New Features
Kimi K3 and deferred tool loading — Use Kimi K3 across built-in providers, including progressive extension tool activation through Kimi’s native protocol. See Dynamic Tool Loading, OpenAI Compatibility, and the
kimi-deferred-tools.tsexample.
Added
Added inherited Kimi K3 support for Kimi Coding, Moonshot AI, Moonshot AI China, OpenRouter, and Vercel AI Gateway.
Added Kimi deferred tool loading for extension-driven tool activation. See Dynamic Tool Loading, OpenAI Compatibility, and the
kimi-deferred-tools.tsexample.
Changed
Changed xAI login to use a prefilled device-authorization link labeled “Sign in with SuperGrok or X Premium,” and changed the default xAI model to Grok 4.5
Fixed
Fixed inherited Kimi K3 output limits for Vercel AI Gateway and OpenRouter models.
Fixed cloning or forking a session before its first assistant response to explain that the session must be saved first.
Removed
- Breaking
Removed Grok 3, Grok 3 Fast, Grok 4.20 variants, and Grok Code Fast 1 from the built-in xAI model catalog
0.80.8
16 Jul 2026- Breaking
Replaced the SDK's
CreateAgentSessionOptions.authStorageandmodelRegistryoptions with the asyncmodelRuntimeoption.AuthStorageand its storage backends are no longer exported; useModelRuntime(or a custom pi-aiCredentialStore), orreadStoredCredential()for one-off reads of auth.json. - Breaking
Removed redundant
ModelRuntime.getAll(),find(),getSnapshot(), andgetAuthOptions()projections. Use the pi-aiModelsmethodsgetModels(),getModel(),getProviders(), andcheckAuth()directly. - Breaking
Replaced SDK request-auth assembly through
ModelRegistry.getApiKeyAndHeaders()withModelRuntime.getAuth(). Passing a provider ID returns provider-scoped auth; passing a model also resolves built-in,models.json, and extension model headers.
New Features
Unified model runtime and provider authentication —
ModelRuntimecentralizes model configuration, provider-owned/login, and dynamic provider catalogs. See Providers.Live model catalog refresh —
/modelrefreshes configured providers in the background, andpi update --modelsforces an immediate refresh. See Install and Manage.xAI device-code OAuth and Grok 4.5 Responses support — Sign in to xAI with a device code and use Grok 4.5 with low, medium, or high thinking. See xAI.
Breaking Changes
- Breaking
Replaced the SDK's
CreateAgentSessionOptions.authStorageandmodelRegistryoptions with the asyncmodelRuntimeoption.AuthStorageand its storage backends are no longer exported; useModelRuntime(or a custom pi-aiCredentialStore), orreadStoredCredential()for one-off reads of auth.json. - Breaking
Removed redundant
ModelRuntime.getAll(),find(),getSnapshot(), andgetAuthOptions()projections. Use the pi-aiModelsmethodsgetModels(),getModel(),getProviders(), andcheckAuth()directly. - Breaking
Replaced SDK request-auth assembly through
ModelRegistry.getApiKeyAndHeaders()withModelRuntime.getAuth(). Passing a provider ID returns provider-scoped auth; passing a model also resolves built-in,models.json, and extension model headers. - Breaking
Changed extension-facing
ModelRegistry.refresh()from synchronousvoidtoPromise<void>becausemodels.jsonloading is asynchronous. Extensions must await it before making synchronous registry reads. - Breaking
Moved canonical dynamic catalog refresh to async
ModelRuntime.refresh()/pi-aiModels.refresh(). Legacy extension OAuthmodifyModelsremains supported as a synchronous compatibility projection after credential initialization.
Added
Added
ModelRuntimeas the canonical async SDK and internal model/auth facade while preserving the synchronous extension-facingModelRegistryAPI.ModelRuntime.create()accepts any pi-aiCredentialStorethrough itscredentialsoption.Added provider-owned
/logindiscovery directly from registered pi-ai providers, including ambient auth status and informational links.Added file-backed dynamic catalogs in
models-store.json, per-provider pi.dev catalog overlays, and Radius gateway support including offline migration from legacy credential-cached catalogs.Added extension provider
refreshModels(context)support for dynamic model discovery with optional provider-controlled persistence.Added
pi update --modelsto force an immediate model catalog refresh without updating pi or extensions.Added inherited xAI device-code OAuth login and Grok 4.5 OpenAI Responses support, with low, medium, and high thinking levels
Changed
Changed
ModelRuntimeto compose built-in providers, immutablemodels.jsonconfiguration, and extension overlays through ad-hoc pi-ai provider methods.Changed
ModelRuntimeto own final request assembly:getAuth(model)includes configured model headers, stream methods resolve auth once, andbefore_provider_headersruns as the Models-only header transform before provider dispatch.Changed
/modelto render the current model snapshot immediately, refresh configured providers in the background, and update the open selector with partial results or timeout errors.
Fixed
Fixed configured-provider catalog refresh to parse pi.dev's model-ID keyed responses, throttle checks to once per four hours, send the versioned pi user agent, treat unimplemented routes as unavailable overlays, and show concise refresh status in
/model.Fixed adjacent assistant thinking blocks to render as one thinking section.
Fixed inherited OpenAI Codex session IDs longer than 64 characters to meet the API limit (#6630).
Fixed inherited terminal output to normalize tab characters consistently
Fixed the Windows terminal title after checking npm packages (#6629).
Fixed Bun standalone binaries to bundle OAuth adapters for interactive logins.
0.80.7
14 Jul 2026- Breaking
Removed the
openai-responsescompat.sendSessionIdHeaderflag frommodels.json. Session-affinity behavior is now controlled bycompat.sessionAffinityFormat("openai","openai-nosession", or"openrouter"). ReplacesendSessionIdHeader: falsewithsessionAffinityFormat: "openai-nosession" Cache-friendly dynamic tool loading - Extensions can add tools during execution while supported Anthropic and OpenAI Responses models preserve prompt-cache prefixes. See Dynamic Tool Loading.
Message copy shortcut -
Ctrl+Xcopies the last assistant message in the transcript or the selected message in/tree, making older and branched messages directly copyable. See Display and Message Queue.
Breaking Changes
- Breaking
Removed the
openai-responsescompat.sendSessionIdHeaderflag frommodels.json. Session-affinity behavior is now controlled bycompat.sessionAffinityFormat("openai","openai-nosession", or"openrouter"). ReplacesendSessionIdHeader: falsewithsessionAffinityFormat: "openai-nosession"
New Features
Cache-friendly dynamic tool loading - Extensions can add tools during execution while supported Anthropic and OpenAI Responses models preserve prompt-cache prefixes. See Dynamic Tool Loading.
Message copy shortcut -
Ctrl+Xcopies the last assistant message in the transcript or the selected message in/tree, making older and branched messages directly copyable. See Display and Message Queue.Fable 5
xhighandmaxthinking - Nativexhighandmaxthinking levels are available across generated provider catalogs. See Model Options.
Added
Added cache-friendly dynamic tool loading for extension tools activated by tool results. Supported Anthropic and OpenAI Responses models load definitions where they become available, preserving the cached prompt prefix. See Dynamic Tool Loading (#6474).
Added inherited native
xhighandmaxthinking levels for Claude Fable 5 across all generated provider catalogsAdded
Ctrl+Xto copy the last assistant message, or the selected message in/tree.Added inherited
toolChoicesupport for OpenAI and Codex Responses, including required and named tool selection
Fixed
Fixed inherited OpenRouter model context windows to use the top provider's actual context length
Fixed inherited OpenRouter OpenAI-compatible session IDs to use the
x-session-idheader instead of OpenAI-specific session-affinity fieldsFixed
Ctrl+Vto paste clipboard text when the pasteboard does not contain an image.Fixed
/login amazon-bedrockto prompt for and save a Bedrock API key instead of only displaying ambient AWS credential setup instructions.Fixed inherited Amazon Bedrock ambient AWS credentials to keep using SigV4 authentication, including for custom model IDs
Fixed inherited Cloudflare Workers AI and AI Gateway authentication to use ambient account and gateway IDs when stored credentials contain only an API key
Fixed inherited legacy terminal decoding for Alt+symbol key combinations such as
Alt+,andAlt+.Fixed the GitHub Copilot
mai-code-1-flash-pickermodel to route through the/responsesendpointFixed branch summaries to work with providers that use ambient authentication instead of API keys
Fixed inherited Amazon Bedrock errors to report unhandled provider stop reasons instead of only
An unknown error occurredFixed npm package removal when installed packages have conflicting peer dependencies
Fixed inherited Azure OpenAI Responses reasoning replay when
encrypted_contentappears only in the terminal response eventFixed inherited Anthropic-compatible proxies that omit
usagefrommessage_deltaeventsFixed inherited OpenCode OpenAI Responses models to omit the unsupported
session-idheader while preserving other cache-affinity dataFixed system prompt cache invalidation across dates by removing the current date from the default prompt (#6621).
0.80.6
9 Jul 2026maxthinking level - New opt-in thinking level abovexhigh, natively supported on GPT-5.6 and adaptive Claude models, available across CLI (--thinking max), SDK, RPC, and model selection. Custom themes can definethinkingMax. See CLI Reference.Input-based pricing tiers - Request-wide input-token pricing tiers for accurate long-context cost accounting (e.g. GPT-5.4/5.5/5.6 long-context rates), also configurable for custom models in
models.jsonandmodelOverrides. See Model Configuration.Added the opt-in
maxthinking level across CLI, SDK, RPC, model selection, and themes. Custom themes can definethinkingMax; existing themes fall back tothinkingXhigh.
New Features
maxthinking level - New opt-in thinking level abovexhigh, natively supported on GPT-5.6 and adaptive Claude models, available across CLI (--thinking max), SDK, RPC, and model selection. Custom themes can definethinkingMax. See CLI Reference.Input-based pricing tiers - Request-wide input-token pricing tiers for accurate long-context cost accounting (e.g. GPT-5.4/5.5/5.6 long-context rates), also configurable for custom models in
models.jsonandmodelOverrides. See Model Configuration.
Added
Added the opt-in
maxthinking level across CLI, SDK, RPC, model selection, and themes. Custom themes can definethinkingMax; existing themes fall back tothinkingXhigh.Added request-wide input-token pricing tiers to custom model costs in
models.json,modelOverrides, and extension-registered providers.Added
~(home directory) expansion for theshellPathsetting
Fixed
Fixed inherited post-compaction output-token budgeting to ignore stale assistant usage from before the compaction boundary (#6464).
Fixed inherited GPT-5.4 and GPT-5.5 long-context cost accounting while retaining the intentional 272K default context limit for models that require an explicit override.
Fixed inherited GPT-5.6 metadata to keep direct OpenAI requests in the 272K short-context tier while exposing the Codex backend's 372K context window with long-context pricing, and removed the nonexistent bare
gpt-5.6alias.Fixed inherited Anthropic message conversion to preserve thinking blocks with empty thinking text but a valid signature instead of dropping them, avoiding thinking-block errors on newer Claude models
0.80.5
9 Jul 20260.80.3
30 Jun 2026Anthropic Claude Sonnet 5 support - Claude Sonnet 5 is available through inherited Anthropic-compatible and Bedrock provider catalogs with adaptive thinking enabled. See Providers and Model Options.
Configurable output spacing -
outputPadcontrols horizontal padding for user messages, assistant messages, and thinking blocks. See Settings.External editor configuration -
externalEditorlets Ctrl+G use a configured editor before$VISUAL/$EDITORfallbacks. See Settings and Keybindings.
New Features
Anthropic Claude Sonnet 5 support - Claude Sonnet 5 is available through inherited Anthropic-compatible and Bedrock provider catalogs with adaptive thinking enabled. See Providers and Model Options.
Configurable output spacing -
outputPadcontrols horizontal padding for user messages, assistant messages, and thinking blocks. See Settings.External editor configuration -
externalEditorlets Ctrl+G use a configured editor before$VISUAL/$EDITORfallbacks. See Settings and Keybindings.Richer RPC session tree access - RPC clients can inspect session entries and tree snapshots with
get_entriesandget_tree. See get_entries and get_tree.Extension session metadata updates - Extensions can observe session name changes through
session_info_changed. See session_info_changed.Modern Azure Foundry endpoint support - Azure OpenAI Responses provider setup supports current Microsoft Foundry endpoint URLs. See Azure OpenAI.
Added
Added inherited Anthropic Claude Sonnet 5 model support.
Added
get_entriesandget_treeRPC commands for reading session entries and tree snapshots over RPCAdded a package
./rpc-entryexport for launching Pi directly in RPC mode.Added session-name change events for extensions
Added inherited Azure OpenAI Responses support for modern Microsoft Foundry endpoint URLs
Added inherited
Usage.reasoningtoken counts for providers that report reasoning/thinking token usage (#6057).Added an
externalEditorsettings.json override for Ctrl+G external editor commands, with default fallbacks to Notepad on Windows andnanoelsewhere (#6122).Added an
outputPadsetting for user message, assistant message, and thinking horizontal padding (#6168).
Changed
Changed the default OpenAI model to
gpt-5.5.Changed inherited OpenAI Codex Responses SSE response-header waits to use the configured HTTP timeout instead of the previous fixed 20 second timeout, reducing false timeouts on slow connections (#4945).
Fixed
Fixed inherited Claude Sonnet 5 metadata to use adaptive thinking payloads for Anthropic-compatible and Bedrock requests.
Fixed inherited generated Xiaomi MiMo model pricing to match current pay-as-you-go pricing from models.dev (#6138).
Fixed inherited provider HTTP errors to include response bodies instead of opaque SDK messages
Fixed inherited
streamSimple()max-token caps so providers that count input and output against one context window do not reject long requests (#5595).Fixed inherited OpenAI Responses streams to preserve reasoning replay state when output items finish out of order (#6009).
Fixed inherited Z.AI preserved thinking requests to send
thinking.clear_thinking: falsewhen thinking is enabled, allowing replayedreasoning_contentto participate in provider caching (#6083).Fixed pre-prompt compaction to stop after compaction instead of continuing immediately
Fixed resource notifications to stay before messages when resuming sessions
Fixed startup benchmark timing output to print after TUI shutdown, preserve extension timings, and drain terminal-query replies before stopping benchmark mode
Fixed extension tool changes to apply before the next provider request in the same agent run without dropping
before_agent_startsystem-prompt overrides (#6162).Fixed a crash when undici emits an internal client error while terminating a mid-stream HTTP response (#6133).
Fixed the compaction event regression test to cover status indicator cleanup and keep CI passing.
Fixed interactive status indicators so ending work, retry, compaction, or branch-summary indicators no longer shrink the TUI when clear-on-shrink is enabled (#6026).
Fixed
--sessionandSessionManager.open()to reject non-empty invalid session files without overwriting them (#6002).Fixed user-message transcript rendering to keep visible backslashes in Markdown escape sequences such as
\"(#6105).Fixed assistant messages stopped by output length to show a visible incomplete-response error (#4290).
Fixed
--no-session --session-idso ephemeral CLI runs can use deterministic session IDs for provider cache affinity (#6070).Fixed disk BMP image files to be detected, converted to PNG, and attached through
readand CLI@fileinputs (#6047).Fixed auto-retry for provider stream errors that explicitly tell callers to retry the request (#6019).
0.80.2
23 Jun 2026Changed inherited pi-ai
ApiKeyCredentialto use theauth.json-compatible discriminatortype: "api_key"and provider-scopedenvvalues instead oftype: "api-key"and metadata.Renamed the inherited agent-core public harness shell execution options type from
ExecutionEnvExecOptionstoShellExecOptions.Fixed inherited Anthropic-compatible custom models to use explicit compatibility metadata instead of provider-name heuristics for session-affinity headers and unsupported tool-field omissions.
Changed
Changed inherited pi-ai
ApiKeyCredentialto use theauth.json-compatible discriminatortype: "api_key"and provider-scopedenvvalues instead oftype: "api-key"and metadata.Renamed the inherited agent-core public harness shell execution options type from
ExecutionEnvExecOptionstoShellExecOptions.
Fixed
Fixed inherited Anthropic-compatible custom models to use explicit compatibility metadata instead of provider-name heuristics for session-affinity headers and unsupported tool-field omissions.
Fixed inherited request-scoped
apiKeyandenvvalues to participate in provider auth resolution, so providers such as Cloudflare can derive request-specific base URLs from explicit call options (#6021).Restored inherited temporary legacy per-API stream aliases such as
streamSimpleOpenAICompletionson the pi-ai compat entrypoint (#6016, #6017).Restored inherited runtime
detectCompatfallback inopenai-completionsfor models without explicit compat metadata (#6020).
0.80.1
23 Jun 2026Fixed inherited Amazon Bedrock scoped
AWS_PROFILEendpoint resolution for built-in inference profile endpoints.Fixed inherited Fireworks Anthropic-compatible requests to apply session-affinity and unsupported tool-field defaults for custom Fireworks models.
Fixed inherited Together MiniMax M2.7 metadata to avoid unsupported Together reasoning toggles.
- 0.79.x
0.79.10
22 Jun 2026Extension compaction event context - Extension
session_before_compactandsession_compactevents now includereasonandwillRetry, so extensions can distinguish manual/compact, threshold auto-compaction, and overflow retry flows. See session_before_compact / session_compact and Custom Summarization via Extensions.Safer update flow -
pi updateinstalls the exact checked Pi version, and update notices show the changelog URL, making upgrades more predictable. See Install and Manage.Added
reasonandwillRetrymetadata to extensionsession_before_compactandsession_compactevents so extensions can distinguish manual, threshold, and overflow compaction flows
New Features
Extension compaction event context - Extension
session_before_compactandsession_compactevents now includereasonandwillRetry, so extensions can distinguish manual/compact, threshold auto-compaction, and overflow retry flows. See session_before_compact / session_compact and Custom Summarization via Extensions.Safer update flow -
pi updateinstalls the exact checked Pi version, and update notices show the changelog URL, making upgrades more predictable. See Install and Manage.
Added
Added
reasonandwillRetrymetadata to extensionsession_before_compactandsession_compactevents so extensions can distinguish manual, threshold, and overflow compaction flows
Fixed
Fixed the
findtool to respect nested git repository boundaries when parent.gitignorerules ignore the nested repo (#5960).Fixed the usage docs slash command table to include
/trustand/import(#5959).Fixed inherited OpenAI-compatible streaming to preserve encrypted
reasoning_detailsthat arrive before matching tool call deltas (#5114).Fixed broken TUI documentation links to the plan-mode extension example (#5957).
Fixed transient extension UI and session-start messages emitted during session replacement or reload so they remain visible, and kept reload input blocked until reload completes (#5943).
Fixed the plan-mode example to preserve active custom tools, skip the action prompt when no plan is found, and queue refinement/execution follow-ups correctly from
agent_end(#5940).Fixed
pi updateto install the exact version returned by the Pi update check, make--forcereinstall that checked version, fail instead of falling back to an unversioned reinstall when no version is available, and report both the old and updated versions.Fixed update notifications to display the actual changelog URL as the hyperlink text.
0.79.9
20 Jun 2026Chat-template thinking compatibility - OpenAI-compatible custom providers can map Pi thinking levels into
chat_template_kwargs, enabling vLLM/Hugging Face chat-template models such as DeepSeek to use provider-native thinking controls. See Custom Provider API Types and OpenAI Compatibility.GLM-5.2 provider improvements - GLM-5.2 now has corrected Fireworks OpenAI-compatible routing and OpenRouter
xhighthinking support, improving/modelbehavior and high-effort reasoning for GLM-5.2 users. See Model Options.Added inherited configurable
chat-templatethinking support for OpenAI-compatible providers that usechat_template_kwargs, such as DeepSeek models behind vLLM (#5673).
New Features
Chat-template thinking compatibility - OpenAI-compatible custom providers can map Pi thinking levels into
chat_template_kwargs, enabling vLLM/Hugging Face chat-template models such as DeepSeek to use provider-native thinking controls. See Custom Provider API Types and OpenAI Compatibility.GLM-5.2 provider improvements - GLM-5.2 now has corrected Fireworks OpenAI-compatible routing and OpenRouter
xhighthinking support, improving/modelbehavior and high-effort reasoning for GLM-5.2 users. See Model Options.
Added
Added inherited configurable
chat-templatethinking support for OpenAI-compatible providers that usechat_template_kwargs, such as DeepSeek models behind vLLM (#5673).
Fixed
Fixed inherited Fireworks GLM-5.2 metadata to use the OpenAI-compatible Chat Completions endpoint with
reasoning_effortsupport (#5923).Fixed same-directory session switches to reuse imported extension modules while preserving fresh extension instances and lifecycle events (#5905).
Fixed deep session branches taking quadratic time to build context or branch paths (#5909).
Fixed inherited OpenRouter GLM-5.2 metadata to expose
xhighreasoning and send OpenRouter's nativexhigheffort (#5770).Fixed inherited Markdown streaming code fence rendering so partial closing fences no longer make code blocks shrink or flicker while content streams
Fixed fuzzy
editmatches to preserve untouched line blocks instead of rewriting the whole file through normalized content (#5899).Fixed bash commands through legacy WSL
bash.exeto pass scripts over stdin so shell variables expand in the target bash (#5893).Fixed
/modelto hide GitHub Copilot models that are unavailable to the authenticated account (#5897).Fixed
/modelselector search to rank exact provider-prefixed matches before proxy-provider model ID matches (#5892).
0.79.8
19 Jun 2026Selective provider base entry points - SDK users can pair
@earendil-works/pi-ai/baseand@earendil-works/pi-agent-core/basewith explicit provider registration to keep bundled applications from including unused provider transports. Seepi-aiBase Entry Point andpi-agent-coreBase Entry Point.Mistral prompt caching - Mistral sessions now use provider-side prompt caching with session affinity and cached-token usage/cost accounting. See API Keys and Environment Variables.
Post-compaction token estimates - Compact results and compaction events now include estimated post-compaction token counts so clients can show the approximate context reduction. See RPC compact and compaction events.
New Features
Selective provider base entry points - SDK users can pair
@earendil-works/pi-ai/baseand@earendil-works/pi-agent-core/basewith explicit provider registration to keep bundled applications from including unused provider transports. Seepi-aiBase Entry Point andpi-agent-coreBase Entry Point.Mistral prompt caching - Mistral sessions now use provider-side prompt caching with session affinity and cached-token usage/cost accounting. See API Keys and Environment Variables.
Post-compaction token estimates - Compact results and compaction events now include estimated post-compaction token counts so clients can show the approximate context reduction. See RPC compact and compaction events.
OpenRouter Fusion alias -
openrouter/fusionis available as a built-in OpenRouter model alias. See API Keys.
Added
Added inherited
@earendil-works/pi-ai/baseand@earendil-works/pi-agent-core/baseentry points for selective provider registration in bundled applicationsAdded inherited Mistral prompt caching using the pi session ID as
prompt_cache_key, including cached-token usage and cost accounting (#5854).Added estimated post-compaction token counts to compact results and compaction events (#5877).
Added the inherited OpenRouter Fusion alias as
openrouter/fusion
Fixed
Updated vulnerable runtime dependencies, including
undiciand the packagedprotobufjstransitive dependency.Fixed compaction to refuse sessions with no eligible messages instead of producing empty summaries (#4811).
Fixed successful overflow-triggered auto-compaction to avoid retrying completed assistant responses (#5720).
0.79.7
18 Jun 2026Automatic theme mode -
/settingscan choose separate light and dark themes and follow terminal color-scheme changes. See Selecting a Theme.Self-only updates by default -
pi updatenow updates pi only, withpi update --allfor updating pi and packages together. See Install and Manage.Extension API helpers - extensions can use
CONFIG_DIR_NAMEfor project config paths and import edit diff helpers for edit-style diffs. Seectx.cwdand SDK Exports.
New Features
Automatic theme mode -
/settingscan choose separate light and dark themes and follow terminal color-scheme changes. See Selecting a Theme.Self-only updates by default -
pi updatenow updates pi only, withpi update --allfor updating pi and packages together. See Install and Manage.Extension API helpers - extensions can use
CONFIG_DIR_NAMEfor project config paths and import edit diff helpers for edit-style diffs. Seectx.cwdand SDK Exports.Warp inline images - Warp terminals now get inline image rendering through Kitty graphics detection. See Image.
Added
Added automatic theme mode so
/settingscan use separate light and dark themes and follow terminal color-scheme changes (#5874).Added inherited Warp terminal image capability detection so inline images render through Warp's Kitty graphics support
Exported
CONFIG_DIR_NAMEfrom the coding-agent public API so extensions can resolve project config paths without hardcoding.piExported edit diff helpers (
generateDiffString,generateUnifiedPatch, andEditDiffResult) from the public API for extensions that need edit-style diffs
Changed
Changed bare
pi updateto update only pi, addedpi update --allfor updating pi and extensions together, and clarified extension update prompts.Reserved
/in theme names for automatic light/dark theme settings.Updated extension docs, examples, runtime help, trust prompts, and config labels to use the configured project config directory instead of hardcoded
.pipaths.
Fixed
Fixed RPC unknown-command errors to include the request id so clients do not hang waiting for a response (#5868).
Fixed
/modelautocomplete and model selection searches to match provider/model queries regardless of whether the provider or model token is typed first.Fixed the tree navigator to horizontally pan deep entries so the selected item remains readable (#5830).
0.79.6
16 Jun 2026Fixed HTTP dispatcher configuration to preserve a caller's deliberate
fetchoverride instead of reinstalling the undici global fetch over it.Fixed inherited OpenCode Go DeepSeek V4 thinking-off requests to send the provider's
thinking: { type: "disabled" }compatibility parameter.
0.79.5
16 Jun 2026Provider-scoped API key environments -
auth.jsonAPI key entries can now includeenvoverrides for provider-specific Cloudflare, Azure OpenAI, Google Vertex, Amazon Bedrock, cache retention, and proxy settings without changing the project shell. See Auth File.Global HTTP proxy setting - Configure
httpProxyonce in global settings to applyHTTP_PROXYandHTTPS_PROXYto Pi-managed HTTP clients. See Network.Vercel AI Gateway attribution - Vercel AI Gateway requests now include Pi attribution headers by default. See API Keys.
New Features
Provider-scoped API key environments -
auth.jsonAPI key entries can now includeenvoverrides for provider-specific Cloudflare, Azure OpenAI, Google Vertex, Amazon Bedrock, cache retention, and proxy settings without changing the project shell. See Auth File.Global HTTP proxy setting - Configure
httpProxyonce in global settings to applyHTTP_PROXYandHTTPS_PROXYto Pi-managed HTTP clients. See Network.Vercel AI Gateway attribution - Vercel AI Gateway requests now include Pi attribution headers by default. See API Keys.
Added
Added Vercel AI Gateway request attribution headers (
http-refererandx-title) for Vercel AI Gateway modelsAdded an
xpfooter marker when experimental features are enabled.Added a global
httpProxysetting that applies asHTTP_PROXYandHTTPS_PROXYfor Pi-managed HTTP clients (#5790).Added
auth.jsonAPI keyenvvalues so provider-specific environment overrides can be scoped to Pi and propagated to inherited provider configuration (#5728).
Changed
Updated the vendored Markdown parser used by HTML session exports to
marked18.0.5.
Fixed
Fixed inherited OpenAI Responses streaming to tolerate null message content from OpenAI-compatible servers before tool calls (#5819).
Fixed inherited OpenCode DeepSeek V4 thinking requests to avoid sending both
thinkingandreasoning_effort(#5818).Fixed device-code login to stop opening the browser automatically.
Fixed inherited editor Cursor Up handling so non-empty drafts jump to the start of the line before browsing input history
Fixed inherited Z.AI GLM-5.2 thinking requests to send
reasoning_effortwith the provider'shigh/maxeffort mapping (#5770).Fixed successful
pi updateon Windows to exit naturally instead of callingprocess.exit(0), avoiding a Node.js/libuv assertion after version-check network requests (#5805).Fixed inherited Google and
google-vertexGemini model metadata to maplatestaliases to the current models, add Gemini 3.5 Flash for Vertex, correct Gemini 2.5 Flash Vertex cache pricing, and remove shut-down Vertex preview models (#5761).Fixed the session selector to stay open and show the all-sessions empty state when both current-folder and all-scope session lists are empty (#5747).
Fixed inherited Moonshot AI China model metadata to include Kimi K2.7 Code, and omitted unsupported thinking-off payloads for Kimi K2.7 Code models (#5760).
0.79.4
15 Jun 2026Automatic first-run theme selection - pi detects the terminal background on first run and defaults to the
darkorlighttheme. See Selecting a Theme.Standalone binary integrity checksums - GitHub release assets now include
SHA256SUMSfiles for verifying standalone binary downloads. See Quickstart Install.Added
SHA256SUMSintegrity files to standalone binary GitHub release assets (#5739).
New Features
Automatic first-run theme selection - pi detects the terminal background on first run and defaults to the
darkorlighttheme. See Selecting a Theme.Standalone binary integrity checksums - GitHub release assets now include
SHA256SUMSfiles for verifying standalone binary downloads. See Quickstart Install.
Added
Added
SHA256SUMSintegrity files to standalone binary GitHub release assets (#5739).Added first-run interactive theme detection from the terminal background
Fixed
Fixed bash tool output collection to keep draining stdout/stderr after the child exits while descendants still write, avoiding truncated late output
Fixed
/treehelp rendering to show compact wrapped controls instead of truncating them on narrow terminals (#5055).Fixed SIGTERM/SIGHUP interactive shutdown to keep signal handlers installed until terminal cleanup completes, preventing
signal-exitfrom re-sending the signal and leaving the terminal in raw/Kitty keyboard mode (#5724).Fixed extensions documentation to clarify that
pi.getActiveTools()returns active tool names whilepi.getAllTools()returns tool metadata (#5729).Fixed question and questionnaire extension examples to wrap long prompt, option, and help text instead of truncating it
Fixed package commands such as
pi list,pi install, andpi updateto terminate after completing even if an extension leaves background handles open (#5687).Fixed
pi updatefor pnpm global installs whose configuredglobal-bin-dirno longer matches the active pnpm home (#5689).Fixed npm package specs that use ranges or tags (for example
@^1.2.7) so installed package resources still load instead of being treated as mismatched exact pins (#5695).Fixed inherited Anthropic 1-hour prompt-cache write cost accounting to price 1-hour cache writes at 2x input instead of the 5-minute cache-write rate
Fixed inherited GitHub Copilot Claude adaptive-thinking effort metadata to match manually checked Copilot model capabilities (#4637).
Fixed inherited OpenCode/OpenCode Go completion model metadata to omit long-retention cache fields for routes that reject
prompt_cache_retention(#5702).Fixed inherited overlay compositing over CJK wide characters so borders stay aligned when an overlay starts inside a full-width cell (#5297).
Fixed inherited WezTerm inline Kitty image rendering during full redraw fallbacks so image padding rows are reserved before the placement is drawn without regressing tall-image placement (#5618, #4415).
Fixed custom provider config so plain uppercase API key and header values remain literals instead of being treated as legacy environment references; use explicit
$ENV_VARsyntax for environment variables (#5661).
0.79.3
13 Jun 2026Fixed inherited OpenAI GPT-5.4/GPT-5.5 and OpenAI Codex GPT-5.4/GPT-5.4 mini/GPT-5.5 context window metadata to use the observed 272k-token Codex backend limit, avoiding a billing hazard from prompts above Codex's accepted limit (reported
0.79.2
12 Jun 2026Clearer Bedrock validation guidance - Amazon Bedrock data retention validation errors now link to AWS data retention documentation. See Amazon Bedrock.
Added an experimental first-time setup flow behind
PI_EXPERIMENTAL=1that asks for a dark/light theme choice (preselecting the detected appearance) and opt-in analytics data sharing on first launch with the default agent directory; opting in stores atrackingIdinsettings.jsonAdded AWS data retention documentation links to inherited Amazon Bedrock unsupported data retention mode validation errors
New Features
Clearer Bedrock validation guidance - Amazon Bedrock data retention validation errors now link to AWS data retention documentation. See Amazon Bedrock.
Added
Added an experimental first-time setup flow behind
PI_EXPERIMENTAL=1that asks for a dark/light theme choice (preselecting the detected appearance) and opt-in analytics data sharing on first launch with the default agent directory; opting in stores atrackingIdinsettings.jsonAdded AWS data retention documentation links to inherited Amazon Bedrock unsupported data retention mode validation errors
Fixed
Fixed project trust detection to ignore global
~/.pi/agentstate when running from$HOME, and madepi updateuse only saved or explicit project trust without prompting (#5619).Fixed experimental first-time setup to skip forked sessions instead of rerunning the setup prompts
Fixed inherited OpenAI-compatible context overflow detection for parenthesized
maximum context length (N)errors (#5677).Fixed inherited OpenAI GPT-5.4/GPT-5.5 and OpenAI Codex GPT-5.4/GPT-5.4 mini/GPT-5.5 context window metadata to match current OpenAI limits (#5644).
Fixed inherited Anthropic refusal stops to preserve provider
stop_detailsexplanations in error messagesIncreased the inherited OpenAI Codex Responses SSE response-header timeout to 20 seconds to reduce false-positive stalls while retaining the bounded wait introduced for zero-event hangs (#4945).
Fixed inherited Claude Fable 5 thinking-off requests to omit Anthropic's unsupported
thinking.type: "disabled"payloadFixed inherited late tool progress callbacks after tool settlement to be ignored instead of emitting stale
tool_execution_updateevents (#5573).Fixed inherited user-message transcript rendering so standalone
+messages no longer render as-(#5657).Fixed inherited slash-separated fuzzy queries so provider/model completions remain matchable after insertion.
Fixed inherited WezTerm inline Kitty image rendering so reserved row clears do not erase all but the top strip of tool image previews (#5618).
Fixed inherited editor wrapping for CJK text to break at character boundaries instead of leaving large trailing gaps
Fixed inherited loose Markdown list rendering to preserve blank-line separation between list items
Fixed
--modelresolution for authenticated custom model IDs whose slash prefix matches an unauthenticated built-in provider (#5643).Fixed
/forkto keep session parent chains connected when the forked path contains labels (#5669).Fixed
/shareand/exportHTML exports to use the active fallback theme when the configured custom theme no longer exists (#5596).Fixed custom fallback model IDs with
:<thinking>suffixes to preserve the requested thinking level when the provider template model does not advertise reasoning
A whole project to check? Paste its package.json in the dependency check: every deprecated, outdated or vulnerable dependency, with what to upgrade to. Free, no account.