v1.26.0

✨ Features

• Local inference provider with llama.cpp backend and HuggingFace model management #6933
• Cerebras provider support #7339
• Moonshot and Kimi Code declarative providers #7304
• LMStudio declarative provider #7455
• Gateway to chat to goose via Telegram etc #7199
• MCP apps sampling support #7039
• Neighborhood extension in the Extensions Library #7328
• Computer controller overhaul with peekaboo #7342
• TUI client for goose-acp #7362
• GOOSE_SUBAGENT_MODEL and GOOSE_SUBAGENT_PROVIDER config options #7277
• Configurable OTel logging level #7271
• GoosePlatform in AgentConfig and MCP initialization #6931
• Gemini CLI streaming support via stream-json events #7244
• Expose context window utilization to agent via MOIM #7418
• Auto submit for recipes that have been accepted #6325
• Claude Code permission prompt routing for approve mode #7420
• Token counts displayed directly for "free" providers #7383
• Typescript SDK for ACP extension methods #7319
• Bedrock prompt cache #6710

🐛 Bug Fixes

• Windows MSVC linking issues #7511
• Searchbar z-index modal overlay #7502
• Out of order messages #7472
• Latest session always displays in sidebar #7489
• Color-scheme declaration for transparent MCP App iframes #7479
• Forward _meta in tool results to MCP Apps #7476
• Summon skill supporting files and directory path in load output #7457
• flake.nix build failure and deprecation warning #7408
• Truncated LLM responses detection in apps extension #7354
• Truncated tool calls that break conversation alternation #7424
• SQLite deadlocks with BEGIN IMMEDIATE #7429
• Correct colors for download progress bar #7390
• MOIM telling models to sleep while waiting for tasks #7377
• stderr noise #7346
• MCP app sampling support restored after revert #7366
• Skip whitespace-only text blocks in Anthropic message #7343
• goose-acp heap allocations #7322
• Trailing space from links #7156
• Low balance detection with prompt for top up #7166
• Display 'Code Mode' instead of 'code_execution' in CLI #7321
• Gemini CLI streaming restored #7291
• CLI handling of Reasoning content and streaming thinking display #7296
• OpenAI support for "reasoning" field alias in streaming deltas #7294
• UI revert app-driven iframe width and send containerDimensions per ext-apps spec #7300
• Ollama input limit override #7281
• Subrecipe relative path with summon #7295
• Extension selector not displaying correct enabled extensions #7290
• Subagent tool call notifications after summon refactor #7243
• UI preserve server config values on partial provider config save #7248
• Allow goose to run inside a Claude Code session #7232
• OpenAI route gpt-5 codex via responses and map base paths #7254
• Filter models without tool support from recommended list #7198
• Google thoughtSignature vagaries during streaming #7204
• Settings tabs getting cut off in narrow windows #7379

🔧 Improvements

• Simplified custom model flow with canonical models #6934
• Simplified text editor to be more like pi #7426
• CLI update replaced shell-based with native Rust implementation #7148
• OpenAI responses models with hardened event streaming handling #6831
• Canonical Models for context window sizes #6723
• Command to validate bundled extensions json #7217
• Unified streaming across all providers #7247
• New navigation settings layout options and styling #6645
• MCP-compliant theme tokens and CSS class rename #7275
• Redirect llama.cpp logs through tracing #7434
• Display working directory in UI #7419
• Remove allows_unlisted_models flag, always allow custom model entry #7255
• Local model settings access from bottom bar model menu #7378
• Use model ID everywhere for local model API #7382
• Improved link confirmation modal #7333
• Upgrade to rmcp 0.16.0 #7274
• Use working dir from session #7285
• Apps token limit #7474
• 3rd-party license copy for JavaScript/CSS minified files #7352
• Client settings improvements #7381
• Open recipe in new window passes recipe id #7392

📚 Documentation

• Groq models #7404
• Voice dictation updates #7396
• Excalidraw MCP App Tutorial #7401
• Gastown blog post: How to Use Goosetown for Parallel Agentic Engineering #7372
• Type-to-search goose configure lists #7371
• Search conversation history #7370
• Reasoning environment variable #7367
• Update skills detail page for Goose Summon extension #7350
• Sandbox topic update #7336
• Monitoring subagent activity section #7323
• Desktop UI recipe editing for model/provider and extensions #7327
• CLAUDE_THINKING_BUDGET and CLAUDE_THINKING_ENABLED environment variables #7330
• Permission Policy for MCP Apps #7325
• CLI syntax highlighting theme customization #7324
• Code Execution extension renamed to Code Mode extension #7316
• Escape variable syntax in recipes #7314
• OTel environment variable and config guides #7221
• System proxy settings #7311
• Summon extension tutorial and Skills references #7310
• Agent session id #7289
• Top Of Mind extension #7283
• Gemini 3 thinking levels #7282
• Stream subagent tool calls #7280
• Delete custom provider in desktop #7279
• Disable AI session naming #7194
• Playwright CLI skill tutorial #7261
• Community all-stars and page update #7483
• Neighborhood extension page with video embed and layout improvements #7473
• YouTube short embed to Neighborhood extension tutorial #7456
• Generate manpages #7443
• Goose v1.25.0 release blog [#7433](https://github.com/block/...

v1.25.2

🔧 Improvements

• Disable tool pair summarization via feature flag to improve stability

v1.25.1

fix: use BEGIN IMMEDIATE to prevent SQLite deadlocks (#7429)

Signed-off-by: rabi <ramishra@redhat.com>

v1.25.0

goose v1.25.0 is here

✨ Features

• macOS sandboxing — Goose Desktop now runs in a macOS sandbox for enhanced security [#7197](#7197)
• Unified summon extension — Replaced subagent and skills with a single, unified summon extension [#6964](#6964)
• MCP Apps UI integration — Integrated AppRenderer from @mcp-ui/client SDK with fallback request handler support [#7013](#7013) [#7208](#7208)
• Edit recipes from the GUI — Model and extensions of a recipe can now be edited directly in the desktop app [#6804](#6804)
• Load provider/model from recipe config — Recipes can now specify their own provider and model [#6884](#6884)
• Claude Code streaming support — Streaming output for the Claude Code CLI provider [#6833](#6833)
• Claude Code dynamic model switching — Dynamic model listing and mid-session model switching for Claude Code [#7120](#7120)
• Gemini CLI stream-json and session reuse — Gemini CLI provider now uses stream-json output and re-uses sessions [#7118](#7118)
• MCP support for agentic CLI providers — MCP extensions now work with agentic CLI providers (Claude Code, Codex, Gemini CLI) [#6972](#6972)
• Reasoning content in API — Expose reasoning_content in the API for reasoning models [#6322](#6322)
• Gemini OpenAI compat extra field — Support for the extra field in chat completion tool calls for Gemini OpenAI compatibility [#6184](#6184)
• OTel standard env var support — Respect standard OpenTelemetry environment variables for exporter selection [#7144](#7144)
• ACP model selection — Model selection support for session/new and session/set_model endpoints [#7112](#7112)
• Override bat themes — Allow overriding default bat syntax highlighting themes using environment variables [#7140](#7140)
• AGENT=goose env variable — Set AGENT=goose environment variable for cross-tool compatibility [#7017](#7017)
• Manpage generation — Automatic manpage generation for goose-cli [#6980](#6980)
• Top Of Mind (tom) extension enabled by default — The tom platform extension is now on by default [#7111](#7111)
• Standalone Goose Apps filtering — Apps page now filters to only show standalone Goose Apps [#6811](#6811)
• MCP Apps UI extension capability — Upgraded rmcp to 0.15.0 and advertise MCP Apps UI extension capability [#6927](#6927)
• Custom headers for OpenAI-compatible providers — Desktop now supports custom headers for custom OpenAI-compatible providers [#6681](#6681)
• SLSA build provenance — Release workflows now include SLSA build provenance attestations [#7097](#7097)

🐛 Bug Fixes

• Subrecipe relative path resolution with summon [#7295](#7295)
• Extension selector not displaying the correct enabled extensions [#7290](#7290)
• reasoning_content handling for Kimi/thinking models [#7252](#7252)
• Shell-escaping special characters except quotes [#7242](#7242)
• Dynamic port for Tetrate auth callback server [#7228](#7228)
• Fork session not working correctly [#7219](#7219)
• Deprecated skills extension now ignored gracefully [#7139](#7139)
• Filetype detection returning PLAIN_TEXT instead of HTML [#6885](#6885)
• Concurrent tool execution within the same MCP extension [#7202](#7202)
• Missing arguments in MCP tool calls causing GUI crash [#7143](#7143)
• Save and run recipe not working [#7186](#7186)
• Speech-to-text local functionality [#7181](#7181)
• MCP resource content decoding [#7155](#7155)
• Provider custom headers configuration [#7157](#7157)
• Keyring fallback now handled as success [#7177](#7177)
• Panic replaced with proper error handling in get_tokenizer [#7175](#7175)
• Text editor view broken [#7167](#7167)
• Assistant messages with tool_calls now always include content field [#7076](#7076)
• GCP Vertex AI model mapping handled correctly [#6836](#6836)
• Missing content from Langfuse traces [#7135](#7135)
• Extension selector syncing issues [#7133](#7133)
• Per-session Agent for model isolation and load_session restore (ACP) [#7115](#7115)
• Claude Code defensive coding improvements for model switching [#7131](#7131)
• Isolated claude-code sessions via stream-json session_id [#7108](#7108)
• Duplicated output in Code Mode by filtering content by audience [#7117](#7117)
• Strip empty extensions array in deeplinks [#7096](#7096)
• Goose Desktop missing Calendar and Reminders entitlements [#7100](#7100)
• Flatpak icon [#7154](#7154)
• Ask-ai bot hitting tool call limits [#7162](#7162)
• OTel monotonic_counter prefix and temporality env var support [#7234](#7234)
• Node installed in goose dir [#7267](#7267)
• Windows bundle random failure and optimised canary build [#7105](#7105)

🔧 Improvements

• Streaming markdown rendering [#7233](#7233)
• Improved compaction messages for better post-compaction agent behavior [#7259](#7259)
• Smaller system prompt for reduced token usage [#6991](#6991)
• Static Regex compilation for performance [#7205](#7205)
• Show directory in window title [#7138](#7138)
• Improved error message logging from Electron [#7130](#7130)
• Updated ElevenLabs API module and remove button UX [#6781](https://github.c...

v1.24.0

goose v1.24.0 Release Notes

✨ Features

• Unified summon extension — replaces subagent and skills with a single, streamlined extension [#6964](#6964)
• Top of Mind (tom) platform extension — new built-in extension enabled by default for contextual awareness [#7073](#7073) [#7111](#7111)
• MCP support for agentic CLI providers — Claude Code, Codex, and Gemini CLI can now use MCP tools [#6972](#6972)
• Claude Code: dynamic model listing and mid-session model switching [#7120](#7120)
• Claude Code: stream-json protocol for persistent sessions [#7029](#7029)
• Reasoning content in API for reasoning models [#6322](#6322)
• Configurable thinking level for Gemini 3 models [#6585](#6585)
• Image support and improved error resilience for Codex [#7033](#7033)
• Google provider: MessageContent::Image support for user messages [#6986](#6986)
• Edit model and extensions of a recipe from the GUI [#6804](#6804)
• Load provider/model specified inside recipe config [#6884](#6884)
• Escape Jinja variables in recipes to include them in prompts [#6975](#6975)
• Session content search via API [#7050](#7050)
• Inline rename for chat sessions in sidebar [#6995](#6995)
• Desktop UI for deleting custom providers [#7042](#7042)
• Custom headers field for custom OpenAI-compatible providers in Desktop [#6681](#6681)
• Display subagent tool calls in CLI and UI [#6535](#6535)
• Type-to-search filtering in CLI select/multiselect dialogs [#6862](#6862)
• AGENT_SESSION_ID env var exposed to extension child processes [#7072](#7072)
• AGENT=goose environment variable for cross-tool compatibility [#7017](#7017)
• Pass env to shell commands [#6863](#6863)
• Global config switch to disable automatic session naming [#7052](#7052)
• Port of Context (pctx) for Code Mode [#6765](#6765)
• Overridable default bat themes using environment variables [#7140](#7140)
• MCP Apps: Permission Policy support for sandbox iframes [#6947](#6947)
• MCP Apps: integrate AppRenderer from @mcp-ui/client SDK [#7013](#7013)
• Upgrade rmcp to 0.15.0 and advertise MCP Apps UI extension capability [#6927](#6927)
• Groq declarative data updated with Preview Models [#7023](#7023)
• ACP: model selection support for session/new and session/set_model [#7112](#7112)
• Support extra field in chat completion tool_calls for Gemini OpenAI compat [#6184](#6184)
• Manpage generation for goose-cli [#6980](#6980)
• SLSA build provenance attestations added to release workflows [#7097](#7097)
• CUDA support — allow building with CUDA as the candle backend [#6914](#6914)
• Proper proxy support on Windows & macOS [#6840](#6840)

🐛 Bug Fixes

• Save and run recipe not working [#7186](#7186)
• Local speech/whisper functionality [#7181](#7181)
• Handle keyring fallback as success [#7177](#7177)
• Replace panic with proper error handling in get_tokenizer [#7175](#7175)
• Text editor view broken [#7167](#7167)
• Configure add provider custom headers [#7157](#7157)
• MCP: decode resource content [#7155](#7155)
• Flatpak icon [#7154](#7154)
• Missing content from Langfuse traces [#7135](#7135)
• Extension selector syncing issues [#7133](#7133)
• Claude Code: defensive coding improvements for model switching [#7131](#7131)
• Duplicated output in Code Mode by filtering content by audience [#7117](#7117)
• ACP: per-session Agent for model isolation and load_session restore [#7115](#7115)
• Isolate Claude Code sessions via stream-json session_id [#7108](#7108)
• Windows bundle random failure and optimised canary build [#7105](#7105)
• Goose Desktop missing Calendar and Reminders entitlements [#7100](#7100)
• Strip empty extensions array when deeplink [#7096](#7096)
• Improve text-muted contrast for better readability [#7095](#7095)
• Allow unlisted models for CLI providers [#7090](#7090)
• command.process_group(0) for CLI providers, not just MCP [#7083](#7083)
• Only send command content to command injection classifier [#7082](#7082)
• Switch to Windows MSVC [#7080](#7080)
• Ensure assistant messages with tool_calls include content field [#7076](#7076)
• Remove Option from model listing return types, propagate errors [#7074](#7074)
• Lazy provider creation for goose ACP [#7066](#7066)
• XML tool call parsing fallback for Qwen3-coder via Ollama [#6882](#6882)
• Ensure animated elements are visible when prefers-reduced-motion is enabled [#7047](#7047)
• Show recommended model on failure [#7040](#7040)
• snake_case function names in code_execution instructions [#7035](#7035)
• Azure OpenAI model listing 404 during configure [#7034](#7034)
• Codex: propagate extended PATH to subprocess [[#6874](https://github.com/block/goose/pu...

v1.23.2

Don't swallow Tetrate errors  (#6998)

v1.23.1

Lifei/cherry pick fix dep (#7020)

Signed-off-by: rabi <ramishra@redhat.com>
Co-authored-by: Rabi Mishra <ramishra@redhat.com>

v1.23.0

✨ Features

• Ask AI Discord bot #6842
• WebSocket transport for goose-acp #6895
• Streamable HTTP transport for ACP + goose-acp usage #6741
• Laminar for Observability #6514
• Platform extension migrator + code mode rename #6611
• CLI flag to skip loading profile extensions #6780
• Thinking status hook #6815
• Hourly skill fetching #6814

🐛 Bug Fixes

• Apps now work in built copies of goose #6901
• Handle thoughtSignature differently for Gemini 2.5 vs 3 #6890
• Preserve working directory when creating new chat #6789
• Case-insensitive matching for builtin extension names #6825
• Correct tool support detection in Tetrate provider model fetching #6808
• Session manager fixes #6809
• Handle quoted paths with spaces in extension commands #6430
• Hardcoded J for newline key in missed locations #6853
• UI settings prompt injection update #6830
• Release build fix #6887

🔧 Improvements

• Extract ProviderDef trait and OpenAiCompatibleProvider #6832
• Remove dependency on goose-mcp from goose crate #6637
• Sync desktop_prompt with UI #6898
• Rejig dictation #6844
• Clean up client side code for creating schedule #6805
• Upgrade React and Electron to latest #6845
• Move persisting extension to session outside of route #6685
• ACP: load configured extensions and refactor tests #6803
• Swap canonical model from openrouter to models.dev #6625
• Centralize audience filtering before providers receive messages #6728
• Don't output compaction in term mode #6799
• Clean up build canonical warnings #6880
• Remove autogenerated .gooseignore files from repo #6824
• GitHub repo configurable for auto-updater and publisher #6828

📚 Documentation

• 5 tips for building MCP Apps that work #6855
• Session isolation #6846
• CLI newline keybinding #6823
• Usage data collection #6822
• Contributor instructions for hermit activation and npm/node versions #6727
• Copilot instructions: "No prerelease docs" update #6795

v1.22.2

fix: fixed the broken release (#6887) (#6888)

v1.22.1

upgrade react and electron to latest (#6845)