Skip to content

Releases

v18.22.0

Choose a tag to compare

@github-actions github-actions released this 16 Nov 21:54
3e8cff3

@oh-my-pi/pi-ai

Fixed

  • Fixed slow streaming of Gemma tool calls with large arguments.

@oh-my-pi/pi-coding-agent

Breaking Changes

  • Moved binary analysis from the standalone ida and ghidra tools into the backend-neutral dsm namespace (xd://eval/dsm).
  • Replaced the ida.enabled, ida.maxOpen, and ida.idleCloseSec settings with disassembler.backend, disassembler.maxOpen, and disassembler.idleCloseSec; shared disassembler database APIs are now available under @oh-my-pi/pi-coding-agent/disasm.

Added

  • Added Tern plan review editing and comments: double-click plans to edit them, select text to add Google Docs-style comments, and send comments to Refine.
  • Added Ghidra support with automatic setup of Ghidra, JDK, and PyGhidra when analysis is explicitly requested; Auto mode uses an installed IDA Pro when available and otherwise falls back to Ghidra.
  • Added support for sharing live analysis databases with child agents through db.handle() and dsm.open(handle).
  • Added a Force Remove option to worktree.onExit that removes session-created worktrees without prompting, including worktrees with uncommitted changes; new commits remain on the worktree branch.

Changed

  • Updated Tern plan review with a single Approve control, comment-aware Refine option, model selection, and a Save action that writes the plan while keeping review open and plan mode active.
  • Plain executable reads now display Mach-O, ELF, and PE metadata without opening an analysis database or downloading a backend; request an overview, function details, or dsm.open() to begin analysis.
  • Improved the model switcher (Alt+P / /switch) with provider filtering and per-switch thinking-level selection. In Tern, it now opens above the composer with integrated search, provider and Task/Main filters, and a thinking-level slider.

Fixed

  • Deleting or moving large directory trees inside a project no longer stalls language-server file watching (#19490 by @H4vC).
  • Reduced CPU and memory usage for long-running subagents as their sessions grow (#19486 by @H4vC).
  • Fixed commit.draft() incorrectly classifying files deleted or renamed through bash or eval as “Not yours”; they now appear under Review.

Removed

  • Removed the legacy standalone disassembler tools and their associated prompt instructions.
  • Removed the Copy and Edit-in-editor buttons from the Tern plan review sheet; use selection copying and double-click editing instead.

@oh-my-pi/collab-web

Fixed

  • Improved streaming performance in the collab viewer, preventing long replies from becoming progressively slower as they are received.

@oh-my-pi/pi-natives

Breaking Changes

  • TtyWriter now requires a drain threshold and callback and only supports independently reopenable TTYs. Its pending UTF-8 byte bounds are reported conservatively.

Changed

  • Large terminal repaints are handled more efficiently, reducing JavaScript-thread work and releasing oversized buffers after output drains.

Fixed

  • Fixed stalled terminal writers continuing to emit output or leaking file descriptors after shutdown.
  • Fixed terminal output preserving Unicode across chunk boundaries and safely replacing malformed surrogate sequences.

@oh-my-pi/pi-tui

Breaking Changes

  • Terminals that support output backpressure must now implement onOutputDrain(callback), which returns an unsubscribe function; terminals without backpressure can omit both pendingOutputBytes and onOutputDrain.
  • Updated PlanReviewOverlay options to use PlanReviewOverlayChoice objects and report choice IDs to onPick and selectOption. Refine and Save actions now reflect plan feedback and saving behavior; removed disabledIndices and externalEditorName, and updated describe to receive DescribeContext.

Added

  • Added configurable model search placement, provider scopes, and an optional thinking-level slider to ModelPickerComponent; selected thinking levels are reported through onPick metadata. The native picker sheet now supports anchoring, scope styling, level display, and icon actions.
  • Added a native plan review document and dock with editable and commentable plans, quote annotations, plan-edit notifications, Approve, Save, and Cancel actions, plus model selection.

Changed

  • Provider scopes now show only the selected provider's models when a provider is chosen, with scope cycling able to return to the full model list.
  • Updated thinking-level controls to use left and right arrow keys, and refreshed the slider and footer hint presentation.
  • Retitled the plan-saving overlay to “Save plan” and updated its action label.

Fixed

  • Fixed slow scrolling and typing in long agent transcripts.
  • Fixed slow typing and cursor movement in drafts longer than 256 lines.

@oh-my-pi/pi-wire

Added

  • Added optional picker icons for tooltips and accessible names, plus an option to control the placement of the search interface.
  • Added buttons, menu, and document component kinds, including menu-row action state changes and document text editing, commenting, and uncommenting events.

What's Changed

  • perf(coding-agent): stop copying the full message list and journal per publisher flush by @H4vC in #19486
  • perf(tui): stop editor wrap-cache thrash and full visual-line rebuilds by @H4vC in #19488
  • perf(ai): scan Gemma in-band tool calls incrementally by @H4vC in #19487
  • perf(coding-agent): forget vanished directories in the LSP file watcher without full scans by @H4vC in #19490
  • perf(collab-web): keep finished Markdown blocks' DOM while streaming by @H4vC in #19485
  • perf(tui): reuse the agent transcript viewer's rows between paints by @H4vC in #19489

Full Changelog: v18.21.0...v18.22.0

v18.21.0

Choose a tag to compare

@github-actions github-actions released this 16 Nov 15:27
6f03614

@oh-my-pi/pi-ai

Added

  • Added automatic detection of thinking capabilities for previously unknown models, with persistent results and support for alternate effort and token-budget request formats.

@oh-my-pi/pi-catalog

Added

  • Added Modal as a hosted inference provider.

Changed

  • Changed the local speech model from Kokoro to Phonon 31M.
  • Improved reasoning capability detection by preserving unverified thinking-level support and using endpoint-specific probe results when available.

Fixed

  • Fixed xAI Grok OAuth model grok-composer-2.5-fast being incorrectly identified as non-reasoning; it now supports thinking levels from minimal through xhigh.

@oh-my-pi/pi-coding-agent

Breaking Changes

  • Updated the AgentSession run lifecycle API: agent_end is emitted once when a run can no longer resume, and agent_pause reports runs that stop while awaiting work. Replaced hasPendingAsyncWork(), settleAsyncWork(), and suppressAutoResumeUntilPrompt() with isRunActive, wakeSources, and waitForRunEnd(). Calling abort() now also ends runs paused on background work; RPC, JSON print, collaboration, and extension streams retain their existing per-loop agent_end behavior.

Added

  • Added configurable automatic subagent prewalk through task.agentPrewalkAuto, allowing per-agent on/off settings or model patterns while preserving unconditional subagent prewalk.
  • Unknown custom and discovered models now automatically detect reasoning support and remember supported reasoning effort levels for future sessions.
  • Added inline image and web-page replies: supported terminals can display local, data, and HTTP(S) images and embedded pages, while other terminals receive linked fallbacks. Added tui.renderEmbeds to disable this rendering.
  • Added interactive embedded pages in Tern replies, including click-to-visit and Escape-to-return navigation; common host, path, and file: URL forms are accepted.
  • Bash redirects can now write to all writable internal URLs, including job cancellation endpoints and handler-backed agent://, cfg://, ssh://, and xd:// URLs.
  • Added image pasting to ask form notes and ui.formDialog options via acceptImages; pasted images are returned with [Image #N] markers and described for text-only models. Unsupported prompts now clearly reject image pastes.

Changed

  • Changed the default local text-to-speech model to Phonon 31M for faster synthesis, smaller downloads, and additional voices. Existing Kokoro voice settings migrate automatically; speaking rates now range from 0.8× to 1.3×.
  • Tern /btw responses now appear as folded, interactive notes at the point where they were requested, with follow-up, copy, branch, inject, and stop actions; /btw without a prompt still opens the history view.
  • Overlapping Bash calls now share shell state, including variables, functions, aliases, options, and traps, and background jobs started by those calls continue running independently.
  • Extensions enabled or reconfigured through extensions or disabledExtensions now load into the active session without requiring a restart.
  • Rich-text pastes made during approval steering or in dialogs now preserve Markdown formatting and images when inserted into the chat editor.
  • Moving a task or resuming a session in another project now switches its agents, extensions, custom tools, and MCP servers to match the destination project without a restart.

Fixed

  • Fixed status indicators incorrectly showing the agent as idle while subagents or background work could still resume the run.
  • Fixed subagents and security scans completing while child agents were still running or queued peer messages remained.
  • Fixed conflict-region resolution when a selected side is empty or shares edge lines with surrounding code.
  • MCP stdio server failures before a response now report the process exit code or terminating signal instead of only reporting closed stdout.
  • Fixed prewalk handoffs so model and reasoning-effort changes are preserved, and automatic handoffs return to the planning model when automatic prewalk is changed or disabled.

@oh-my-pi/pi-mnemopi

Changed

  • Improved file memory bank performance by reducing unnecessary disk synchronization during checkpointing, lowering per-turn I/O overhead for banks shared across multiple processes.

@oh-my-pi/pi-natives

Added

  • Added the PhononTts binding for on-device Phonon text-to-speech.
  • Added Shell.adopt(fork) to apply variables, functions, aliases, options, and traps from a completed fork back to its parent shell. When both shells set a name, the value from the command that finished last is retained.

Fixed

  • Fixed concurrent commitCreate() calls incorrectly overwriting each other’s commit subject and body.

@oh-my-pi/pi-tui

Added

  • Added support for pasting images into HtmlFormDialog option notes when opened with acceptImages, including image attachment, removal, and access to retained images through OptionNote.images.
  • Added live web-page and image rendering in assistant replies for supported terminals, including inline images and embedded page previews. Added the /btw transcript note for sharing side topics without interrupting the main conversation.

Changed

  • Improved reasoning-capability discovery so models awaiting discovery follow the session's thinking default and offer automatic thinking in addition to disabled and effort-level settings.
  • Renamed the public html-form/media APIs to media, including loadMedia, Media, and MAX_MEDIA_BYTES.

Fixed

  • Fixed fallback model selection so roles are offered only models they can use, excluding incompatible image and speech models.
  • Fixed macOS spelling assistance, autocorrect, replacements, and word completion for shell (!) and Python ($) drafts.

@oh-my-pi/pi-utils

Fixed

  • Fixed bare URLs swallowing the following word when Korean, Japanese, or Chinese text follows a parenthesized URL without a space.

What's Changed

  • add modal as a hosted inference provider by @andrewhinh in #18785
  • feat(task): add task.agentPrewalkAuto to judge subagent prewalk per assignment by @1vivy in #18913
  • fix(coding-agent): rescope agents, extensions, custom tools and MCP servers on /move by @yuzu-octopus in #19416
  • fix(tui): disable macOS spelling assistance for local-execution drafts by @aleister1102 in #12474
  • fix(mcp): report the stdio subprocess exit status on EOF by @adi-IL in #16482
  • fix(utils): end bare URLs at an unbalanced ) before CJK text by @nnnnoel in #19336
  • fix(vcs): isolate commit message files between concurrent commits by @eggpeat in #19345
  • fix(tui): filter model hub fallback picks by the role's accepted kinds by @Xytronix in #19371
  • fix(coding-agent): keep image pastes in dialogs out of the hidden chat editor by @Dangerboi in #19374
  • feat(coding-agent): let ask option notes take pasted screenshots again by @Dangerboi in #19376
  • fix(coding-agent): honor explicit choices during prewalk handoff by @taibenvenuti in #17292

New Contributors

Full Changelog: v18.20.3...v18.21.0

v18.20.3

Choose a tag to compare

@github-actions github-actions released this 16 Nov 01:38
9093536

@oh-my-pi/pi-ai

Breaking Changes

  • Removed the public isXdDevicePath export from providers/cursor-pi-args. Ranged reads of xd:// documents now use the standard line-selector behavior.

Fixed

  • Fixed rate-limit handling for Mistral and other OpenAI-compatible providers so requests with no available allowance fail immediately with an actionable error suggesting a model or plan change.
  • Fixed Anthropic OAuth requests for pooled Claude accounts to report accurate device and account information, including after credentials are rotated during a request.

@oh-my-pi/pi-catalog

Added

  • Added publicly available image-sizing helpers for determining model-delivered image dimensions and fitting images to a specified long edge before coordinate mapping.

Fixed

  • Fixed image inputs being dropped for DeepSeek V4 Flash, restoring consistent multimodal image support.
  • Restricted reasoning-effort options for Mistral GLM 5.3, GLM 5.3 Flash, and Mistral Large 4 to the levels supported by those models.

@oh-my-pi/pi-coding-agent

Breaking Changes

  • Renamed the browser tab.clickAt() option from clickCount to count, matching tab.click().
  • Browser viewport screenshot coordinates now use the displayed image coordinates when the screenshot is scaled for the model; pass css: true (or { x, y, css: true } for drag points) to provide viewport CSS pixels explicitly.
  • Added opt-in mnemopi.recallRelevance (off, auto, or on; default off) to filter automatic memory recall for task relevance using the judge role.
  • Added path-scoped entries for bash.patterns and bashInterceptor.patterns, resolved per working directory. Project and --config overlays now prepend these rules to global rules instead of replacing them.
  • Added bash.allowedEnvVars and per-rule env lists to permit explicitly named environment-variable prefixes on matching allow rules.
  • Subagents now preserve @upstream provider pinning on inherited models.
  • Added --no-judge, --no-spotlight, and --offline options to restrict auxiliary services to the active model's provider or disable cross-provider services.
  • Added per-rule TTSR repeatMode frontmatter (once, after-gap, or always) to override the global repeat policy.
  • Compound bash approval prompts now show each command segment separately, distinguishing allowed segments from those requiring approval.
  • Removed the seance agent, /seance command, and the task tool's sourceSession parameter.

Added

  • Added mnemopi.embeddingQueryPrefix and mnemopi.embeddingInputType configuration for asymmetric and instruction-tuned embedding models, including providers that require an input_type.

Changed

  • Browser screenshots are delivered to the model in CSS pixels up to the supported unscaled size, making viewport screenshot coordinates directly usable with browser actions; scaling guidance remains for scaled full-page and element captures.
  • Browser waitForUrl, waitForResponse, and requests({ filter }) now support * glob patterns in URL strings, while other strings retain substring matching.
  • The WebKit browser engine now automatically installs and updates to the newest stable ABKit release, with versioned installations that allow updates without waiting for existing browser hosts to exit.
  • Improved browser responsiveness and recovery: unresponsive WebKit pages fail fast with actionable errors, navigation can recover busy pages, and browser operations provide clearer diagnostics for timeouts, closed tabs, unsupported event listeners, invalid routes, cookie data, and viewport changes.
  • Pressing Delete at the end of the Ctrl+R search field removes the selected prompt from history.
  • /copy now works while viewing a subagent and copies from that subagent's transcript.
  • find now returns unranked literal keyword matches when all judge requests fail.
  • take() and drop() now reject nonexistent hunk units with a clear error.
  • edit rejection messages now suggest the file's current tag, and edits made before conflict resolution carry over to the resolved file.
  • /marketplace and marketplace install/uninstall commands now report picker and catalog-load errors without exiting the session.
  • ACP sessions honor mcp.startupTimeoutMs for MCP servers declared by the client, and ACP form elicitations no longer repeat field labels unnecessarily.
  • Task progress from every task.steps subagent step is now preserved, rather than only the final step's report.
  • Pasting HTML into the composer now preserves spaces from whitespace-only inline elements and indentation inside <pre> blocks.
  • Queued follow-ups edited with Alt+Up are restored as queued messages instead of being sent into the active turn.

Fixed

  • Fixed ida exec output on IDA 9.5+ by suppressing helper deprecation warnings and preserving the updated function metadata API.
  • Expired Perplexity logins and cookies no longer return an error page as a search answer; the provider is reported as expired and the search chain continues.
  • Fixed browser WebMCP discovery and invocation on Chrome and Chromium 157 and later.
  • Private hosts written with a trailing dot are now correctly handled by Ask and always-ask-cached before file tools access them.
  • Fixed the TUI freezing when switching sessions from an open tool-approval prompt.
  • Fixed /commit and eval commit dropping UTF-8 BOMs or mishandling quoted and Unicode filenames during change tracking and selective commits.
  • Fixed misleading Wayland clipboard warnings when pasting an image or when the clipboard is empty.
  • Fixed JavaScript evaluation after an in-place Linux upgrade by keeping session workers tied to the session's build.
  • Deleting a session that cannot be removed no longer deletes its prompt history and statistics.
  • Fixed read and write handling of line selectors on xd:// device URLs.
  • Fixed eval wait() rejecting or mishandling numeric delays; browser waits now direct users to the appropriate wait helpers.
  • Fixed browser open with an existing Arch Chromium profile, Tern navigation, closed external Tern blocks, and docked Tern viewport emulation.
  • Fixed Tern and WebKit browser calls that incorrectly exposed unsupported page event-listener APIs; errors now identify the supported alternatives.
  • Fixed browser timeout and busy-page errors to include the current URL, affected tab, and actionable recovery guidance.
  • Fixed Mnemopi embedding failures being silently downgraded to keyword search; provider errors now produce a warning, including in print mode on stderr.

@oh-my-pi/pi-mnemopi

Added

  • Added configurable embedding query prefixes and input types for instruction-tuned and asymmetric embedding models, including Qwen3-Embedding, E5, and NVIDIA NIM.
  • Added the embeddings.onRequestFailed callback, allowing hosts to handle and surface failed embedding API requests.

@oh-my-pi/pi-natives

Fixed

  • Committing or writing trees now correctly reports unresolved index conflicts instead of silently omitting conflicted files.
  • Fixed shell compatibility with GNU grep for end-of-line-anchored patterns, including -o and -z modes, and corrected sed substitutions on multibyte-character lines.
  • Fixed native diff and patch handling for filenames containing spaces and other characters requiring quoting, including binary changes.

@oh-my-pi/pi-utils

Fixed

  • Fixed HTML-to-Markdown conversion for fetched and pasted HTML to preserve whitespace in inline elements, maintain code indentation in preformatted blocks, and avoid inserting stray blank lines between block elements.

What's Changed

  • fix(edit): offer the current tag for the rows shown when a header tag is unknown by @Xytronix in #19332
  • fix(approval): gate private hosts written with a trailing dot by @XYZ1024-alt in #19228
  • fix(find): list literal keyword matches when every judge request fails by @yuzu-octopus in #19231
  • fix(browser): launch Chromium with the WebMCP feature enabled by @tuschla in #19232
  • fix(coding-agent): kept retained snapshots when a conflict:// write resolves a block by @tuschla in #19233
  • fix(coding-agent): report /marketplace picker load failures instead of crashing by @tuschla in #19234
  • fix(browser): reuse Chromium started through a compiled launcher by @tuschla in #19235
  • fix(coding-agent): closed pending tool approval when the turn aborts by @roboomp in #19242
  • fix(coding-agent): kept the JS eval worker on the running build after an in-place upgrade by @tuschla in #19259
  • fix(coding-agent): stopped Ctrl+V on Wayland warning about a missing xsel when the clipboard holds no text by @tuschla in #19260
  • fix(builtins): stop sed panicking on empty matches inside a character by @tuschla in #19261
  • fix(grep): match -o and -z end anchors at the end of the line by @tuschla in #19262
  • fix(utils): keep whitespace-only inline elements and
     descendants in turndown by @tuschla in #19263
  • fix(advisor): preserve skill-prompt user text past the body cap by @yuzu-octopus in #19267
  • fix(session): purge database rows only after transcript deletion by @eggpeat in #19271
  • fix(vcs): preserve filenames across native diff and patch round-trips by @eggpeat in #19274
  • fix(ai): derive Claude device identity from the serving account by @Xytronix in https://...
Read more

v18.20.2

Choose a tag to compare

@github-actions github-actions released this 14 Nov 20:08
7c49c11

@oh-my-pi/pi-ai

Added

  • Added the Antigravity CLI ACP provider with managed installation, Google sign-in, refreshable distributed credentials, host tools, and retained conversations across tool steps and history changes.

@oh-my-pi/pi-catalog

Added

  • Added Antigravity CLI models for Gemini, Claude and GPT-OSS with selectable native thinking levels.

@oh-my-pi/pi-coding-agent

Added

  • Added the antigravity-cli provider for Google Antigravity over ACP, with managed binary installation, /login antigravity-cli Google sign-in, host tool execution, and retained conversations across tool steps and history changes.

@oh-my-pi/pi-utils

Added

  • Added a profile-aware data directory for managed Antigravity CLI installations and isolated account state.

Full Changelog: v18.20.1...v18.20.2

v18.20.1

Choose a tag to compare

@github-actions github-actions released this 14 Nov 17:29
b7fe8c7

@oh-my-pi/pi-agent-core

Fixed

  • Corrected context cost estimates for requests that omit reasoning history, so billing estimates now include thinking text only when it is sent or reconstructed for a tool call.

@oh-my-pi/pi-ai

Added

  • Added setUnpausedProviders() to @oh-my-pi/pi-ai/oauth, allowing applications to re-enable providers paused by authentication rules; loaded credential pools and extension logins recognize the change on their next read.

Changed

  • Updated Vertex AI Gemini requests to use the v1beta1 API, restoring support for long Gemini responses.
  • Improved Google Antigravity quota recovery so Gemini 4 Argon access can resume after quota or plan changes when shared Gemini capacity is available.
  • Improved Amazon Bedrock region selection to honor a configured regional endpoint before falling back to us-east-1.

Fixed

  • Fixed Gemini 4 responses that exceed a single output segment across Google, Vertex AI, Gemini CLI, and Antigravity, ensuring long responses continue generating and are delivered as one complete reply.
  • Fixed Cursor conversations so hook-added prompt content remains in the current user message and conversation context is correctly preserved across turns.
  • Fixed Amazon Bedrock streaming so malformed responses fail promptly and interrupted responses are retried rather than treated as complete.
  • Fixed Claude account recovery when usage reports contain only one shared usage window.
  • Fixed OpenRouter conversations with Kimi and DeepSeek so reasoning context is preserved when switching models.
  • Fixed Xiaomi MiMo Token Plan authentication to use the correct regional endpoint, including for models loaded from bundled or cached configuration.
  • Fixed Perplexity sign-in in the login flow and setup wizard, including email-code and browser-based SSO authentication.
  • Stopped native judges from retrying gateway errors caused by exhausted upstream credentials while retaining retries for other transient 502 errors.
  • Fixed simple completion and authenticated gateway chat routes so supported providers receive stop-sequence and frequency-penalty settings.
  • Fixed Muse sessions after credential changes, key rotation, or idle resume so subsequent turns can continue normally.

@oh-my-pi/pi-catalog

Added

  • Added Gemini 4 Hydrogen on Google Antigravity as a single model with selectable low, medium, high, and xhigh thinking levels, priced at $1.50/$7.50 per million tokens.
  • Added Chutes provider support with live model discovery and /login chutes authentication using CHUTES_API_KEY.

Changed

  • Paused Google Antigravity: login is no longer available and stored credentials are not used until the provider is re-enabled. Use GEMINI_API_KEY with Google AI Studio or the google-vertex provider instead (#18341).
  • Updated Command Code model pricing and usage reporting to match the provider's current rates, including long-context pricing tiers for Claude Haiku 5.5, GPT-6.1 Sol and Terra, Grok 4.7, and Claude Sonnet 5.5.
  • Increased the maximum output for Gemini 4 Argon, Carbon, and Hydrogen on the Gemini API and Vertex AI to 262,144 tokens, enabling substantially longer responses.
  • Updated OpenCode Go Claude Haiku 5.5 to support thinking levels from low through max, including explicit thinking disablement.
  • Updated Devin requests to identify as Devin CLI 3000.12.57, the current stable version.

Fixed

  • Fixed Gemini 4 Argon on Command Code so image attachments are preserved and usage is billed correctly at $2 per million input tokens, $10 per million output tokens, and $0.10 per million cache-read tokens (#19125 by @wolfiesch).
  • Fixed Factory Droid GPT-6 models failing on the next turn after changing the thinking level during Bedrock or Snowflake sessions (#19188 by @will-bogusz).
  • Fixed OpenRouter model listings for workspace-restricted API keys so each key shows its correct allowed model roster and maintains an independent model cache (#19013 by @YibaiMeng).
  • Fixed OpenCode Go Muse Spark tool-call turns losing encrypted reasoning between steps (#17748 by @nico2525nn).
  • Fixed Vertex AI Gemini 3 and newer models rejecting replayed or cross-model tool calls that lack a thought signature (#17150 by @duck1eee).
  • Fixed Cursor model listings omitting models available to the provider, including GLM 5.3 and Grok 4.7 (#18248 by @joseotaviorf).

@oh-my-pi/pi-coding-agent

Added

  • Added /commit to create commits containing only changes made by the agent or a selected subagent, while preserving other agents’ and the user’s work. The workflow is also available through the eval commit namespace, supports targeting repositories outside the session directory, handles binary and non-UTF-8 files, symlinks, file mode changes, and initial commits, and can optionally skip commit hooks with no_verify; use commit.enabled to disable it.
  • Added auth.unpausedProviders (the Unpaused Providers setting) to re-enable paused providers such as Google Antigravity at your own risk while preserving their stored login state.
  • Added configurable long-turn todo reminders through todo.midRunNudges, including options to disable reminders, limit them to two, or continue with increasing intervals.
  • Pasting styled content into the composer now preserves Markdown formatting and places embedded images at their original positions, while code copied from editors remains plain text.
  • Added the session_clear hook and extension event, emitted when /clear resets the current session context without changing its session ID.
  • Added path-scoped Bash allow rules, project and configuration overlay precedence, and environment-variable controls through bash.allowedEnvVars and per-rule env settings.
  • Subagents now preserve @upstream provider pinning when inheriting models.
  • Added deletion of individual prompts from Ctrl+R history with the Delete key.
  • Added --no-judge, --no-spotlight, and --offline options to restrict or disable auxiliary services outside the active model’s provider.
  • Added per-rule TTSR repeatMode settings (once, after-gap, or always) to control reminder repetition.
  • Bash approval prompts now show compound command segments individually and identify which segments are allowed or require approval.
  • Todo done, drop, and rm operations now accept multiple tasks in a single request, including newline-, semicolon-, comma-, or space-separated input without splitting exact task-name matches.
  • Todo status lines now display task counts when multiple targets are involved.

Changed

  • /fork and fork-from-transcript now abort an active response and fork immediately.
  • Reduced default grep output by showing only matching lines and limiting matches per file; configure grep.contextBefore and grep.contextAfter to include surrounding context.
  • Changed the default spotlight.thinking level from low to med for deeper result analysis and improved ranking, at a higher request cost.
  • When Spotlight is enabled, commonly reread source and issue or pull-request content is shown more completely before outlining, including support for additional source file types.

Fixed

  • /loop now pauses after a provider error such as a rate limit instead of resubmitting the same prompt and attaching its @file mentions again (#19272).
  • Resuming a session whose directory was deleted now continues in the current directory with a warning instead of failing.
  • Model mentions and /switch completions now recover discovered models after provider discovery fails at startup, retrying discovery when needed.
  • Background job results that arrive after an ACP prompt ends are now carried into the next prompt instead of starting an invisible turn.
  • OpenAI server compaction failures now proceed to the next available compaction method instead of retrying server compaction on another model.
  • Language servers now recognize edits made shortly after a file is created, including on macOS.
  • Bash commands displayed in the Pseudo shell view now remain accurate while streaming through heredocs, substitutions, loops, continuations, arithmetic expressions, and background execution.
  • Resuming a session or reviving a subagent after a retry fallback with a thinking suffix (e.g. :xhigh) now restores the primary model's own thinking level instead of keeping the fallback's (#19273).

@oh-my-pi/collab-web

Fixed

  • Fixed guests retaining stale agents, tool cards, and working state after a failed rejoin; guests now reliably restore the current session state when reconnecting.

@oh-my-pi/pi-natives

Breaking Changes

  • Changed lsTree() to return { path, mode } entries instead of bare paths; mode matches git ls-tree output (for example, 100755 or 120000).

Added

  • Added noVerify to commitCreate() options to skip the pre-commit and commit-msg hooks, matching git commit --no-verify.
  • Added optimistic concurrency and alternate-index support to commitCreate(), allowing commits to use a separate index and reject stale commits when HEAD has changed.
  • Improved grep() and glob() performance for ovnfs checkouts by using the serving daemon when available while preserving native fallback behavior and existing streaming, filtering, and link-following options.

Changed

  • Improved shell search across local a...
Read more

v18.20.0

Choose a tag to compare

@github-actions github-actions released this 12 Nov 23:40

@oh-my-pi/pi-agent-core

Changed

  • prepareCompaction() runs as the compaction.prepare loop phase, so a ui.loop-blocked stall from a cold token walk over a long branch names compaction instead of "unknown" (#15027, #18867 by @wolfiesch)

@oh-my-pi/pi-ai

Added

  • Added the google-decisions judgment API for Google's Gemini Decisions API (gemini-decision-1), with text and image input (#19060 by @kevcube)
  • Added apiKey to successful OAuthAccessResolution results from oauth.accessById() and oauth.accessAll(): the key normal requests send for that account, which differs from accessToken for providers with structured credentials (#19030 by @Xytronix)
  • Added an optional exemptionAlias to ToolCallLoopGuard so transport calls can match the dispatched tool's name in exemptTools (#18860)
  • Added the liquid-systemone judgment API for Liquid AI's decision models, which rejects image sets over Liquid's 10,000 32×32-patch request cap before sending (#18076, #18100 by @SuperUserA)
  • Added the nace-systemone judgment API for Nace's Drex decision models, which sends images as Drex's top-level images data URLs and replaces the base64 image, audio, video and PDF data URLs that Drex rejects in judged text with a description (#18950 by @alphastorm)

Fixed

  • Fixed a spent Codex account retrying every minute for days when its siblings were only refused the requested model: their short model-policy blocks no longer count as a sibling becoming available, so the usage-limit wait follows the spent account's reset and model fallback can take over (#17637, #19019 by @wolfiesch)
  • Auth-gateway client tokens (serve --clients) can no longer set who a request is from, who pays, or how the gateway reaches the provider: chat routes drop their request headers, userProfileId, initiatorOverride, metadata, serviceTier, cacheRetention and the other gateway-only options, while the gateway token keeps them (#19067 by @alphastorm)
  • Fixed the auth gateway's /v1/pi/stream dropping taskBudget, include and anthropicPrefixMismatchBehavior, which its other chat routes accept (#19067 by @alphastorm)
  • Fixed requests to Vertex AI Mistral partner models (mistral-medium-3, mistral-small-2503, codestral-2) failing because they were sent in Anthropic's request format (#19010, #19022 by @yuzu-octopus)
  • Added Gemini's xhigh thinking level, sent on the Gemini API and Vertex for models whose ladder exposes it (Gemini 4 Hydrogen).
  • Fixed omp failing to start with ENOTDIR: not a directory, stat '~/.aws/credentials' when ~/.aws is a file instead of a directory (#19083 by @iamshatter)

@oh-my-pi/pi-catalog

Added

  • Added the Gemini Decision 1 judge model (google/gemini-decision-1, GEMINI_API_KEY); assign it to the judge role to use it (#19060 by @kevcube)
  • Added Nace (nace) with the drex-max-v1.1, drex-max-finance and drex-v2.0 judge models and /login nace (NACE_API_KEY) (#18950 by @alphastorm)
  • Added Gemini 4 Hydrogen (gemini-4-hydrogen) on Google, Vertex, OpenCode Zen, OpenRouter, Vercel AI Gateway, and Kilo, with low, medium, high, and xhigh thinking levels on the native Google transports.

Changed

  • Factory Droid requests identify as droid 1.7.1 and report the interactive CLI as their surface, and prefix-bound Claude models (Opus, Sonnet and Haiku 5.5, Fable 5.1) on Factory's direct Anthropic route now ask Anthropic to drop stale signed thinking instead of rejecting a turn whose earlier context changed (#18771 by @will-bogusz)

Fixed

  • Fixed alibaba-token-plan/qwen4-max-preview (including its qwen4.0-max-preview alias) showing no context or output limits, image input, or thinking levels (#18986).
  • Fixed Vertex AI Mistral partner models (mistral-medium-3, mistral-small-2503, codestral-2) misrouting to the first-party Gemini endpoint; they now ride the publishers/mistralai raw-predict endpoint with ADC auth and an OpenAI-shaped body (#19010, #19022 by @yuzu-octopus).
  • Added Liquid AI (liquid) with the d1.1 and d1-ultra judge models and /login liquid (LIQUID_API_KEY) (#18076, #18100 by @SuperUserA)

@oh-my-pi/pi-coding-agent

Breaking Changes

  • secrets.yml regex rules are now Rust regex patterns run natively, in linear time: lookarounds and backreferences are no longer supported, and a (?P<secret>…) group marks the part to redact (text around it is context that stays); \d, \w and \s are Unicode-aware. A regex rule that does not compile stops startup with an error naming the file and entry instead of being skipped. Migrate (?<=password=)\S+ to password=(?P<secret>\S+); see docs/secrets.md (#18965 by @H4vC)
  • Secret placeholders are now four camelCased words after the friendly name, e.g. GitHubToken_correctHorseStapleBattery, with no $$ delimiters, and the transcript shows the real value wherever the model wrote one. Placeholders in the old $$NAME_HASH:CASE$$ form, in sessions saved before this change, stay literal text, including in tool calls the model copies them into; start a new session to get working placeholders (#19014 by @H4vC)
  • What the model writes from placeholders (replies, thinking, tool calls, summaries) is stored as written and replayed byte-identical, keeping thinking signatures and prompt caches valid; model output that still holds a value (written while redaction was off, or before the secret was configured) is redacted on replay. Values a regex rule finds later no longer rewrite earlier history. A friendlyName is shown to the model as written and is dropped only when it spells out its own secret (#19014 by @H4vC)
  • Nix builds with withAppleFm = false no longer compile the Apple Foundation Models bridge with the host's Xcode (#18911)
  • Apple Silicon Nix builds include Apple Foundation Models support instead of silently building without it (#16584 by @honsunrise)
  • ask over RPC hosts without a form dialog now shows each prompt with the form's surrounding paragraphs, lists, preformatted blocks and tables instead of only the question (#18927)

Added

  • Added the Gemini Decision 1 judge (google/gemini-decision-1) for the judge role, and models.yml accepts the google-decisions API (#19060 by @kevcube)
  • ssh:// host listing, ssh:// autocomplete and /ssh list now include the Host aliases from ~/.ssh/config and /etc/ssh/ssh_config (following Include); ssh.json entries win on a name clash
  • Added find.verifyModel, a model selector whose judge answers only find's passage verification while filename and sketch ranking stay on the judge role (#19006 by @alphastorm)
  • Added omp bench --account to benchmark a specific stored OAuth account, and report rejected priority requests instead of silently benchmarking standard speed.
  • Text a browser or computer call types into a page or app, such as a password, no longer appears in plain text in an eval cell's status lines or the session file (#18981 by @will-bogusz)
  • Browser page-script observations in Tern and WebKit now mask password values, including values used as accessible names or descriptions (#18983 by @will-bogusz)
  • Chromium main-world waitForFunction calls no longer expose Puppeteer's internal ARIA query bindings to page scripts (#18883 by @will-bogusz)
  • On macOS a computer run that ends while secure input is on now starts with a note naming the app holding it and saying the user's...
Read more

v18.19.0

Choose a tag to compare

@github-actions github-actions released this 12 Nov 03:35
04cd1a7

@oh-my-pi/pi-ai

Added

  • Added the liquid-systemone judgment API for Liquid AI decision models, including validation that rejects image sets exceeding Liquid’s 10,000 32×32-patch request limit before submission.

Fixed

  • Fixed Gemini CLI and Antigravity responses losing or misordering thought signatures and reply text, including responses that follow tool calls.
  • Fixed Gemini CLI and Antigravity merging reply text around a thought signature, dropping signatures at the start of a reply or after a tool call, and placing a thought ahead of reply text that streamed before it (#18942 by @JustasMasiulis)
  • Fixed Google Antigravity judging Gemini models by quota buckets they do not consume: Gemini 3.x requests ignore the Gemini 4 Argon weekly bucket, and Argon requests ignore gemini-weekly, so an exhausted or low bucket no longer blocks, reserves, or caps an account for the other models (#16104).

@oh-my-pi/pi-catalog

Added

  • Added Liquid AI provider support with the d1.1 and d1-ultra judge models and Liquid authentication via /login liquid using LIQUID_API_KEY.

Fixed

  • Fixed Google Antigravity Gemini 4 Argon models sharing the general Gemini quota scope instead of their dedicated weekly bucket (#16104).

@oh-my-pi/pi-coding-agent

Breaking Changes

  • Changed the browser open API: ignore_https_errors is now insecure, allow_file_access has been removed, and file: pages can access other local files.
  • Changed the browser open and targets app option from an object to a string or CDP endpoint/executable path. Use argv instead of app.args and the top-level target instead of app.target.
  • Removed task.file and the task tool's taskFile field; subagent instructions must now be provided inline.
  • Removed blocked todo tasks and the blocked status/blocker metadata. Existing blocked tasks are restored as pending.

Added

  • Added systemPromptTemplates, which maps model-selector patterns to Handlebars template files for customizing system prompts. Templates re-render when switching models, with warnings and fallback to the bundled prompt for missing or invalid templates.
  • Added ordered multi-step tasks through task.steps. Subagents receive steps sequentially, and the final step determines the task result; interrupted tasks resume with their remaining steps.
  • Added Liquid AI judge models liquid/d1.1 and liquid/d1-ultra, including image resizing for Liquid's request limits.
  • Added per-step progress, state, and duration reporting for GitHub Actions watches, with expandable details and native Tern Surface Protocol views.
  • Added task.autoParallel (off, suggest, or on) to identify subtasks that can run concurrently and optionally delegate them in a single parallel task call.
  • Added portable awk, tar, lsof, shasum, md5, and strings support to the bash tool, including Windows support and internal-URI paths.

Changed

  • Browser sessions now prefer the bundled ABKit WebKit engine when available, with automatic installation and SHA-256 verification. WebKit provides headed, user-visible tabs and broader browser compatibility; Chromium remains the fallback and can be selected explicitly with app: "chromium", while app: "webkit" requires WebKit. Configure this behavior with browser.webkit and browser.webkitPath or the corresponding environment variables.
  • Browser visibility controls now show or hide WebKit tab windows immediately. Headed WebKit tabs can be interacted with directly, while Chromium tabs retain their existing mode-switching behavior.
  • In Tern environments with native DOM access, browser observation and interaction use trusted native input across frames and closed shadow roots, with caret-aware typing and case-insensitive text selectors.
  • edit.modelVariants now uses the longest matching model pattern, so specific patterns take precedence over generic ones regardless of configuration order.
  • The ask tool now submits the form's current values when its timeout expires; updated settings and documentation reflect this behavior.
  • The read tool loads detailed format selectors and URI sub-path syntax on demand, reducing the default request description while retaining inline details when read is disabled.
  • The default system prompt and tool descriptions have been streamlined, with guidance moved closer to the tools and detailed task/procedure references provided when relevant.
  • The bash pseudo view renders words built from literals and expansions as f-strings (f"repos/org/{r}/rulesets") instead of concat(...) calls.

Fixed

  • Fixed computer.close() so subsequent computer-control calls start a fresh desktop session instead of failing.
  • Fixed extension providers with oauth.modifyModels so they do not load the full model catalog before login.
  • Fixed resuming sessions so it no longer scans unrelated later session transcripts.
  • Fixed session saving on Windows to avoid UI stalls caused by repeated session-file access.
  • Fixed automatic chart selection so generated charts use compatible data and chart types.
  • Fixed bash previews for env -i commands so they display the requested command rather than generated shell setup code.

@oh-my-pi/collab-web

Fixed

  • Fixed task tool cards so spawned tasks consistently display their name, assigned agent, and instructions, with steps spawns shown as numbered stages.

@oh-my-pi/pi-natives

Added

  • Added shellBuiltins(sessionEnv?) to list commands provided by the embedded shell, including process controls and utilities.
  • Expanded the embedded shell with awk, tar, shasum, md5, strings, and lsof across Linux, macOS, and Windows. Destructive utilities can be disabled with PI_DISABLE_UUTILS_DESTRUCTIVE; when disabled, tar is delegated to the system.

Fixed

  • Fixed macOS text entry in Chrome and other Chromium-based applications so typing followed by Return opens the entered text instead of reloading the current page.

@oh-my-pi/pi-tui

Added

  • GitHub Actions watch cards now show current-step progress and can be expanded with Ctrl+O to display ordered step states and durations in ANSI and native Tern views.

Fixed

  • Fixed comparison charts mislabeling gaps against a rival column as improvements or regressions.
  • Fixed Tern edit diffs showing unrelated one-line hunks and incorrect line numbers around distant changes.
  • Fixed table charts incorrectly interpreting words such as “with,” “to,” and “from” in column names as comparison conditions.
  • Fixed chart row names containing trailing numbers, such as “EmbeddingGemma 2” and “Jev 15,” being split or incorrectly nested.
  • Fixed read rejecting colon-separated line ranges and displaying a single requested range as multiple rows.

Removed

  • Removed the blocked status from TodoStatus, the blocker field from TodoItem, and the block/unblock operations from TodoOperation.

@oh-my-pi/pi-utils

Added

  • Added browser utilities exposing trusted browser archive extraction limits and the installed WebKit host executable path.

What's Changed

  • feat(ai,catalog,coding-agent): add Liquid AI (d1-ultra, d1.1) judge provider by @SuperUserA in #18100
  • fix(collab): isolate test home directories and guard real ~/.omp paths by @iliaal in #18811
  • fix(ai): keep signature-only thought parts as signed thinking on Gemini CLI routes by @JustasMasiulis in #18942
  • perf(tui): release regenerable render caches on retired tool cards by @iliaal in #18820
  • perf(coding-agent,catalog): keep model cache reads provider-scoped and prune expired legacy rows by @iliaal in #18821
  • feat(coding-agent): add per-model system prompt templates by @DeprecatedLuke in #17835
  • perf(session): read only headers when looking for a session's continuations by @iliaal in #18952
  • feat(github): show step-level Actions watch progress by @anatoli-tsinovoy in #18945
  • fix(ai): isolate antigravity argon quota bucket by @roboomp in #16106

Full Changelog: v18.18.9...v18.19.0

v18.18.9

Choose a tag to compare

@github-actions github-actions released this 11 Nov 19:06

@oh-my-pi/pi-agent-core

Added

  • Tool calls get ctx.toolCall.earlierEffectsSettled, which settles once the earlier calls of their message that run beside them and may write files (their approval, asked with their arguments, is not read) have finished; speculative results wait for those effects before validation, and cancellation while waiting prevents their commit (#16274 by @will-bogusz)

@oh-my-pi/pi-ai

Added

Fixed

  • Devin requests no longer send temperature: 0: the catalog's temperature-floor raises it to 0.0001, so session titles and Mnemopi memory extraction work on Devin instead of failing with Cascade's opaque invalid_argument and silently falling back to another model (#18423)
  • Auth-gateway /v1/messages now answers 400 invalid_request_error for output_config.format (and the pre-GA output_format) instead of returning 200 with output the schema never constrained, so structured-output clients fall back to prompt-side schemas (#18842)
  • Auth-gateway requests denied only the requested model (Codex ChatGPT account, Cursor plan) now switch to another account and keep the denied one usable for other models, instead of marking a working credential suspect; gateway rotation now follows the same rules as local sessions, including waiting out a sibling's short usage block

@oh-my-pi/pi-catalog

Added

  • Added the web/sofya search model (#18185 by @yusufgurdogan)
  • Added the temperature-floor compat axis, assigned 0.0001 on Devin so a request that pins temperature: 0 reaches Cascade as a value it accepts (#18423)
  • Added the keyless web/hackernews search engine model for Hacker News stories and comments

Fixed

  • Devin requests now identify as Devin CLI 3000.12.43, the current stable release, instead of 3000.11.205 (#18766 by @will-bogusz)
  • Fixed Gemini 4 Argon and Carbon on Command Code showing no reasoning levels; both now offer low, medium, and high
  • Fixed OpenCode Zen Gemini 4 Carbon requests going to the chat-completions endpoint instead of the native Gemini endpoint Zen serves it on

Removed

  • Removed Devin's retired SWE-1.6 Fast from the models omp lists before Devin's live model list loads (first start, or when Devin is unreachable) (#18795 by @will-bogusz)

@oh-my-pi/pi-coding-agent

Added

  • Added Sofya as a web search engine (web/sofya), using SOFYA_API_KEY or a stored sofya credential (#18185 by @yusufgurdogan)
  • Documented autoresearch mode in docs/autoresearch.md: activation and its git gates, the three stop conditions, the per-segment iteration cap that defaults to none, the harness METRIC/ASI contract, and the on-disk store that OMP_AUTORESEARCH_DB_DIR relocates (#18655 by @sjawhar)
  • Extensions can import the form theme, the native span helpers (styledSpans, compactText, rowsText and the rest of the ANSI/span conversions), the xd:// URL helpers (parseXdUrl and friends) and the cfgExtensionTranscriptPresenters setting handle from @oh-my-pi/pi-coding-agent instead of internal module paths, which stop the extension from loading when a file moves (#18849 by @materemias)
  • /context shows loaded instruction files and skills in their own categories; System Prompt marks where each one sits without counting it twice, and notes when an extension or SDK callback changed the prompt after it was built (#17971 by @c-w-z)
  • Added Hacker News search (web/hackernews): web_search queries scoped to site:news.ycombinator.com now search HN stories and comments through the keyless HN Search API before the configured engines

Changed

  • A role edit in another omp instance no longer rewrites the task and eval tool descriptions of every running session, which was invalidating the Anthropic prompt cache from the tool definitions onward and dropping signed thinking blocks on prefix-bound models; the per-call model advertisement is now frozen into the base prompt surface and changes only when that surface rebuilds (#18716, #18732 by @yuzu-octopus)
  • Agent-to-agent messages now always reach the recipient as user turns: one that ends a wait arrives right after it instead of inside the wait tool result, and none is sent with developer or system authority
  • Launch services now persist for non‑interactive sessions
  • The shell's built-in jq now behaves like jq 1.8.2 instead of jaq: the same output (number formatting, colors, -S, --indent, --seq, --stream), error messages and exit statuses, jq's full option set (--version reports jq-1.8.2), and its grammar and builtins (?//, $__loc__, modules, getpath/setpath creating missing paths, |= deletions, Oniguruma regexes, dates); it passes jq 1.8.2's own test suite

Fixed

  • Nix builds with withAppleFm = false no longer compile the Apple Foundation Models bridge with the host's Xcode (#18911)
  • Apple Silicon Nix builds include Apple Foundation Models support instead of silently building without it (#16584 by @honsunrise)
  • On Windows, the bash tool runs external commands when omp itself runs in an AppContainer sandbox, instead of failing every one with Permission denied (exit 126) (#18800 by @devjay4561)
  • Explicit Bash prompt rules remain enforced in yolo mode when a command also matches a critical pattern (#18788 by @jasonharrison).
  • Fixed job widgets and captions failing outright when the widget model errors: they now try the role's next model (retry.fallbackChains.widget, else the built-in smol defaults) and fail only once every candidate has (#18793)
  • Failed job-widget, job-caption, cache-warm, composer-prediction, and subagent completion-probe calls now keep the provider's error message in their session usage records (#18792)
  • /changelog, /changelog full, and /changelog last N now list releases newest-first, matching CHANGELOG.md and the startup release notes (#18809)
  • Fixed a read placed after a bash call in the same message returning the file as it was before that command changed, created or deleted it, in primary and advisor turns (#16274 by @will-bogusz)
  • RPC/ACP /session delete now switches the live session before deleting its file, so later turns save to a new, resumable session (#18665).
  • Browser tab.click, tab.dblclick, tab.check, tab.uncheck, tab.uploadFile and frame.click on Chromium tabs now look their selector up again when the page replaces the element before the click goes out, including during and after delivery watcher setup, instead of waiting out the action timeout on the replaced element; a click through an element handle or an aria-ref whose element was replaced fails at once (#18676 by @will-bogusz)
  • A browser tab.run whose code uses document, window, location or another page-only global now says the code runs in omp's runtime, not the page, and to use tab.evaluate, instead of only "document is not defined" (#18840 by @will-bogusz)
  • JavaScript eval cells and tab.run/computer.run strings ending in assignments, blocks, loops, or declarations now return nothing; trailing expressions and explicit returns still produce values (#18865 by @will-bogusz)
  • A page calling window.print() no longer hangs the browser omp launches: print() returns at once and tab.pdf() still makes PDFs (#18838 by @will-bogusz)
  • tab.select() and element-handle select() return the options selected after the page's input and change handlers ran, instead of the choice the page then reset (Chromium tabs) or an option the page's handler had just replaced (cmux and Tern tabs) (#18829 by @will-bogusz)
  • tab.waitForDownload() fails, naming the path it looked for, when no file is there on omp's machine, as with a browser reached through app.cdp_url on another machine, instead of returning a path ...
Read more

v18.18.8

Choose a tag to compare

@github-actions github-actions released this 11 Nov 03:31

@oh-my-pi/pi-ai

Fixed

  • Fixed concurrent OAuth requests rejected as claude_code_version_too_old failing instead of retrying with the required version (#18651)
  • Chat Completions gateway requests preserve explicit tool strict-mode choices, so Codex tools can omit optional arguments when callers set strict: false (#18672 by @0xeb).

@oh-my-pi/pi-catalog

Added

  • Added Muse Ember (muse-ember-1.0) on Meta Model API and Muse Code with the minimal–max reasoning effort ladder and its >200K long-context pricing
  • Added Gemini 4 Carbon pricing on the Gemini API, Vertex AI, Google Antigravity, and Cursor ($2 input / $0.10 cached / $10 output per 1M tokens through 2026-11-30, then $4 / $0.20 / $20), instead of showing $0 (#18734 by @eggpeat)

Fixed

  • Fixed regenerating the catalog marking 18 reasoning models as non-reasoning when NanoGPT's <id>:thinking sibling row had larger limits (#18673, #18728)

@oh-my-pi/pi-coding-agent

Changed

  • curl -fsSL https://omp.sh/install | sh and install.ps1 now install the standalone binary by default, even when bun is present, so installs get delta updates; pass --source/-Source for a bun package install
  • Ask forms can attach notes to selected options; those notes accompany the answers separately from submitted form values (#18779 by @Shadorain)

Fixed

  • Claude auto-redeem now spends a full saved reset on an account blocked by its weekly limit when the reset also clears the Fable cap; such resets were treated as Fable-only and skipped while Fable usage had room (#18801)
  • Fixed agents showing as idle in the Agent Hub while still running after a failed streaming edit preview (edit.streamingAbort) auto-recovered the turn (#18790 by @H4vC)
  • A browser call on a tab that is not open now says to open it with await browser.open({ name }) (Python: await browser.open(name=…)) and the url and app it was opened with, instead of pointing at the removed action:"open" (#18739 by @will-bogusz)
  • Fixed computer.window({ app }) and computer.windows({ app }) matching similarly named apps when the exact app name has an open window, such as Google Chrome alongside Google Chrome for Testing (#18744 by @will-bogusz)
  • In Tern, browser role/ selectors keep input roles and names under Turkish and Azeri locales (#18658 by @will-bogusz)
  • Chrome role/RootWebArea waits with visible return the document instead of failing, and hidden waits keep waiting (#18658 by @will-bogusz)
  • An omp -p or other omp that your session's bash tool or eval cell starts inside Tern or cmux no longer picks Tern or cmux from the pane it inherited, so its browser tabs stop floating over your pane (#18773 by @will-bogusz)
  • Browser relay connections sharing a tab no longer lose cross-site frames when another connection disables auto-attachment on its page; automatic children detach locally when their last armed page leaves, while explicitly attached frame subtrees remain usable after opting out (#18657 by @will-bogusz)
  • Browser agents now say when a step can't be done through the page, instead of patching the page with scripts and reporting the step as done (#18761 by @will-bogusz)
  • An extension's assistant_message rewrite now reaches the next Codex Responses request: same-model native replay used to resend the captured original text while the transcript showed the rewritten one (#18312)
  • MCP reconnects and tool-list refreshes now keep extension-deactivated tools inactive while enabling newly discovered tools (#18678).
  • Exiting after aborting a bash command now waits for its TERM-resistant foreground processes to be killed (#18669).
  • An edit touching several files that anchors lines no read displayed now reveals those lines for every file in one rejection, so the identical resend applies instead of being rejected again for the next file (#18742 by @MaisonnatM)
  • omp auth-gateway stdio now serves judge, image, speech, embedding, rerank and video models: /v1/models lists them, and /v1/systemone answers typesafe/jev-* or @judge instead of 404, or 422 from a chat model (#18740 by @andyhite)
  • omp auth-gateway stdio no longer cuts its last responses off at 128 KiB when its input ends before a slow reader has read them (#18740 by @andyhite)
  • omp --session <id> --restore restores that session, as omp --restore --session <id> does, instead of opening the session picker: a bare --restore keeps a session --session or --resume already named (#18685 by @AdamDureyAtWillship)
  • The legacy Pi virtual module now builds in an npm install that keeps the bundled Pi packages nested: it is anchored to the installed package, so its imports resolve through that package's own dependency graph (#15208, #18733 by @yuzu-octopus)
  • User-level agents, commands, and other ~/.omp/agent config (including agents saved from the agents hub) now follow PI_CODING_AGENT_DIR, matching where omp agents unpack writes (#18684)
  • Fixed grep and glob treating a ; list that starts with an internal URL and includes a local path, such as artifact://1;notes.txt, as one address; each listed target is now searched (#18714 by @eggpeat)
  • Isolated tasks that edit a file whose name contains spaces now commit that file, including in a nested git repository, instead of leaving it unstaged or committing a different file that shares part of the name (#18715 by @eggpeat)
  • A disposed or switched session now releases every persistent shell it owns, including the ones a surface keeps under a name of its own (/loop conditions), instead of holding one native shell per surface for the life of the process (#18717 by @eggpeat)

@oh-my-pi/collab-web

Changed

  • Ask result cards render saved notes beneath their matching selected answers (#18779 by @Shadorain)

@oh-my-pi/pi-natives

Changed

  • macOS keys, typing and held keys for a window now go to a sheet attached to it that holds keyboard focus, such as an Open sheet or Finder's Go to Folder, as macOS sends them, instead of throwing InvalidTarget and naming the sheet; so do keys for a control of that sheet, which reports the window under it as its own (#18675 by @will-bogusz)
  • Computer input no longer fails with InputBusy the moment another omp session's call holds the desktop, such as an apps.open waiting for its window: it waits up to 5 seconds for that call to finish, on macOS, Linux and Windows (#18746 by @will-bogusz)

Fixed

  • Fixed a crash when parsing Markdown with very deeply nested lists or block quotes, such as a line of several hundred - markers, which made editing such files kill the process (#18707)
  • Fixed the shell's git builtin on Windows with bash.gitGuard (or smart git) on failing every allowed command with exit 127 (…\Git\cmd\git: No such file or directory); it now runs git.exe (#18722)
  • Computer element searches accept native role names, spaced role names such as text field, and radiobutton for macOS radio controls without changing printed accessibility roles (#18745 by @will-bogusz)
  • Fixed macOS background input and takeover refusing every call, with nothing sent, while Finder is in front with its focus on the desktop and a Finder window open; Finder gets its front status back with its focus left on the desktop (#18680 by @will-bogusz)

@oh-my-pi/pi-tui

Changed

  • Ask forms support notes on options: n on radio/checkbox rows and Alt+n on select ...
Read more

v18.18.7

Choose a tag to compare

@github-actions github-actions released this 10 Nov 20:23

@oh-my-pi/pi-ai

Added

  • LiteLLM usage adapter (@oh-my-pi/pi-ai/usage/litellm) reporting a gateway's key and user USD budgets (#17873 by @maykonmichel)
  • UsageApi.replaceConfiguredProviders registers configuration-sourced usage providers between runtime overrides and the built-in resolver, and UsageProvider.usesRequestKey makes aggregate usage probe only the key inference would send (#17873 by @maykonmichel)

Changed

  • Providers with a login-warning auth rule show their account risk before a new login and start it only after the user answers yes; any other answer or a cancel stores nothing, and their login-list entry is marked "(at your own risk)" (#18499)

Fixed

  • Claude Fable launch resets now report the Fable weekly limit they clear instead of an unknown window (#18661 by @TheAifam5)

@oh-my-pi/pi-catalog

Added

  • Added Muse Ember (muse-ember-1.0) on Meta Model API and Muse Code with the minimal–max reasoning effort ladder and its >200K long-context pricing
  • Added the login-warning "<text>" auth rule, which makes a new login accept the stated account risk first (#18499)

Changed

  • Resumed Google Antigravity behind a login warning instead of pausing it (#18499)

Fixed

  • Morph (api.morphllm.com) requests now send x-session-id, keeping a conversation on the worker that holds its prompt cache (#18704 by @wolfiesch)

@oh-my-pi/pi-coding-agent

Added

  • /usage and omp usage show LiteLLM key and user budgets for the built-in litellm provider and every models.yml provider with discovery.type: litellm, unless listed in disabledProviders (#17873 by @maykonmichel)
  • Extensions can ask the operator before an interactive quit, or cancel it, with session_before_shutdown; a further /exit or Ctrl+D skips the pending prompts and closes (#18424 by @aw1cks)

Changed

  • Saved-reset auto-redeem is set per window: codexResets.autoRedeem5h/autoRedeemWeekly and claudeResets.autoRedeem5h/autoRedeemWeekly/autoRedeemFable replace each provider's autoRedeem, whose value migrates into the 5h and weekly keys; a Claude Fable launch reset is governed by autoRedeemFable alone and spent only for a block that includes the Fable weekly cap, or before expiry when Fable usage is above zero (#18661 by @TheAifam5)
  • Sessions without a prompt UI (such as omp -p) no longer spend saved resets while auto-redeem is unset, even ones about to expire (#18661)
  • Google Antigravity is back in /login and omp login, marked "(at your own risk)": a new login first explains that Google may suspend accounts that sign in from third-party tools and continues only after you type yes; saved Antigravity logins work again (#18499)
  • On Windows, the bash tool's description tells the model it runs in an emulated bash environment, not cmd or PowerShell, so it writes bash instead of dir, %VAR% and other cmd syntax

Fixed

  • Extension prompts shown from session_before_* handlers no longer time out after 30 seconds while the operator is still answering (#18424 by @aw1cks)
  • A second exit request now skips pending worktree.onExit prompts and keeps those worktrees instead of waiting for an answer (#18424 by @aw1cks)
  • Advice that makes the agent keep working after it finished no longer comes back a second time, often as a blocker, while the agent is fixing it: advisor reviews of the earlier answer are cancelled and the advisor reviews the fix instead (#18735 by @H4vC)
  • On Windows, the bash tool's shell resolves MSYS drive paths (/c/Users/..., /mnt/c/...) everywhere, so running /c/.../tool.exe, cd /c/..., [ -f /c/... ] and < /c/... reach drive C: like they already did for ls and other built-in utilities
  • On Windows, cmd-style command lines such as dir | find /c /v "", sort /r, timeout /t 5 and whoami /all in the bash tool run the Windows programs they were written for, instead of pi's GNU-style find/sort/timeout/whoami (which read /c as drive C: and could walk the whole drive)
  • Changing or unsetting PATH in the bash tool's shell now forgets remembered command locations, as bash does, so a command no longer runs from a directory that has left PATH
  • One slow or flooding collab guest no longer ends sharing for the room: the host drops that guest's backlog and asks it to rejoin, sessions larger than the send budget can be joined, and guest questions nobody received settle as unavailable (#18629 by @iliaal)

@oh-my-pi/pi-tui

Changed

  • grep and ast_grep results with no matches render as one line with 0 matches in the warning color, dropping the separate No matches found row (#18700 by @JustasMasiulis)

Fixed

  • Saved-reset messages name the Claude Fable weekly limit as "Claude Fable weekly" instead of its raw id (#18661 by @TheAifam5)

What's Changed

  • feat(extensions): add an interactive pre-shutdown hook by @aw1cks in #18424
  • fix(tui): show grep's 0 matches in yellow on one line by @JustasMasiulis in #18700
  • fix(shell): resolve MSYS drive paths (/c/...) in the engine on Windows by @H4vC in #18703
  • fix(catalog): pin Morph conversations to one worker with x-session-id by @wolfiesch in #18704
  • fix(shell): run cmd-syntax command lines through the Windows programs pi's utilities shadow by @H4vC in #18711
  • feat(coding-agent): split saved-reset auto-redeem into 5h, weekly and Claude Fable settings by @TheAifam5 in #18661
  • fix(advisor): stop superseded reviews from re-raising advice the primary is already fixing by @H4vC in #18735
  • fix(collab): keep one guest from ending sharing for the room by @iliaal in #18629
  • feat(ai): report LiteLLM key/user budgets in /usage and omp usage by @maykonmichel in #17873

New Contributors

Full Changelog: v18.18.6...v18.18.7