Goose

No entries match these filters. Clear filters

1. v1.52.0 Sep 23, 2026 · release notes

   Goose v1.52.0 adds live voice conversations, Z.AI provider, OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code./Jev decision crates, and custom OpenAI base URLs.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.52.0 https://github.com/aaif-goose/goose.git
   # already have the repo? check out this version:
   $ git checkout v1.52.0

   • Adds live voice conversations capability in the desktop app for real-time audio interaction
   • Introduces decisions provider crate with OpenRouter and Jev implementations for agent decisioning logic
   • Adds Z.AI Coding Plan provider with streaming tool calls support for code generation workflows
   • Enforces recipe parameter limits (max 32 params, 200 select options, 128 KiB) to constrain recipe inputs
   • Supports OpenAI provider custom base URL configuration in the SDK for proxy or local endpoint routing

   +1 more show less

   • Adds support for Opus 5.5, GPT-6-sol, and GPT-6-luna models via the canonical registry

   6 details in Everything

2. v1.51.0 Sep 17, 2026 · release notes

   Goose v1.51.0 adds GPT-live API support, EUrouter declarative provider, and ACP prompt meta routing.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.51.0 https://github.com/aaif-goose/goose.git
   # already have the repo? check out this version:
   $ git checkout v1.51.0

   • Adds GPT-live API support for real-time streaming interactions
   • Introduces EUrouter as a declarative provider for gateway pairing
   • Routes Desktop sessions to the state-machine loop via ACP prompt meta
   • Adds operator allowlist configuration for gateway pairing security

   4 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !Removes the --planning-mode flag and associated planning functionality from the CLI
   • !Removes the create recipe subcommand entirely from the CLI interface
3. v1.50.0 Sep 8, 2026 · release notes

   Goose v1.50 adds GPT-6 Astra models, tool callingtool callingA model's ability to answer with a structured request to run a named function or API, which the application executes and returns. It is how agents take actions. for goose-agent, Databricks AI Gateway path configuration, and AIML API integration.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.50.0 https://github.com/aaif-goose/goose.git
   # already have the repo? check out this version:
   $ git checkout v1.50.0

   └──▷ USE IT
   Configure the Databricks AI Gateway path for Kotlin-based integrations to direct requests to a specific gateway endpoint.
   shell copy

   export DATABRICKS_AI_GATEWAY_PATH="/v1/chat/completions"

   Integrate with the AIML API using the new client library published as a separate npm package.
   shell copy

   npm install @goose/ai-ml-client

   • Supports GPT-6 Astra models
   • Allows Kotlin callers to configure the Databricks AI Gateway path via a new setting or environment variable
   • Integrates with AIML API

   3 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !Removes fast model routing functionality
   • !Removes managed model registry from local inference
4. v1.49.0 Sep 3, 2026 · release notes

   Goose v1.49.0 adds declarative providers (Opper, Databricks, TrustedRouter, etc.), on_failure hooks, web-search skills, ARM64ARM64The 64-bit instruction set architecture from ARM Holdings, used in modern mobile chips, Apple Silicon, and cloud server CPUs; tools target it to run natively on those platforms without emulation overhead. desktop, and --with-extension naming.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.49.0 https://github.com/aaif-goose/goose.git
   # already have the repo? check out this version:
   $ git checkout v1.49.0

   └──▷ USE IT
   Capture pre-execution failures in custom hooks to handle or log errors before they bubble up.
   yaml copy

   # Configure your hook.yaml to include an on_failure block
   cat <<EOF > ~/.config/goose/hook.yaml
   pre_tool_use:
     my_hook:
       on_failure: |
         echo "Tool failed with exit code ${{exit_code}}" >&2
   EOF

   Load a specific extension by name during the CLI run without relying on auto-discovery order.
   copy

   $ goose cli --with-extension my_custom_extension_name

   • Adds on_failure block for PreToolUse hooks
   • Web-search and browser-use built-in skills
   • Model-native audio transcription provider
   • --with-extension can now name its extension
   • OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. session_id forwarding and app category header

   +15 more show less

   • ACp thinking effort derivation and forwarding
   • GPT-5.6 follow-up support for Codex and Responses API
   • Pre-registered OAuth clients for streamable_http extensions
   • OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. request params, response metadata, tool call parity, and agent identification
   • Opper as a declarative OpenAI-compatible provider
   • Databricks Unity Catalog model services discovery
   • TrustedRouter declarative provider
   • OpenCode Zen gateway as a declarative provider
   • Gondola as declarative OpenAI-compatible provider
   • SayGM as declarative OpenAI-compatible provider
   • Lynkr as declarative OpenAI-compatible provider
   • PleumRouter declarative provider
   • Z.ai GLM-5.3 (replacing GLM-5.2)
   • OpenAI GPT-5.6 (sol/terra/luna) via Bedrock
   • Custom provider cost fields for cost tracking

   20 details in Everything

5. v1.48.0 Aug 27, 2026 · release notes

   Goose v1.48.0 adds declarative providers, PreToolUseResult hooks, /new CLI command, web-search skill, and OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. metadata.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.48.0 https://github.com/aaif-goose/goose.git
   # already have the repo? check out this version:
   $ git checkout v1.48.0

   • Adds on_failure block for PreToolUse hooks to capture tool execution outcomes
   • Introduces PreToolUseResult event with stable tool_call_id across the tool lifecycle
   • Adds /new CLI command to start a fresh session without restarting the agent
   • Expands --with-extension CLI flag to accept an extension name directly
   • Includes built-in Web-search and browser-use skills

   +16 more show less

   • Supports pre-registered OAuth clients for streamable_http extensions
   • Derives and forwards thinking effort from the ACP harness
   • Titles new sessions via _meta.sessionTitle in the ACP harness
   • Adds _goose/unstable/session/extensions/apply method to the ACP interface
   • Exposes OpenTelemetry request parameters, response metadata, and agent identification
   • Forwards session_id and app category header to OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. via telemetry
   • Replaces Z.ai GLM-5.2 with GLM-5.3 support for custom provider cost tracking
   • Adds model-native audio transcription capability to the provider system
   • Enables routing for OpenAI GPT-5.6 (sol/terra/luna) via AWS Bedrock
   • Provides GPT-5.6 follow-up support for Codex and Responses API endpoints
   • Adds Select saved recipes when creating a schedule in the Desktop UI
   • Sorts configured providers to the top of the provider list in the Desktop UI
   • Collapses scheduled job sessions into accordion view in chat history
   • Auto-focusses chat input when user starts typing in the Desktop client
   • Shows interactive git branch indicator in the chat bottom bar
   • Displays recent used models in the chat footer model picker

   21 details in Everything

6. v1.47.0 Aug 21, 2026 · release notes

   Goose v1.47 adds interactive git branch indicator, pre-registered OAuth for streamable_http, and recent model picker

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.47.0 https://github.com/aaif-goose/goose.git
   # already have the repo? check out this version:
   $ git checkout v1.47.0

   • Adds an interactive git branch indicator in the chat bottom bar
   • Supports pre-registered OAuth clients for streamable_http extensions
   • Shows recently used models in the chat footer model picker

   3 details in Everything

7. v1.46.0 Aug 12, 2026 · issue 043

   Goose v1.46.0 adds hooks with PreToolUse denial, 25+ new providers, goose review, major CLI additions, and broad env-var configurability.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.46.0 https://github.com/aaif-goose/goose.git
   # already have the repo? check out this version:
   $ git checkout v1.46.0

   └──▷ TRY IT
   Block a specific tool from executing during a session by wiring a PreToolUse hook that denies it — useful for enforcing policy on sensitive operations.
   copy

   $ GOOSE_MAX_TOOL_RESPONSE_SIZE=65536 goose session

   Pin a stable OAuth callback port so OAuth redirect URIs stay consistent across restarts, useful in environments with strict firewall rules.
   copy

   $ GOOSE_OAUTH_CALLBACK_PORT=9876 goose session

   Run a local code review against the current working directory, delegating checks via the default orchestrator.
   copy

   $ goose review

   Set the transcription language for local Whisper so voice input is recognized correctly without auto-detection overhead.
   copy

   $ LOCAL_WHISPER_LANGUAGE=fr goose session

   • Adds LOCAL_WHISPER_LANGUAGE environment variable to configure language for multilingual local voice transcription.
   • Adds GOOSE_MAX_TOOL_RESPONSE_SIZE environment variable to configure the maximum size of tool output returned to the agent.
   • Adds MAX_CODE_BLOCK_LINES environment variable to cap the number of lines rendered in code blocks.
   • Adds GOOSE_FAST_MODEL environment variable (consumed by ModelConfig::with_fast) to designate a fast model for lightweight tasks.
   • Adds GOOSE_DOCS_ROOT environment variable to redirect documentation lookups for air-gapped deployments.

   +41 more show less

   • Adds GOOSE_OAUTH_CALLBACK_PORT environment variable to pin a stable OAuth redirect_uri port across sessions.
   • Adds AZURE_OPENAI_AD_TOKEN environment variable to support Azure AD (Entra IDEntra IDMicrosoft's cloud-based identity and access management service, formerly Azure Active Directory, used by cyber tools to authenticate users and enforce access policies in Microsoft 365 and Azure environments.) bearer-token authentication for the Azure OpenAI provider.
   • Adds _meta.hidden field to the ACP session/new request to create hidden (background) sessions.
   • Adds Hooks framework with PreToolUse denial, letting operators intercept and block tool calls before execution.
   • Adds --edit flag to the session command to edit conversation history before forking a session.
   • Adds /goal slash command for agent self-evaluation before a session concludes.
   • Adds /status slash command to report current session state.
   • Adds /model slash command to switch the active model mid-session.
   • Adds goose review command for local code review workflows.
   • Adds tab completion and provider switching to the model command.
   • Loads global agent hints from ~/.agents/AGENTS.md automatically each session.
   • Adds Markdown as a format option for session export.
   • Surfaces per-message token count, cost, time-to-first-tokentime-to-first-tokenTime to first token, the wait between submitting a request and receiving the first generated token. It includes delays such as queuing and processing the prompt. (TTFTTTFTTime to first token, the wait between submitting a request and receiving the first generated token. It includes delays such as queuing and processing the prompt.), and tokens-per-second stats in the UI.
   • Tracks cache tokens separately for accurate cost reporting.
   • Streams shell command output to the UI in real time as commands execute.
   • Adds TLS support for acp serve.
   • Adds a resizable sidebar with a drag handle and persisted width in the desktop UI.
   • Adds chat history search in the navigation panel.
   • Adds a search filter to the provider selection grid.
   • Adds a TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. diff viewer.
   • Emits GenAI semantic convention attributes in OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. (OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor.) spans.
   • Forwards images and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. embedded-resource blobs in Anthropic and Google provider formats.
   • Adds logo_uri to OAuth client ID metadata documents (CIMDCIMDA text-based Short Message Service protocol from Nokia for exchanging SMS messages between applications and SMS centers, used to build tools that send, receive, or test carrier messaging flows.) so authorization servers can display the goose logo on consent screens.
   • Adds proactive OAuth token refresh to avoid re-authentication on every session.
   • Adds goose://new-session and goose://resume deep links, including support for an initial prompt via goose://new-session.
   • Adds encrypted Nostr session sharing and session import.
   • Adds Hugging Face OAuth support with a dedicated auth tab in settings.
   • Supports MLXMLXApple's machine-learning framework for Apple silicon, which uses the unified memory shared by CPU and GPU to run and fine-tune models on a Mac. models in the local inference provider.
   • Adds Linux Vulkan support for local inference.
   • Adds a UniFFI SDK for cross-language bindings, Python wheel publishing, and Kotlin FFI with Maven publish.
   • Adds a quarterly recurrence option to the session scheduler.
   • Adds Termux detection in the installer to automatically select the musl portable build.
   • Binds MCP apps to trusted ownership metadata to restrict which extensions can be loaded.
   • Adds a CLI command to list available skills with their token counts.
   • Adds an image read tool for ingesting images directly in sessions.
   • Adds Summon extension peek mode for async background tasks, plus context and working_dir parameters for delegate tasks.
   • Adds new providers: Celeris, Friendli, Azure AI Foundry (multi-LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code.), OllamaCloud (dynamic model discovery), Sakana AI (Fugu API), iFlytek Spark, Astron MaaS, Fireworks AI, Together AI, OrcaRouter, Perplexity, Alibaba/Qwen (DashScope), Databricks AI Gateway, Scaleway, NEAR AI Cloud, EmpirioLabs, xAI SuperGrok (OAuth subscription), Vercel AI Gateway, Routstr, FuturMix, oMLX, Atomic Chat, Muse Spark 1.1 (Meta Models API), and AWS Bedrock support for OpenAI GPT models via mantle endpoint.
   • Adds support for OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. request parameters.
   • Adds new models: GPT-5.6, GPT-5.5, GLM-5.2, Opus5 (with adaptive thinking), claude-sonnet-5, claude-fable-5, Cursor composer-2.5, Cursor-agent (fetched from CLI), latest Gemini models, Gemini 3.x, MiniMax-M3/M2.7, and OVHcloud Qwen3.6-27B.
   • Adds desktop localizations for French, German, Italian, Portuguese, Indonesian, Malay, Vietnamese, zh-TW, Korean, Spanish, Japanese, Hindi, Russian, Turkish, Simplified Chinese (zh-CN), and language selection in the UI.
   • Adds structured summary output with template rendering for context compactioncontext compactionCondensing an AI conversation or working history to free space in the model's context window. A summary preserves useful information, but details can be lost..

   46 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !The goose project command (and its p alias) has been removed; CLI project support is no longer available.
8. docs update Jul 30, 2026 · issue 030

   Goose adds offline documentation skill and Azure AI Foundry integration for OpenAI, Anthropic, and partner models.

   • New goose-doc-guide skill reads official Goose documentation before answering, enabling offline/air-gapped doc lookups.
   • Adds Azure AI Foundry integration, supporting OpenAI, Anthropic, and partner model deployments from an Azure AI Foundry project.

   2 details in Everything

9. v1.45.0 Jul 29, 2026 · issue 030

   Goose v1.45.0 adds tab completion for model switching, air-gapped docs support, disabling built-in skills, and Gemini/Opus5 model support.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.45.0 https://github.com/aaif-goose/goose.git
   # already have the repo? check out this version:
   $ git checkout v1.45.0

   └──▷ TRY IT
   Point Goose at a local mirror of the docs in an air-gapped environment so help content loads without internet access.
   copy

   $ GOOSE_DOCS_ROOT=/mnt/internal/goose-docs goose session

   Run the ACP server with scheduled recipe execution enabled (opt-in after the new default disables it).
   copy

   $ goose server --enable-scheduler

   • Adds tab completion and inline provider switching to the model command.
   • Supports latest Gemini models and Opus5 (including adaptive thinking).
   • Adds GOOSE_DOCS_ROOT environment variable for air-gapped / offline documentation access.
   • Gates tool-call label enrichment on ACP client capability, improving ACP interoperability.
   • Adds stable identity for agent event messages, enabling reliable message tracking across sessions.

   5 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !The ACP scheduler is now disabled by default; deployments relying on scheduled recipe execution must explicitly pass --enable-scheduler.
10. docs update Jul 24, 2026 · issue 024

    Goose adds rp-why skill: an AI collaboration coach that scores your tool sophistication and prompt depth.

    └──▷ TRY IT
    Install the rp-why skill and initialize your baseline before your first analysis.
    copy

    $ npx skills add https://github.com/block/agent-skills --skill rp-why
    /rp-why init

    At the end of a Goose session, check whether you stretched or coasted compared to your typical patterns.
    copy

    $ /rp-why current

    On Friday, compare the week's sessions against your baseline to identify next week's focus areas.
    copy

    $ /rp-why compare

    • New rp-why skill analyzes conversation history to score sessions on Gas Town Stage (tool sophistication, stages 1–8) and DOK Level (prompt cognitive depth, levels 1–4).
    • New /rp-why init command establishes a personal baseline by analyzing available conversation history (~30 seconds).
    • New /rp-why current command shows how the active session compares to your typical usage patterns.
    • New /rp-why compare command diffs the current session against your baseline to surface progress or regression.
    • Session output includes a quadrant map (Underutilizing / Frontier / Learning Zone / Thinking Ahead), DOK distribution histogram, and actionable growth nudges.

    5 details in Everything

11. docs update Jul 24, 2026 · docs issue 004

    Goose adds rp-why skill: an AI collaboration coach that scores your tool sophistication and prompt depth.

    └──▷ TRY IT
    Install the rp-why skill and initialize your baseline before your first analysis.
    copy

    $ npx skills add https://github.com/block/agent-skills --skill rp-why
    /rp-why init

    At the end of a Goose session, check whether you stretched or coasted compared to your typical patterns.
    copy

    $ /rp-why current

    On Friday, compare the week's sessions against your baseline to identify next week's focus areas.
    copy

    $ /rp-why compare

    • New rp-why skill analyzes conversation history to score sessions on Gas Town Stage (tool sophistication, stages 1–8) and DOK Level (prompt cognitive depth, levels 1–4).
    • New /rp-why init command establishes a personal baseline by analyzing available conversation history (~30 seconds).
    • New /rp-why current command shows how the active session compares to your typical usage patterns.
    • New /rp-why compare command diffs the current session against your baseline to surface progress or regression.
    • Session output includes a quadrant map (Underutilizing / Frontier / Learning Zone / Thinking Ahead), DOK distribution histogram, and actionable growth nudges.

    5 details in Everything

12. v1.44.0 Jul 23, 2026 · issue 024

    Goose v1.44.0 adds new model support (GPT-5.6, Gemini, Muse Spark 1.1), new providers, --edit session flag, per-message usage stats, and OllamaCloud integration.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.44.0 https://github.com/aaif-goose/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.44.0

    └──▷ TRY IT
    Edit a prior session's conversation before forking it — useful for pruning context or replaying with adjusted prompts.
    copy

    $ goose session --edit

    • Adds support for latest Gemini models.
    • Adds Muse Spark 1.1 support via the Meta Models API.
    • Adds MiniMax-M3 and missing M2.7 model variants.
    • Adds declarative Sakana AI provider for the OpenAI-compatible Fugu API.
    • Adds iFlytek Spark and Astron MaaS providers.

    +10 more show less

    • Adds Fireworks AI as a declarative provider.
    • Adds OllamaCloudProvider with dynamic model discovery and context limits.
    • Adds OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. request parameters support.
    • New --edit session flag to edit a conversation before forking it.
    • New per-message usage stats UI showing tokens, cost, TTFTTTFTTime to first token, the wait between submitting a request and receiving the first generated token. It includes delays such as queuing and processing the prompt., and tokens/second.
    • Chunks command-classifier input with overlapping windows for improved security robustness.
    • Adds search filter for the provider grid in the UI.
    • Adds delete support for custom apps from the Apps UI.
    • Reconnects desktop ACP sessions automatically after sleep or connection loss.
    • Adds French, German, Italian, Portuguese, Indonesian, Malay, Vietnamese, and zh-TW desktop locales.

    15 details in Everything

13. v1.43.0 Jul 14, 2026 · issue 015

    Goose v1.43.0 adds per-message cost/token stats, OllamaCloud provider, --edit session flag, TLS for ACP serve, and six new model providers.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.43.0 https://github.com/aaif-goose/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.43.0

    └──▷ TRY IT
    Review and trim a session's conversation before branching it into a new task, avoiding stale context.
    copy

    $ goose session --edit

    • Adds --edit session flag to edit conversation history before forking a session.
    • New per-message usage stats UI showing tokens, cost, TTFTTTFTTime to first token, the wait between submitting a request and receiving the first generated token. It includes delays such as queuing and processing the prompt., and tok/s for each turn.
    • New OllamaCloudProvider with dynamic model discovery and automatic context-limit detection.
    • Adds iFlytek Spark and Astron MaaS as supported providers.
    • Adds Fireworks AI as a declarative provider.

    +12 more show less

    • Adds OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. request parameters support.
    • Adds OVHcloud Qwen3.6-27B model with 262K context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it..
    • Adds GPT-5.6 family support (gpt-5.6-luna, gpt-5.6-sol, gpt-5.6-terra) with 1,050,000-token context and xhigh reasoning effort.
    • Adds GLM-5.2 model support.
    • Adds TLS support for goose serve ACP sessions.
    • Adds streamable HTTP support in the deeplink generator.
    • Desktop sessions now automatically reconnect after system sleep or connection loss.
    • Chat sessions are now grouped by project in the navigation panel.
    • Delete support for custom apps added to the Apps UI.
    • Desktop localizations added for French, German, Italian, Portuguese, Indonesian, Malay, Vietnamese, and Traditional Chinese.
    • Provider bindings MVP added to goose-sdk with Python wheel publishing.
    • Diagnostics output migrated to structured JSON report format.

    17 details in Everything

14. v1.42.0 Jul 13, 2026 · issue 015

    Goose v1.42.0 adds Ollama Cloud provider, per-message token/cost stats UI, project-grouped sessions, and Python SDK provider bindings.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.42.0 https://github.com/aaif-goose/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.42.0

    • Adds OllamaCloudProvider with dynamic model discovery and automatic context-limit detection.
    • Adds per-message usage stats in the desktop UI: token counts, cost, time-to-first-tokentime-to-first-tokenTime to first token, the wait between submitting a request and receiving the first generated token. It includes delays such as queuing and processing the prompt. (TTFTTTFTTime to first token, the wait between submitting a request and receiving the first generated token. It includes delays such as queuing and processing the prompt.), and tokens/sec.
    • Adds per-message usage and cost tracking with derived session totals.
    • Groups chat sessions by project in the navigation panel.
    • Shows project names in the directory-switcher dropdown.

    +4 more show less

    • Adds provider bindings MVP to goose-sdk with Python wheel publishing.
    • Adds AskAI bot hints for follow-up suggestions.
    • Hardens goose update to fail closed when Sigstore/SLSASLSASupply-chain Levels for Software Artifacts, a framework for assessing how software was built and protecting that process from tampering. Provenance records help verify where a build came from. provenanceprovenanceEvidence recording where a software artifact came from and how it was built, including its source and build process. It helps a consumer check that the artifact matches a trusted build. cannot be verified.
    • Increases the /mcp-app-guest body limit to support larger single-file MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. apps.

    9 details in Everything

15. v1.41.0 Jul 3, 2026 · issue 003

    Goose v1.41.0 adds TLS for ACP, 15+ new providers, hooks, summon sub-agents, and a wave of new CLI commands.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.41.0 https://github.com/aaif-goose/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.41.0

    └──▷ TRY IT
    Serve Goose's ACP endpoint over TLS so remote clients can connect securely.
    copy

    $ goose serve --tls --tls-cert-path /etc/goose/server.crt --tls-key-path /etc/goose/server.key --port 3284

    Edit a previous session's conversation context before forking it, useful for re-running with corrected instructions.
    copy

    $ goose session --edit --resume <session-id>

    Cap tool response payload size to prevent context bloat when tools return large outputs (e.g., grep over a huge codebase).
    copy

    $ GOOSE_MAX_TOOL_RESPONSE_SIZE=16384 goose run --recipe my-recipe.yaml

    • Adds --edit session flag to edit a conversation before forking it into a new session.
    • New tui subcommand on goose-cli to launch the terminal UI directly.
    • Supports TLS for goose serve via --tls, --tls-cert-path, and --tls-key-path flags.
    • Adds iFlytek Spark, Astron MaaS, Fireworks AI, Together AI, OrcaRouter, EmpirioLabs AI, Perplexity, Alibaba (Qwen/DashScope), Scaleway, NEAR AI Cloud, Vercel AI Gateway, Databricks AI Gateway, Routstr, FuturMix, oMLX, Atomic Chat, and xAI SuperGrok OAuth as new providers.
    • Adds OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. request parameters support for fine-grained API payload control.

    +22 more show less

    • OpenAI Responses API store param now configurable via OPENAI_STORE env var or config.
    • MLXMLXApple's machine-learning framework for Apple silicon, which uses the unified memory shared by CPU and GPU to run and fine-tune models on a Mac. model support added for the local inference provider.
    • GLM-5.2 model and gpt-5.5 (ChatGPT Codex) model support added.
    • Summon gains peek mode for async background tasks, context and working_dir parameters for delegate tasks, subagent instructions, and structured metadata for task load results.
    • Session import functionality added.
    • OTLPOTLPOpenTelemetry Protocol, a standard way to send traces, metrics and logs between applications, collectors and monitoring systems. It lets telemetry move across compatible tools. logging schema added for cross-tool detection.
    • Worktree-aware directory switcher added.
    • Chat history search feature added.
    • TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. diff viewer added.
    • Paginated session list in ACP.
    • ACP schema now exposes available tools.
    • ACP thinking effort config option added.
    • Structured per-provider config block with non-destructive provider switching.
    • Language selection feature in desktop app with locales for French, German, Italian, Portuguese, Indonesian, Malay, Vietnamese, Traditional Chinese, Spanish, Korean, Japanese, Hindi, Russian, Turkish, and Simplified Chinese.
    • Stable OAuth redirect URI via GOOSE_OAUTH_CALLBACK_PORT environment variable.
    • Unified thinking effort control across all providers.
    • Image read tool added.
    • Open-plugins generalization plus skills support.
    • Option to disable automatic update downloads.
    • Last message snippets for ACP sessions.
    • GOOSE_FAST_MODEL environment variable support for specifying a fast model.
    • Optional api_key configuration for declarative OpenAI-engine providers.

    27 details in Everything

16. v1.40.0 Jul 2, 2026 · issue 002

    Goose v1.40.0 adds 13+ new AI providers, hooks system with tool-use denial, local code review, new CLI slash commands, and prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. mitigation.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.40.0 https://github.com/aaif-goose/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.40.0

    └──▷ TRY IT
    Cap how much output a noisy tool can return so long responses don't blow the context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it..
    copy

    $ GOOSE_MAX_TOOL_RESPONSE_SIZE=32768 goose run --recipe audit.yaml

    Run an AI-powered local code review on the current working directory before opening a pull request.
    copy

    $ goose review

    • New /status slash command in the CLI to check session state at a glance.
    • New goose review subcommand for running local AI-powered code review.
    • Adds a hooks system for extensibility, including PreToolUse denial to block tool calls before execution and blocking Stop hook decisions.
    • Prompt injection mitigation enabled by default.
    • Supports 13+ new model providers: Together AI, EmpirioLabs, OrcaRouter, NEAR AI Cloud, Perplexity, Alibaba/Qwen via DashScope, Databricks AI Gateway, Scaleway, Vercel AI Gateway, xAI SuperGrok OAuth, OpenAI GPT via AWS Bedrock mantle endpoint, GLM-5.2, and MLXMLXApple's machine-learning framework for Apple silicon, which uses the unified memory shared by CPU and GPU to run and fine-tune models on a Mac. models for local inference.

    +11 more show less

    • New GOOSE_FAST_MODEL environment variable to designate a fast model for lightweight operations.
    • Unified thinking effort control and canonical thinking modes across all supported providers.
    • Cache token tracking for accurate per-session cost reporting.
    • Encrypted Nostr session sharing.
    • Peek mode for async background tasks in summon, plus context parameter and working_dir override for delegate tasks.
    • New TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. subcommand (goose-cli tui) with built-in diff viewer.
    • Chat history search in the desktop navigation panel.
    • Session import and deep links (goose://new-session, goose://resume) including initial prompt support.
    • Image read tool for ingesting local images.
    • Streamable HTTPStreamable HTTPAn MCP transport protocol that lets a client send JSON-RPC requests over HTTP POST while the server streams responses back using Server-Sent Events, replacing the older stdio and SSE transports for remote tool communication. support in the deeplink generator.
    • Linux musl CLI builds for broader Linux compatibility.

    16 details in Everything

17. v1.39.0 Jun 25, 2026 · issue 001

    Goose v1.39.0 adds /status and /model CLI commands, MLXMLXApple's machine-learning framework for Apple silicon, which uses the unified memory shared by CPU and GPU to run and fine-tune models on a Mac. local inference, 8 new providers, Azure Entra IDEntra IDMicrosoft's cloud-based identity and access management service, formerly Azure Active Directory, used by cyber tools to authenticate users and enforce access policies in Microsoft 365 and Azure environments. auth, and a unified OTLPOTLPOpenTelemetry Protocol, a standard way to send traces, metrics and logs between applications, collectors and monitoring systems. It lets telemetry move across compatible tools. logging schema.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.39.0 https://github.com/aaif-goose/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.39.0

    └──▷ TRY IT
    Check session status interactively without leaving a running CLI session.
    copy

    $ /status

    Switch the active model mid-session to compare responses without restarting.
    copy

    $ /model

    Run a recipe and print token/generation statistics on completion to benchmark cost and speed.
    copy

    $ goose run --recipe goose-self-test.yaml --stats

    • Adds /status slash command to the CLI for quick session status reporting.
    • Adds /model slash command to the CLI for switching models mid-session without restarting.
    • Adds --stats flag to print generation statistics after a run completes.
    • Supports MLX models for local inference, enabling Apple Silicon on-device model execution.
    • Supports OpenAI GPT models via AWS Bedrock mantle endpoint.

    +23 more show less

    • Adds Azure Entra ID bearer tokenbearer tokenA credential that grants access to whoever presents it, without a separate proof of possession. Protecting the token matters because a stolen copy can be used by someone else. authentication via AZURE_OPENAI_AD_TOKEN environment variable.
    • Adds EmpirioLabs, OrcaRouter, Together AI, Perplexity, Alibaba (Qwen/DashScope), Databricks AI Gateway, and xAI SuperGrok as new declarative providers.
    • Adds HuggingFace OAuth support with a dedicated auth tab in settings.
    • Introduces GOOSE_SERVER__SECRET_KEY support at the goose serve ACP endpoint for API authentication.
    • Adds configurable tool output size limit via GOOSE_MAX_TOOL_RESPONSE_SIZE environment variable.
    • Adds configurable per-agent timeout for Open Model Gym via GYM_AGENT_TIMEOUT environment variable.
    • Loads global agent hints automatically from ~/.agents/AGENTS.md.
    • Adds an image read tool for processing image files directly in sessions.
    • Adds peek mode for async background tasks in summon, enabling non-blocking status checks.
    • Adds context and working_dir override parameters for summon delegate tasks.
    • Adds structured metadata to summon task load results for richer programmatic consumption.
    • Adds a quarterly scheduling option to the task scheduler.
    • Adds a unified OTLP logging schema for cross-tool detection and observability.
    • Enables canonical thinking modes configuration and ACP thinking effort config option.
    • Enables prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. mitigation by default for internal users.
    • Adds _meta.conversationBefore for ACP session forking.
    • Adds ACP session search, list with keyword/type filter, session import, and system prompt setter in the Desktop UI.
    • Adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. extensions support in open plugins.
    • Enables goose://new-session deep link for pre-populated session launch.
    • Adds Korean, Japanese, Hindi, Spanish, Russian, and Turkish localization support.
    • Adds CLI command to list skills with token counts.
    • Surfaces Anthropic stream refusals as visible errors instead of silent failures.
    • Tracks cache tokens for accurate cost reporting.

    28 details in Everything

18. v1.38.0 Jun 17, 2026 · issue 001

    Goose v1.38.0 adds goose review, hooks, 8 new providers, ACP secret-key auth, egress directionality, and a PreToolUse denial hook.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.38.0 https://github.com/aaif-goose/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.38.0

    └──▷ TRY IT
    Run an AI code review on the current repo, capping each check subprocess to 10 turns to prevent runaway agents.
    copy

    $ goose review --turn-limit 10

    Start the ACP server with a secret key so only authenticated clients can connect.
    copy

    $ GOOSE_SERVER__SECRET_KEY=mysecretkey goose serve

    Prevent context bloat from large tool responses by capping output size, and disable noisy per-call summary cards.
    copy

    $ GOOSE_MAX_TOOL_RESPONSE_SIZE=8192 GOOSE_DISABLE_TOOL_CALL_SUMMARY=true goose run --recipe my-recipe.yaml

    • Adds goose review command for local AI-powered code review with per-check turn limits and orchestrated subprocesses.
    • New hooks system with PreToolUse denial hook to intercept and block tool calls before execution.
    • Adds GOOSE_SERVER__SECRET_KEY support at the goose serve ACP endpoint for authenticated API access.
    • Adds egress logging directionality (outbound/inbound/unknown) to security tracing events, enabling data-exfiltration vs. data-fetch distinction.
    • Unified OTLPOTLPOpenTelemetry Protocol, a standard way to send traces, metrics and logs between applications, collectors and monitoring systems. It lets telemetry move across compatible tools. logging schema for cross-tool detection across security monitoring pipelines.

    +27 more show less

    • New /model slash command in CLI sessions for switching models mid-conversation without restarting.
    • New /goal slash command for agent self-evaluation before finishing a task.
    • Adds GOOSE_MAX_TOOL_RESPONSE_SIZE env var to cap tool output size and prevent context bloat.
    • Adds GOOSE_DISABLE_TOOL_CALL_SUMMARY env var to opt out of per-tool-call summary cards.
    • Adds GOOSE_OAUTH_CALLBACK_PORT env var for a stable OAuth redirect URI across sessions.
    • New image read tool enabling agents to process and reason over image files.
    • Session import now supports Claude Code, Codex, and Pi .jsonl.jsonlA text format where each line is a self-contained, valid JSON object, making it easy to stream, append, and process large datasets line by line without parsing a full JSON array into memory. transcript formats in addition to Goose exports.
    • Encrypted Nostr session sharing for secure cross-device session transfer.
    • Supports 8 new providers: Perplexity, Alibaba/Qwen via DashScope, Databricks AI Gateway, NEAR AI Cloud, Scaleway, Routstr, Vercel AI Gateway, and xAI SuperGrok OAuth.
    • Adds Hugging Face OAuth support with a new Settings > Auth tab for credential management.
    • Canonical thinking modes available across all supported providers.
    • Steering messages via ACP enable mid-session behavioral guidance from ACP clients.
    • ACP session management (list, search, filter by keyword/type, paginate, get session info) now available in Desktop and via API.
    • New Scholar Sidekick MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. extension for research workflows.
    • Linux musl CLI builds for glibc-free deployments.
    • Windows CUDACUDANVIDIA's platform and programming interface for running general computation on its GPUs. Nearly every AI training and inference stack is built on it first. release artifacts for GPUGPUA graphics processing unit, a processor that performs many calculations in parallel. AI workloads use GPUs to speed up model training and inference.-accelerated local inference.
    • Nushell terminal and tab-completion support.
    • Quarterly scheduling option for recurring recipes.
    • New goose://resume and goose://new-session deep links for Desktop session navigation.
    • Projects as backend sources with system prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. for scoped agent contexts.
    • Opt-in GOOSE_FAST_MODEL env var honored in ModelConfig::with_fast for fast-model overrides.
    • Harbor eval runner for structured agent benchmarking.
    • CLI command to list skills with token counts for context budget planning.
    • TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. command (goose tui) and TUI diff viewer added to goose-cli.
    • Configurable MAX_CODE_BLOCK_LINES via env var to control output verbosity.
    • Minimal UniFFI setup enabling cross-language SDK bindings.
    • Simplified Chinese (zh-CN) UI translation.

    32 details in Everything

19. v1.37.0 Jun 3, 2026 · issue 001

    Goose v1.37.0 adds 9+ new providers, PreToolUse hooks, goose review, /model switching, skills listing, and encrypted session sharing.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.37.0 https://github.com/aaif-goose/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.37.0

    └──▷ TRY IT
    Run an interactive code review on your current working directory to get AI feedback before opening a PR.
    copy

    $ goose review

    Switch to a different model mid-session without losing context — useful when a task needs a more capable or faster model.
    copy

    $ /model claude-opus-4-5

    Audit token cost of installed skills to understand context budget usage before starting a long session.
    copy

    $ goose skills list

    Cap how much output a single tool call can return to prevent context overflow from verbose tools.
    copy

    $ GOOSE_MAX_TOOL_RESPONSE_SIZE=50000 goose run --recipe my-recipe.yaml

    • Adds goose review command for local code review directly from the CLI.
    • Adds /model slash command to switch the active model mid-session without restarting.
    • Adds goose skills list subcommand to display all installed skills with token counts.
    • Adds PreToolUse denial hooks so practitioners can intercept and block tool calls before execution.
    • Adds a Hooks system for event-driven automation around agent actions.

    +32 more show less

    • Adds /goal slash command for agent self-evaluation before finishing a task.
    • Adds configurable tool output size limit via GOOSE_MAX_TOOL_RESPONSE_SIZE environment variable.
    • Adds encrypted Nostr session sharing for secure cross-device session handoff.
    • Adds xAI SuperGrok OAuth subscription provider.
    • Adds Databricks AI Gateway (databricks_v2) provider with routing across OpenAI Responses, Anthropic Messages, and MLflow chat completions.
    • Adds Perplexity as a declarative OpenAI-compatible provider.
    • Adds Alibaba (Qwen via DashScope) declarative provider.
    • Adds NEAR AI Cloud, Scaleway, Routstr, FuturMix, oMLX, Atomic Chat, and Vercel AI Gateway as new declarative providers.
    • Adds Scholar Sidekick MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. extension.
    • Adds GitHub Copilot /responses API support.
    • Adds Linux Vulkan support for local inference.
    • Adds goose://resume and goose://new-session deep links for desktop session management.
    • Adds quarterly scheduling option for the recipe scheduler.
    • Adds optional --parameters flag to scheduled recipes.
    • Adds ACP session system prompt setter for programmatic prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools..
    • Adds ACP session list pagination.
    • Exposes raw provider supported models over ACP.
    • Adds GOOSE_FAST_MODEL env var to configure the fast model in ModelConfig::with_fast.
    • Adds configurable MAX_CODE_BLOCK_LINES via environment variable.
    • Adds GOOSE_OAUTH_CALLBACK_PORT env var for a stable OAuth redirect URI.
    • Adds proactive OAuth token refresh to avoid re-authentication on every session.
    • Adds TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. command (goose tui) gated behind a tui feature flag.
    • Adds TUI diff viewer for reviewing file changes in the terminal UI.
    • Adds Nushell terminal and shell completion support.
    • Adds Russian language and Turkish desktop locale support.
    • Adds Simplified Chinese (zh-CN) translation.
    • Adds Harbor eval runner for benchmarking.
    • Adds summon subagent instructions for multi-agent orchestration.
    • Adds slash commands (built-in, skill, recipe) in the ACP server.
    • Adds projects as backend sources with system prompt injection.
    • Groups consecutive tool calls into one summarized chain card in the UI.
    • Strips chain-of-thoughtchain-of-thoughtIntermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens. markers from custom provider output automatically.

    37 details in Everything

20. v1.36.0 May 27, 2026 · release notes

    Goose v1.36.0 adds goose review, a hooks system for PreToolUse denial, TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. diff viewer, /goal self-evaluation, and 7+ new AI providers.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.36.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.36.0

    └──▷ TRY IT
    Run a local code review on your current working directory before opening a PR.
    copy

    $ goose review

    Suppress verbose tool call summaries in CI pipelines where only final output matters.
    copy

    $ GOOSE_DISABLE_TOOL_CALL_SUMMARY=1 goose run --recipe my-recipe.yaml

    • New goose review command for local code review without leaving the terminal.
    • New TUI diff viewer to inspect file changes interactively in the terminal UI.
    • New /goal slash command lets the agent self-evaluate against a stated goal before finishing a session.
    • New hooks system supports PreToolUse denial and extensibility, enabling policy enforcement on tool calls.
    • Adds NEAR AI Cloud, Scaleway, Vercel AI Gateway, Atomic Chat, Routstr, FuturMix, and oMLX as new AI providers.

    +23 more show less

    • Slash commands (built-in, skill, recipe) now available in the ACP server.
    • Summon subagent instructions enable structured multi-agent delegation.
    • Unified thinking effort control across all providers via a single configuration surface.
    • New goose://resume and goose://new-session deep links for session management from external apps.
    • New --parameters flag for scheduled recipes to pass runtime parameters.
    • Quarterly scheduling option added to the recipe scheduler.
    • GitHub Copilot /responses API support added.
    • Chat history search feature added to the desktop UI.
    • Projects can now act as backend sources with system prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools..
    • Linux Vulkan support enables GPUGPUA graphics processing unit, a processor that performs many calculations in parallel. AI workloads use GPUs to speed up model training and inference.-accelerated local inference on Linux.
    • Linux musl CLI builds available for Alpine/musl-based environments.
    • Windows CUDACUDANVIDIA's platform and programming interface for running general computation on its GPUs. Nearly every AI training and inference stack is built on it first. release artifacts now published for GPU-accelerated local inference on Windows.
    • New GOOSE_DISABLE_TOOL_CALL_SUMMARY env var to suppress tool call summaries.
    • New GOOSE_OAUTH_CALLBACK_PORT env var for a stable OAuth redirect URI across sessions.
    • Proactive OAuth token refresh avoids re-authentication on every new session.
    • New MAX_CODE_BLOCK_LINES env var controls code block rendering length.
    • Honors GOOSE_FAST_MODEL env var in ModelConfig for fast-model selection.
    • GPT-5.5 added to known models list.
    • ACP session list now supports pagination.
    • Encrypted Nostr session sharing for secure cross-client session handoff.
    • Nushell terminal and shell completion support added.
    • Worktree-aware directory switcher for multi-worktree Git workflows.
    • Harbor eval runner integration added.

    28 details in Everything

21. v1.35.0 May 22, 2026 · release notes

    Goose v1.35.0 adds a hooks system for pre/post tool execution control, goose review, /goal self-evaluation, and five new providers.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.35.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.35.0

    └──▷ TRY IT
    Run a local code review on your current working directory to catch issues before committing.
    copy

    $ goose review

    • Adds a hooks system for extensible pre/post tool execution, including PreToolUse denial hooks to block tool calls before they run.
    • New goose review command for local code review directly from the CLI.
    • New /goal slash command lets the agent self-evaluate against its goal before finishing a session.
    • Diff viewer added to the TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. for inspecting code changes inline.
    • Unified thinking effort control across all providers for consistent reasoning depth.

    +27 more show less

    • Adds GOOSE_DISABLE_TOOL_CALL_SUMMARY environment variable to opt out of per-tool-call summaries.
    • Groups consecutive tool calls into a single summarized chain card in the UI.
    • Supports --parameters flag for scheduled recipes.
    • Adds quarterly scheduling option for recipes.
    • New goose://new-session deep link opens a fresh chat window.
    • Adds GOOSE_OAUTH_CALLBACK_PORT environment variable for a stable OAuth redirect URI.
    • Chat history search added to the navigation panel.
    • Encrypted Nostr session sharing for secure collaboration.
    • Nushell terminal and shell completion support.
    • Summon subagent instructions for directing spawned subagents.
    • Open-plugins generalization with skills support.
    • Projects usable as backend sources with system prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools..
    • Worktree-aware directory switcher in the TUI.
    • Slash commands (built-in, skill, recipe) now available in the ACP server.
    • Paginated session list in ACP.
    • Recipe discovery and execution in the ACP server.
    • Agents CRUDCRUDThe four basic database operations — Create, Read, Update, Delete — used as shorthand for any system that manages persistent data records, giving cyber tools a standard model for interacting with stored information. operations via ACP.
    • ACP streamable HTTP spec compliance.
    • New Vercel AI Gateway provider with opt-in leaderboard participation.
    • New GitHub Copilot /responses API support.
    • GPT-5.5 added to known models.
    • Linux musl CLI builds now available.
    • Linux Vulkan support for local inference.
    • Windows CUDACUDANVIDIA's platform and programming interface for running general computation on its GPUs. Nearly every AI training and inference stack is built on it first. release artifacts.
    • New providers: Atomic Chat, Routstr, FuturMix, oMLX, SaladCloud AI Gateway.
    • Simplified Chinese (zh-CN) translation added.
    • Strip chain-of-thoughtchain-of-thoughtIntermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens. markers from custom provider output.

    32 details in Everything

22. v1.34.0 May 13, 2026 · release notes

    Goose v1.34.0 adds hooks, agent CRUDCRUDThe four basic database operations — Create, Read, Update, Delete — used as shorthand for any system that manages persistent data records, giving cyber tools a standard model for interacting with stored information., project-based system prompts, a TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. binary, and Linux Vulkan local inference.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.34.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.34.0

    • Adds hooks support for customizable agent behavior at runtime.
    • Adds CRUD operations for managing agents via ACP.
    • Enables projects as backend sources with system prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. into sessions.
    • Supports auto-updating for plugins installed to ~/.agents/plugins.
    • Adds Linux Vulkan support for local inference workloads.

    +12 more show less

    • Achieves ACP streamable HTTP spec compliance.
    • Exposes built-in skills through the ACP sources list.
    • Introduces a provider-first onboarding flow for new users.
    • Introduces configurable Ollama host in goose2.
    • Adds Google model inventory refresh support.
    • Adds skills picker directly in the chat composer.
    • Supports mergeable configs for layered configuration management.
    • Adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Apps payload hydration and replay in goose2.
    • Translates ACP host capabilities into MCP initialization for MCP Apps.
    • Releases Windows CUDACUDANVIDIA's platform and programming interface for running general computation on its GPUs. Nearly every AI training and inference stack is built on it first. build artifacts for local inference on Windows.
    • Introduces the goose-tui binary as a standalone terminal UI.
    • Groups consecutive tool calls into a single summarized chain card in the UI.

    17 details in Everything

23. v1.32.0 Apr 23, 2026 · release notes

    Goose v1.32.0 adds Exa AI search, two new AI providers, @agent mentions, /skills and /edit commands, voice dictation, and Unix socket MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. transport.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.32.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.32.0

    • Adds Exa AI-powered web search as a built-in tool.
    • New /skills command to browse and manage available agent skills from the CLI.
    • New /edit command for on-demand prompt editing mid-session.
    • Supports @agent mentions in chat to target specific agents directly.
    • Adds Novita AI as a declarative provider.

    +12 more show less

    • Adds Kimi Code as a provider with OAuth device-flow authentication.
    • Adds Cash App extension to the Extensions Library.
    • Adds Unix domain socket transport support for StreamableHttp MCP connections.
    • Client ID Metadata Document (CIMDCIMDA text-based Short Message Service protocol from Nokia for exchanging SMS messages between applications and SMS centers, used to build tools that send, receive, or test carrier messaging flows.) support for OAuth flows.
    • Desktop notifications when Goose finishes a task.
    • Auto-compaction for the context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. to manage long sessions automatically.
    • Context window usage now shown inline in the status bar and chat input.
    • Voice dictation support via direct-ACP pattern in goose2.
    • Chat attachments support in goose2.
    • Extension management screen added to the TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency..
    • Extensions settings page and context panel widget added.
    • Threads can now be associated with projects.

    17 details in Everything

24. v1.30.0 Apr 8, 2026 · release notes

    Goose v1.30.0 adds goose serve, goose doctor, egress logging, new Gemini/Copilot/Zhipu providers, and major TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. improvements.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.30.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.30.0

    └──▷ TRY IT
    Run Goose as a persistent background service, useful for integrations or automation pipelines.
    copy

    $ goose serve

    Diagnose misconfigured environments or credential issues before starting a session.
    copy

    $ goose doctor

    Prevent tool output from being truncated when inspecting verbose command results during a session.
    copy

    $ GOOSE_SHOW_FULL_OUTPUT=true goose run --text 'List all open ports and their processes'

    • New goose serve subcommand runs Goose as a background service.
    • New goose doctor command diagnoses common configuration and environment issues.
    • New egress logging inspector logs and inspects all outbound network calls for security visibility.
    • Adds Gemini OAuth provider for first-class OAuth-based authentication with Google Gemini.
    • Adds Copilot ACP provider for GitHub Copilot via ACP.

    +23 more show less

    • Adds Zhipu AI as a declarative provider with configurable ZHIPU_BASE_URL override.
    • Adds Gemma 4 as a local model option.
    • Supports configurable fast_model for declarative providers to designate a lightweight model for quick tasks.
    • Tab-expandable tool call output collapses and expands cleanly in the TUI.
    • Independent --text mode decouples headless text output from the normal TUI.
    • New GOOSE_SHOW_FULL_OUTPUT config option disables tool output truncation.
    • GOOSE_CLI_SHOW_THINKING can now be set in config.yaml in addition to env vars.
    • GOOSE_CONTEXT_LIMIT is now readable from config.yaml.
    • GOOSE_DISABLE_KEYRING is now honoured from config.yaml at startup.
    • Configurable per-extension timeouts via ACP _meta or a global default.
    • Skills are now visible and browsable from the desktop app UI.
    • Skills are now managed as a dedicated platform extension.
    • Recursive skill discovery finds skills at any depth under configured roots.
    • Adds react-intl-based internationalization infrastructure to the desktop app.
    • Per-delegate max_turns override is now supported.
    • skip_canonical_filtering option added for declarative providers.
    • Optional native-tls support added as an alternative to rustls for system-TLS environments.
    • AWS provider dependencies are now feature-gated behind the aws-providers feature flag for leaner builds.
    • Telemetry is now behind Cargo feature flags.
    • Parallelized extension loading in ACP server for faster startup.
    • Secrets files are now created with tighter OS-level permissions.
    • Adds Rust ACP client for extension methods.
    • Adds Inference Mesh settings tab.

    28 details in Everything

25. v1.29.0 Mar 31, 2026 · release notes

    Goose v1.29.0 adds orchestration support, Gemini-ACP provider, GOOSE_SHELL config, Sigstore provenanceprovenanceEvidence recording where a software artifact came from and how it was built, including its source and build process. It helps a consumer check that the artifact matches a trusted build. verification, and stabilized ACP session APIs.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.29.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.29.0

    └──▷ TRY IT
    Override the shell Goose uses for command execution — useful when your default shell lacks the environment your tasks need.
    copy

    $ GOOSE_SHELL=/usr/bin/zsh goose run --session my-session

    • Adds GOOSE_SHELL environment variable to configure the preferred shell used by Goose.
    • Adds Sigstore/SLSASLSASupply-chain Levels for Software Artifacts, a framework for assessing how software was built and protecting that process from tampering. Provenance records help verify where a build came from. provenance verification for Rust self-update, strengthening supply-chain integrity.
    • Adds a Gemini-ACP provider with subscription model support.
    • Adds orchestration support for coordinating multi-agent workflows.
    • Stabilizes ACP session/set_config, list, delete, and close endpoints; adds set_config to the ACP session API.

    +8 more show less

    • Adds ability to manage sub-recipes in the desktop UI.
    • Displays delegate sub-agent logs in the desktop UI.
    • Adds expandable sidebar to show full chat names in the desktop UI.
    • Adds tool filtering by MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Apps visibility metadata.
    • Passes toolInfo to MCP Apps via hostContext.
    • Adds CodeMode ToolDisclosure variants support.
    • Feature-gates local inference dependencies for leaner builds.
    • Upgrades ink renderer for performance and switches CLI from HTTP to stdio transport.

    13 details in Everything

26. v1.28.0 Mar 18, 2026 · release notes

    Goose v1.28.0 adds adversarial safety agent, Claude adaptive thinking, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Roots, new LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. providers, and ACP filesystem delegation.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.28.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.28.0

    • Adds an adversarial agent that monitors sessions to prevent information leakage and other unsafe behaviors.
    • Supports Claude adaptive thinking, enabling dynamic reasoning depth for Claude models.
    • Introduces a new onboarding flow for first-time setup.
    • Persists GooseMode per-session via the session database so mode selection survives restarts.
    • Propagates session.id to OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. spans and log records for end-to-end tracing.

    +16 more show less

    • Migrates the autovisualiser extension to MCP Apps.
    • Adds ACP session/set_mode handler for programmatic mode switching via ACP.
    • Supports MCP Roots, enabling scoped filesystem access declarations for MCP servers.
    • Loads hint files from nested subdirectories, not just the top-level directory.
    • Adds ACP read tool for delegating filesystem I/O to ACP clients.
    • Adds a summarize tool for deterministic, reproducible file reads.
    • Adds ACP providers for Claude Code and Codex (claude-acp, codex-acp).
    • Captures reasoning summaries from the OpenAI Responses API.
    • Adds MiniMax as a provider via an Anthropic-compatible API.
    • Adds Tensorix as a declarative LLM provider.
    • Supports individually loadable skill files via load() in summon.
    • Adds Avian as an LLM provider.
    • Adds a base path field for custom provider configuration.
    • Improves private channel detection for the Slack bot integration.
    • Enables preferred microphone selection for voice input.
    • Supports secret interpolation in streamable HTTP extension URLs.

    21 details in Everything

27. v1.27.0 Mar 5, 2026 · release notes

    Goose v1.27.0 adds structured shell output, tree-sitter ASTASTAn abstract syntax tree, a structured representation of parsed code. Tools inspect or transform this tree to understand expressions and statements rather than just matching text. analysis, Anthropic adaptive thinking, and self-signed HTTPSHTTPSHTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. for goosed.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.27.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.27.0

    └──▷ TRY IT
    Run a shell command and consume the structured stdout/stderr output separately for automated analysis or CI pipelines.
    copy

    $ # When Goose executes shell commands, the tool now returns:
    # { "stdout": "...", "stderr": "..." }
    # Reference stdout or stderr independently in follow-up instructions or recipes.

    • Adds structured {stdout, stderr} return from the shell tool with an explicit output schema, making it easier to parse command results programmatically.
    • New platform extension with tree-sitter AST parsing enables deeper code analysis within Goose sessions.
    • Supports Anthropic adaptive (extended) thinking mode for compatible Claude models.
    • Adds fullscreen and Picture-in-Picture (PiP) display modes for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Apps.
    • Introduces self-signed HTTPS support for the goosed server, enabling secure local deployments.

    +3 more show less

    • Adds code-mode feature flag to include pctx/Code Mode as an optional dependency.
    • Adds base_path field to custom provider config for more flexible provider routing.
    • Adds GPT 5-3-Codex model support in the Databricks provider.

    8 details in Everything

28. v1.26.0 Feb 26, 2026 · release notes

    Goose v1.26.0 adds local llama.cpp inference, Cerebras/Moonshot/LMStudio providers, Telegram gateway, and sub-agent model config.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.26.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.26.0

    └──▷ TRY IT
    Pin sub-agent tasks to a cheaper or faster model while keeping your primary agent on a more capable one.
    copy

    $ GOOSE_SUBAGENT_MODEL=claude-haiku-4-5 GOOSE_SUBAGENT_PROVIDER=anthropic goose session start

    • Adds local inference provider backed by llama.cpp with HuggingFace model management.
    • Adds Cerebras provider support.
    • Adds Moonshot and Kimi Code as declarative providers.
    • Adds LMStudio as a declarative provider.
    • Adds a gateway to chat with Goose via Telegram and similar messaging platforms.

    +17 more show less

    • Adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. apps sampling support.
    • Adds Neighborhood extension to the Extensions Library.
    • Overhauled computer controller with Peekaboo integration.
    • Adds a TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. client for goose-acp.
    • Adds GOOSE_SUBAGENT_MODEL and GOOSE_SUBAGENT_PROVIDER config options to control sub-agent model selection.
    • Adds configurable OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. logging level.
    • Adds GoosePlatform in AgentConfig and MCP initialization.
    • Adds Gemini CLI streaming support via stream-json events.
    • Exposes context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. utilization to the agent via MOIM.
    • Adds auto-submit for recipes that have been accepted.
    • Adds Claude Code permission prompt routing for approve mode.
    • Displays token counts directly for free providers.
    • Adds a TypeScript SDK for ACP extension methods.
    • Adds Bedrock prompt cache support.
    • Removes allows_unlisted_models flag — custom model entry is now always permitted.
    • Displays working directory in the UI.
    • Adds local model settings access from the bottom-bar model menu.

    22 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The allows_unlisted_models flag has been removed; custom model entry is now unconditionally enabled and any configuration relying on this flag will no longer be respected.
29. v1.0.25 Feb 26, 2026 · release notes

    Goose v1.0.25 adds Snowflake provider, a full cron scheduler with pause/cancel/edit, LanceDB tool selection, and Claude 4 on Vertex.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.25 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.25

    • Adds Snowflake as a supported LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider.
    • Introduces a cron scheduler for goose-server with schedule creation, editing, pause/unpause, task cancellation, and live run tracking.
    • Supports schedule creation from deep links with comprehensive extension support.
    • Adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. router skeleton and handling of MCP server notification messages.
    • Enables LanceDB vector-based tool selection for smarter tool routing.

    +9 more show less

    • Adds GitHub Copilot streaming support, enabling gpt-4.1 and Claude models via Copilot.
    • Adds Claude 4 support in the Vertex provider dropdown with proper context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. limits.
    • Enables viewing and editing existing recipes in the Desktop UI.
    • Supports recipe discovery via the GOOSE_RECIPE_PATH environment variable and recipe directory paths.
    • Adds screenshot paste support in the Desktop UI.
    • Enables interactive deletion of sessions from the CLI.
    • Introduces configurable tool_params_max_length setting.
    • Adds Google Drive labels support.
    • Adds a Kotlin function for getting structured outputsstructured outputsForcing a model's response to match a given format or JSON schema, so software can parse it reliably instead of hoping the text is well formed..

    14 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Tool router environment variable names have been updated to include a goose prefix (e.g., existing env vars without the prefix will no longer be recognized).
30. v1.25.0 Feb 18, 2026 · release notes

    Goose v1.25.0 adds macOS sandboxingsandboxingRunning code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls., unified summon extension, Claude Code streaming, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. support for agentic CLI providers, and SLSASLSASupply-chain Levels for Software Artifacts, a framework for assessing how software was built and protecting that process from tampering. Provenance records help verify where a build came from. build provenanceprovenanceEvidence recording where a software artifact came from and how it was built, including its source and build process. It helps a consumer check that the artifact matches a trusted build..

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.25.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.25.0

    • Runs Goose Desktop inside a macOS sandbox for enhanced security isolation.
    • Introduces a unified summon extension, replacing the separate subagent and skills extensions.
    • Integrates MCP Apps UI via AppRenderer from the @mcp-ui/client SDK with fallback request handler support.
    • Enables editing a recipe's model and extensions directly in the desktop GUIGUIA graphical user interface, controls such as windows, buttons and menus. It lets people use software visually rather than entering commands..
    • Allows recipes to specify their own provider and model via recipe config.

    +16 more show less

    • Adds streaming output support for the Claude Code CLI provider.
    • Supports dynamic model listing and mid-session model switching for Claude Code.
    • Gemini CLI provider now uses stream-json output and reuses sessions for efficiency.
    • MCP extensions now work with agentic CLI providers (Claude Code, Codex, Gemini CLI).
    • Exposes reasoning_content in the API for reasoning models.
    • Adds support for the extra field in chat completion tool calls for Gemini OpenAI compatibility.
    • Respects standard OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. environment variables for exporter selection.
    • Adds model selection support for ACP session/new and session/set_model endpoints.
    • Allows overriding default bat syntax highlighting themes via environment variables.
    • Sets AGENT=goose environment variable automatically for cross-tool compatibility.
    • Generates manpages automatically for goose-cli.
    • Enables the Top Of Mind (tom) platform extension by default.
    • Filters the Apps page to show only standalone Goose Apps.
    • Advertises MCP Apps UI extension capability (via rmcp 0.15.0 upgrade).
    • Supports custom headers for custom OpenAI-compatible providers in the Desktop.
    • Release workflows now include SLSA build provenance attestations for supply-chain security.

    21 details in Everything

31. v1.24.0 Feb 12, 2026 · release notes

    Goose v1.24.0 adds unified summon extension, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tools for agentic CLI providers, session search API, and SLSASLSASupply-chain Levels for Software Artifacts, a framework for assessing how software was built and protecting that process from tampering. Provenance records help verify where a build came from. provenanceprovenanceEvidence recording where a software artifact came from and how it was built, including its source and build process. It helps a consumer check that the artifact matches a trusted build..

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.24.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.24.0

    └──▷ TRY IT
    Filter long lists instantly when picking models or extensions interactively in the CLI.
    copy

    $ goose configure

    • New unified summon extension replaces separate subagent and skills extensions with a single streamlined interface.
    • New built-in Top of Mind (tom) platform extension enabled by default for contextual awareness.
    • Claude Code, Codex, and Gemini CLI agentic providers can now use MCP tools.
    • Claude Code gains dynamic model listing and mid-session model switching.
    • Claude Code adds stream-json protocol for persistent sessions.

    +28 more show less

    • Reasoning content is now returned in the API for reasoning models.
    • Configurable thinking level for Gemini 3 models.
    • Codex now supports image inputs.
    • Google provider adds MessageContent::Image support for user messages.
    • Recipes can now have their model and extensions edited directly from the GUIGUIA graphical user interface, controls such as windows, buttons and menus. It lets people use software visually rather than entering commands..
    • Recipes can load the provider/model specified inside the recipe config itself.
    • Jinja variables can now be escaped in recipes to pass them through as literal text in prompts.
    • New API endpoint for searching session content.
    • Inline rename for chat sessions in the sidebar.
    • Desktop UI now supports deleting custom providers.
    • Custom headers field added for custom OpenAI-compatible providers in the Desktop UI.
    • Subagent tool calls are now displayed in both the CLI and UI.
    • Type-to-search filtering added to CLI select/multiselect dialogs.
    • AGENT_SESSION_ID environment variable is now exposed to extension child processes.
    • AGENT=goose environment variable set for cross-tool compatibility.
    • Environment variables are now passed to shell commands.
    • Global config option to disable automatic session naming.
    • Port of Context (pctx) support added for Code Mode.
    • Default bat syntax-highlight themes can now be overridden via environment variables.
    • MCP Apps gain Permission Policy support for sandbox iframes.
    • MCP Apps integrate the AppRenderer from the @mcp-ui/client SDK.
    • Upgraded rmcp to 0.15.0 with MCP Apps UI extension capability advertised.
    • Groq provider updated with Preview Models.
    • ACP protocol gains model selection support via session/new and session/set_model.
    • Manpage generation added for goose-cli.
    • SLSA build provenance attestations added to release workflows.
    • Supports building with CUDACUDANVIDIA's platform and programming interface for running general computation on its GPUs. Nearly every AI training and inference stack is built on it first. as the candle backend.
    • Proper proxy support on Windows and macOS.

    33 details in Everything

32. v1.23.0 Feb 4, 2026 · release notes

    Goose v1.23.0 adds WebSocketWebSocketA protocol that upgrades an HTTP connection into a persistent, full-duplex channel, letting a server push data to a client without repeated polling. Cyber tools use it for real-time event streaming and live command output. and Streamable HTTPStreamable HTTPAn MCP transport protocol that lets a client send JSON-RPC requests over HTTP POST while the server streams responses back using Server-Sent Events, replacing the older stdio and SSE transports for remote tool communication. transports for ACP, Laminar observability integration, and a CLI flag to skip profile extensions.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.23.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.23.0

    • Adds WebSocket transport support for goose-acp agent communication.
    • Adds Streamable HTTP transport for ACP and goose-acp usage.
    • Integrates Laminar for agent observability and tracing.
    • New CLI flag to skip loading profile extensions at startup.
    • Adds a thinking status hook for surfacing model reasoning state.

    +3 more show less

    • Adds hourly skill fetching to keep skills automatically up to date.
    • Introduces a platform extension migrator alongside a code mode rename.
    • Adds an Ask AI Discord bot integration.

    8 details in Everything

33. v1.22.0 Jan 29, 2026 · release notes

    Goose v1.22.0 adds session forking, BERTBERTA transformer encoder trained to understand text using context on both sides of a word. Its representations are useful for search, classification and other language-understanding tasks.-based command injectioncommand injectionAn attack that makes an application execute operating-system commands supplied through untrusted input. It often happens when input is passed unsafely to a shell. detection, container extensions, and an OSSOSSOpen-source software: code whose licence permits anyone to inspect, modify, and redistribute it. Cyber tools build on it to reuse auditable libraries without licensing fees. skills marketplace.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.22.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.22.0

    └──▷ USE IT
    Cap how many turns a recipe's subagent is allowed to run, preventing runaway automation.
    yaml copy

    # In your recipe YAML
    max_turns: 10

    Authenticate to AWS Bedrock using a bearer tokenbearer tokenA credential that grants access to whoever presents it, without a separate proof of possession. Protecting the token matters because a stolen copy can be used by someone else. instead of standard AWS credentials.
    copy

    $ export AWS_BEARER_TOKEN_BEDROCK=<your-token>
    goose run --provider bedrock --model anthropic.claude-3-5-sonnet-20241022-v2:0

    • Adds session forking to branch existing conversations into independent sessions.
    • Introduces BERT ML-based command injection detection in tool calls for prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. defense.
    • Supports running extensions inside containers via new Container Extensions option.
    • Adds an open-source skills marketplace for discovering and adding community skills.
    • Integrates OVHcloud AI as a new model provider.

    +15 more show less

    • Adds max_turns setting to recipes and subagents to cap agent execution depth.
    • Supports requires_auth flag for configuring custom providers that need no authentication.
    • Adds AWS_BEARER_TOKEN_BEDROCK environment variable support for AWS Bedrock authentication.
    • Supports custom headers with the Anthropic provider.
    • Enables external editor support (vim, helix, etc.) for composing prompts.
    • Adds a built-in Goose Documentation Skill for in-session reference to Goose docs.
    • Adds a Diagnostics Viewer for debugging Goose sessions.
    • Makes skills available in codemode.
    • Adds PR code review functionality.
    • Adds Vibe MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Apps, a new MCP apps integration.
    • Adds collapsible chat navigation for a cleaner desktop UI.
    • Adds new keyboard shortcuts.
    • Adds reduced motion support for accessibility preferences.
    • Creates a default gooseignore file automatically when one is missing.
    • Adds Codex subscription support as a provider option.

    20 details in Everything

34. v1.21.0 Jan 22, 2026 · release notes

    Goose v1.21.0 adds multi-chat sessions, standalone MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. apps, CLI permission prompts, parallel isolated agent sessions, and an MCP server for recipes.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.21.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.21.0

    • Adds multi-chat support to manage multiple concurrent chat sessions.
    • Supports standalone MCP applications with optional prefersBorder UI option.
    • Adds an MCP server for recipes, enabling recipe management via the MCP protocol.
    • Introduces CLI permission prompts so the agent asks for user approval before sensitive actions.
    • Enables parallel sessions with isolated agent state in goose-acp.

    +8 more show less

    • Adds Flatpak packaging for Linux distributions.
    • Supports user-configurable templates for customizing agent prompts.
    • Adds GCP Vertex model listing with organization policy filtering.
    • Supports optional request params and context limit configuration via GOOSE_PREDEFINED_MODELS.
    • Adds customizable new-line keybinding in the CLI.
    • Standardizes agent-session-id across providers for improved session correlation.
    • Excludes subagent tool from code_execution filtering, enabling subagent delegation in code execution mode.
    • Adds graceful fallback handling when the system keyring is unavailable.

    13 details in Everything

35. v1.20.0 Jan 16, 2026 · release notes

    Goose v1.20.0 adds ML prompt-injectionprompt-injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. detection, HTTP proxy support, streaming for Gemini and Vertex AI, and a configurable max-tokens env var.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.20.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.20.0

    └──▷ TRY IT
    Cap token usage per session to avoid runaway costs when running Goose against large codebases.
    copy

    $ GOOSE_MAX_TOKENS=4096 goose run --text "Summarize all TODOs in this repo"

    Route Goose's provider API calls through a corporate HTTP proxy for environments that block direct outbound traffic.
    copy

    $ HTTPS_PROXY=http://proxy.corp.example.com:8080 goose run --text "Scan this project for secrets"

    • Adds ML-based prompt injection detection to identify and reduce prompt injection attacks, with tuned confidence levels to lower false-positive rates.
    • Supports HTTP proxy via new proxy configuration, enabling Goose to route traffic through corporate or intercepting proxies.
    • Adds GOOSE_MAX_TOKENS environment variable to cap the maximum tokens sent to the model.
    • Adds streaming support for the Google Gemini provider, reducing latency for long responses.
    • Adds streaming support for the Vertex AI provider.

    +8 more show less

    • Adds retry logic for model fetching from providers, improving resilience against transient API errors.
    • Adds a hotkey to toggle full tool output display in the UI.
    • Adds frameDomains and baseUriDomains Content Security PolicyContent Security PolicyA browser security mechanism, standardized by the W3C, that lets a web server declare via HTTP headers which sources of scripts, styles, and other resources a page is allowed to load, blocking unauthorized content injection. support for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Apps, enabling controlled embeddingembeddingLists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. of external domains.
    • Enables changing the working directory and loaded extensions within the same session window without restarting.
    • CLI now surfaces extension load errors directly, making misconfigured extensions immediately visible.
    • Adds spell-check setting to the desktop UI.
    • Hides advanced recipe options under expandable content, reducing UI clutter for common workflows.
    • Caches tools in ExtensionManager for faster code-execution startup.

    13 details in Everything

36. v1.19.0 Jan 6, 2026 · release notes

    Goose v1.19.0 adds shell completions, OpenAI Codex CLI provider, JSONLJSONLA text format where each line is a self-contained, valid JSON object, making it easy to stream, append, and process large datasets line by line without parsing a full JSON array into memory. streaming, and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server support from editors like Zed.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.19.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.19.0

    └──▷ TRY IT
    Set up persistent shell completions for zsh so Goose subcommands and flags tab-complete in your terminal.
    copy

    $ goose completion zsh > ~/.zfunc/_goose

    • Adds completion subcommand to generate shell completions for bash, zsh, fish, and more.
    • Adds OpenAI Codex CLI as a supported provider.
    • Enables prompt caching for Anthropic models served via Databricks.
    • Honors MCP servers configured in clients like Zed over stdio and HTTP transports.
    • Adds MCP app renderer for richer, interactive UI experiences within MCP sessions.

    +2 more show less

    • Adds JSONL streaming option for JSON output mode, enabling line-by-line consumption of streamed responses.
    • Uses MCP server names as extension identifiers in code mode for clearer context.

    7 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !SSESSEServer-Sent Events, a browser API and HTTP-based protocol where a server pushes a stream of text events to a client over a single long-lived connection, giving tools a lightweight alternative to WebSockets for one-way real-time updates. transport for MCP is removed as part of the upgrade to rmcp 0.12.0 and sacp 10.0.0; configurations relying on SSE transport will break.
37. v1.18.0 Dec 19, 2025 · release notes

    Goose v1.18.0 adds slash commands for prompt/history management, OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. & xAI streaming, and skills discovery paths.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.18.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.18.0

    └──▷ TRY IT
    Browse available prompts and run one by name during an interactive session — useful for quickly invoking saved prompt templates.
    copy

    $ /prompts
    /prompt my-security-review-prompt

    Inspect a prompt's parameters and metadata before executing it, helpful when you want to understand a recipe's required inputs.
    copy

    $ /prompt my-security-review-prompt --info

    Compact or clear conversation history mid-session to stay within context limits without starting a new session.
    copy

    $ /compact
    # or, to wipe the full history:
    /clear

    • Adds /prompts slash command to list available prompts, /prompt to execute a prompt or inspect it with --info, /compact to compact conversation history, and /clear to clear conversation history.
    • Adds recipe slash command parsing with improved parameter handling.
    • Adds streaming support for OpenRouter and xAI providers.
    • Adds skills auto-discovery from .agents/skills and ~/.config/agent/skills paths.
    • Adds a Goose PR reviewer workflow for automated code review.

    +2 more show less

    • Adds pricing integration with the canonical model registry.
    • Adds ability to disable the auto-update feature for the CLI.

    7 details in Everything

38. v1.17.0 Dec 18, 2025 · release notes

    Goose v1.17.0 adds Inception provider, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. elicitation, external server support, OpenAI streaming, and new built-in recipes.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.17.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.17.0

    • Adds Code Mode MCP platform extension for enhanced code execution capabilities via tool callingtool callingA model's ability to answer with a structured request to run a named function or API, which the application executes and returns. It is how agents take actions..
    • Supports MCP elicitation, enabling interactive prompts during tool execution.
    • Enables connection to an external goosed server backend.
    • Adds Inception provider, bringing Mercury model support to Goose.
    • Adds streaming support for the OpenAI responses API.

    +5 more show less

    • Introduces canonical (standardized) model definitions across providers.
    • Adds search bar and filtering to the recipes view.
    • Enables automatic provider detection from API key during onboarding.
    • Adds built-in DataHub recipe for searching and understanding data in DataHub.
    • Adds built-in Flutter PR Code Review recipe for reviewing Flutter pull requests.

    10 details in Everything

39. v1.16.0 Dec 10, 2025 · release notes

    Goose v1.16.0 adds native terminal support, remote access, slash commands, custom headers, and a new /agent/stop API endpoint.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.16.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.16.0

    └──▷ TRY IT
    Invoke Goose directly from your native terminal without launching the desktop UI.
    copy

    $ @goose explain the CVEs in this repo's dependencies

    Stop a running agent via the new API endpoint in a headless or CI environment.
    copy

    $ curl -X POST http://localhost:<port>/agent/stop

    • Adds @goose native terminal support, enabling Goose interaction directly from your system terminal.
    • Adds Goose remote access capability for controlling Goose from external clients.
    • Adds slash commands in the chat interface for quick in-session actions.
    • Adds custom headers and toggle keyring CLI options for provider configuration.
    • Adds /agent/stop API endpoint and makes max active agents configurable.

    +13 more show less

    • Adds support for loading sessions in ACP (Agent Communication Protocol).
    • Enables recipe deeplink parameters for pre-population of recipe inputs.
    • Adds SKILLS.md support for Claude compatibility.
    • Adds math and science symbology rendering via KaTeX.
    • Adds markdown export from URL.
    • Adds Firefox sidebar contextual information support.
    • Adds automatic update installation.
    • Adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface.-UI proxy for better MCP message handling.
    • Adds scheduler information to diagnostics output.
    • Persists provider name and model config within the session.
    • Adds extension selector behind an ALPHA feature flag.
    • Adds multiple file popup support for simultaneous file interactions.
    • Supports onboarding auto-detection of provider from API key.

    18 details in Everything

40. v1.15.0 Nov 19, 2025 · release notes

    Goose v1.15.0 adds Grok provider support, Supabase MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface., recipe security scanning, and ACP session management.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.15.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.15.0

    • Adds Grok as a supported AI provider.
    • Adds Supabase MCP server to the extension registry.
    • Adds security scanning of recipes at save time to catch malicious or unsafe content.
    • Adds SessionManager integration for ACP (Agent Communication Protocol) sessions.
    • Adds MCP UI proxy with improved message handling for extension interactions.

    +2 more show less

    • Reads paths from both interactive and login shells, improving environment compatibility.
    • Adds provider error proxy for simulating provider errors during development and testing.

    7 details in Everything

41. v1.14.0 Nov 12, 2025 · release notes

    Goose v1.14.0 adds Mistral AI and GitHub Copilot providers, platform tools via CLI, and env-var substitution in auth config.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.14.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.14.0

    • Adds Mistral AI as a new LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider option.
    • Supports newer GitHub Copilot AI model versions as a provider.
    • Enables access to platform tools through the CLI.
    • Supports dynamic environment variable substitution in authentication configuration blocks.
    • Adds a macOS dock menu with a New Window option for better desktop integration.

    +1 more show less

    • Adds a pending extension indicator in the extension panel for visual loading feedback.

    6 details in Everything

42. v1.13.0 Nov 6, 2025 · release notes

    Goose v1.13.0 adds Mistral AI provider, newer GitHub Copilot models, platform tools in CLI, and env-var auth config.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.13.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.13.0

    • Adds Mistral AI as a supported LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider.
    • Supports newer GitHub Copilot models.
    • Exposes platform tools in the CLI (previously desktop-only).
    • Enables environment variables in auth configuration.
    • Adds Ctrl-C cancellation of in-flight AI requests from the CLI.

    +2 more show less

    • Adds right-click dock icon for opening a new window on macOS.
    • Shows a loading indicator for extensions while they initialize.

    7 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The JetBrains integration has been removed; any existing JetBrains-based workflows will no longer function after upgrading.
43. v1.12.0 Oct 24, 2025 · release notes

    Goose v1.12.0 adds system diagnostics, extension search, code snippet saving, and two new built-in recipes.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.12.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.12.0

    • Adds system diagnostics tools for inspecting Goose's runtime environment.
    • Adds extension search functionality to quickly find and manage extensions.
    • Adds code snippet formatting and saving directly from conversations.
    • Introduces a Test Coverage Optimizer built-in recipe.
    • Introduces a Smart Task Organizer built-in recipe.

    +7 more show less

    • Adds a schedule button for recipe entries, enabling scheduling from the recipe list UI.
    • Auto-creates recipe directories when saving a recipe to a new path.
    • Adds UI parameter input controls for configuring recipe parameters interactively.
    • Redesigns the extension management system.
    • Adds progress tracking with trend indicators.
    • Improves prompt caching with stable system prompts for more efficient provider usage.
    • Adds provider request logging for visibility into outbound LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. requests.

    12 details in Everything

44. v1.11.0 Oct 17, 2025 · release notes

    Goose v1.11.0 adds session search, new built-in recipes, Apify MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. integration, voice dictation, and declarative providers.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.11.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.11.0

    • Introduces a Code Documentation Generator built-in recipe.
    • Introduces a CI/CD Pipeline built-in recipe.
    • Introduces a Daily Standup Report Generator built-in recipe.
    • Adds a self-test recipe for validating Goose installations.
    • Adds an API Documentation Generator prompt.

    +9 more show less

    • Adds Apify MCP integration with supporting documentation.
    • Adds a flag to enable ElevenLabs voice dictation.
    • Introduces declarative provider configuration.
    • Sorts providers alphabetically for consistent discovery.
    • Adds a Copy as Markdown button to documentation pages.
    • Enables starting an agent session directly from a recipe ID or deeplink.
    • Improves Rust analysis output for the analyze tool.
    • Adds Firecrawl MCP tutorial and integration documentation.
    • Expands Ollama integration with a modified default model and additional models.

    14 details in Everything

45. v1.10.0 Oct 14, 2025 · release notes

    Goose v1.10.0 adds prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. detection UI, GOOSE_DEBUG support, custom lead/worker models, AWS Bearer TokenBearer TokenA credential that grants access to whoever presents it, without a separate proof of possession. Protecting the token matters because a stolen copy can be used by someone else. auth, and more.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.10.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.10.0

    └──▷ TRY IT
    Enable verbose debug output when troubleshooting CLI sessions or extension behavior.
    copy

    $ GOOSE_DEBUG=1 goose session start

    List only the most recent sessions under a specific path to audit or resume recent work.
    copy

    $ goose session list --path /projects/myapp --limit 10

    Open a saved recipe directly in the Goose desktop app from the CLI.
    copy

    $ goose recipe open <path>

    • Adds prompt injection detection settings UI with updated logging for security-aware deployments.
    • New GOOSE_DEBUG environment variable for CLI debug output.
    • Adds path and limit flags to the session list subcommand for filtering session history.
    • New subcommand for opening recipes directly in the desktop app.
    • Supports setting custom models for lead and worker agents in multi-model workflows.

    +22 more show less

    • Adds grok-code-fast-1 model support for the xAI provider.
    • Adds claude-sonnet-4-5 model support.
    • Enables AWS Bearer Token (AWS_BEARER_TOKEN_BEDROCK) authentication for Bedrock runtime.
    • Parses and respects retryDelay from Google API RateLimitExceeded errors for smarter rate-limit handling.
    • Adds elapsed time display to CLI session output.
    • Lazy infinite scroller for session history view improves navigation of large session lists.
    • Supports headers in extension deeplinks.
    • Allows empty API key when registering a custom provider.
    • Shell errors thrown on interactive commands to surface blocked interactions explicitly.
    • AgentManager foundation enables unified subagent execution and concurrent session creation.
    • Persists dynamic extension config so recipe sessions with extensions can be resumed.
    • Displays extension install notes in the Add custom extension form.
    • Adds dev and alpha environment indicator in the UI.
    • Using --resume with --name now also accepts session IDs.
    • CLI now returns a non-zero exit code when a session ends with an error.
    • Goose can now tell the system what shell it is using.
    • New Code Documentation Migrator, Smart Meeting Assistant, Code Review Mentor, Smart Task Organizer, advanced data analysis pipeline, advanced software project generator, Messy Column Fixer, and OpenMetadata recipes added to the prompt/recipe library.
    • New multi-channel researcher and content amplification prompts added to the prompt library.
    • New analyze tool hint for focus mode when used on file paths.
    • Screen access capability added (feat: Use the screen, goose).
    • CLI web auth token support added.
    • Session manager added for improved session lifecycle management.

    27 details in Everything

46. v1.9.2 Oct 9, 2025 · release notes

    Goose v1.9.2 adds Claude Sonnet 4.5 model support and macOS Tahoe compatibility.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.9.2 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.9.2

    • Adds claude-sonnet-4-5 as a supported model.
    • Upgrades Electron runtime for macOS Tahoe compatibility.

    2 details in Everything

47. v1.9.0 Sep 24, 2025 · release notes

    Goose v1.9.0 adds prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. detection, auto-compaction, an analyze tool, streaming for Tetrate ARS, and richer recipe management in the UI.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.9.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.9.0

    • Adds prompt injection detection (pattern-matching) to flag malicious inputs before they reach the model.
    • Automatically compacts sessions for streaming providers when context lengthcontext lengthThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. is exceeded, with an offer to summarize or clear overly long conversations.
    • New analyze tool added to the Developer extension for code and file analysis.
    • Adds streaming support to the Tetrate Agent Router Service (ARS) provider.
    • Supports importing file contents from recipe file input-type parameters.

    +8 more show less

    • Desktop UI now supports uploading JSON recipe files directly.
    • Recipe schema validation added to the UI to catch malformed recipes before execution.
    • UI recipe activities now support parameter replacement and always appear at the top of chat when a recipe is active.
    • Adds Message Metadata for Visibility Control, enabling fine-grained control over which messages are shown.
    • I Ching MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server added to the extension catalog.
    • Upgraded agent-client-protocol (ACP) with improved tool experience and nicer tool name formatting.
    • Recipe title field added to the import form.
    • UI filters invalid, duplicate, and unused parameters from recipes automatically.

    13 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The --path flag has been removed to avoid session directory validation issues.
48. v1.8.0 Sep 11, 2025 · release notes

    Goose v1.8.0 adds ACP support, multi-model switching, streaming agent routing, and YAML recipe imports.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.8.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.8.0

    • Adds Agent Client Protocol (ACP) support for interoperability with ACP-compatible agents.
    • Supports multi-model/provider configuration with automatic switching between providers.
    • Enables streaming support in the Tetrate Agent Router.
    • Adds configurable Bedrock retry settings.
    • Supports importing YAML recipes directly in the Desktop app.

    +9 more show less

    • Enables reading local file inputs inside recipes.
    • Adds message metadata for visibility control over what is shown in the UI.
    • Enables deleting sessions from the history list in the Desktop UI.
    • Tracks sessions per agent for better session management.
    • Makes the auto-compact threshold configurable via the UI.
    • Shows enabled extensions first in the extensions list.
    • Adds a unified diff view in the developer editor.
    • Aligns recipe metadata metrics across scheduler, UI, and CLI.
    • Upgrades to RMCP 0.6.2.

    14 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Deprecated Claude 3.5 models have been removed.
49. v1.7.0 Aug 29, 2025 · release notes

    Goose v1.7.0 adds Android/Termux computer control, extension malware scanning, auto-visualization of structured data, and dynamic provider model querying.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.7.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.7.0

    • Adds computer control support on Android devices via Termux.
    • Adds extension malware scanning to detect malicious code before installation.
    • New MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface.-UI Autovisualiser automatically renders structured data returned in sessions.
    • UI now dynamically queries providers for their supported models at runtime.
    • Re-introduces session sharing between users.

    +5 more show less

    • Supports port numbers in custom OpenAI-compatible provider configurations.
    • Adds support for reading OTLPOTLPOpenTelemetry Protocol, a standard way to send traces, metrics and logs between applications, collectors and monitoring systems. It lets telemetry move across compatible tools. configuration from config files and environment variables.
    • Enables loading of nested hint files for better project organization.
    • Adds Tetrate Agent Router Service as a new provider in the registry.
    • Adds scroll state support for improved navigation in the chat session list.

    10 details in Everything

50. v1.6.0 Aug 22, 2025 · release notes

    Goose v1.6.0 adds mTLSmTLSMutual TLS, an encrypted connection where both client and server present certificates to prove their identity. It helps services verify who they are communicating with. for all providers, PKCEPKCEProof Key for Code Exchange, an OAuth safeguard that binds an authorization code to the client that requested it. An intercepted code cannot be redeemed without the client's matching verifier. auth, message queue with interruption handling, and in-place message editing.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.6.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.6.0

    └──▷ TRY IT
    Allow Goose to automatically load project-level agent instructions without any extra flags.
    copy

    $ # Place an AGENTS.md file in your project root; Goose reads it by default on startup
    echo '# Agent Instructions
    Always use TypeScript.' > AGENTS.md
    goose session

    • Adds mTLS support across all providers for mutual-authentication to AI backends.
    • Adds PKCE support for the Tetrate Agent Router Service authentication flow.
    • New message queue system with interruption handling for more reliable agent interactions.
    • Providers can now configure a fast model specifically for summarization tasks.
    • Supports concurrent extension startup, reducing initialization latency.

    +6 more show less

    • Desktop app alerts when suspicious Unicode characters are detected in a Recipe.
    • Reads AGENTS.md by default as a context file, bringing project-level agent instructions into scope automatically.
    • MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. developer extension now accepts -1 for insert_line number to support end-of-file insertions.
    • Adds a Nix flakeNix flakeA Nix packaging construct that pins all dependencies in a lockfile and exposes a standardized interface, giving cyber tools fully reproducible builds across any machine running the Nix package manager. for reproducible Goose builds.
    • Upgrades @mcp-ui/client package with improved UI message handling.
    • Adds remark-breaks plugin to preserve single newlines in rendered markdown.

    11 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The session sharing feature has been removed; any workflows relying on it will no longer function.
51. v1.5.0 Aug 19, 2025 · release notes

    Goose v1.5.0 adds cursor-agent CLI provider, OAuth keyring persistence for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface., and per-recipe tool limits.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.5.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.5.0

    • Adds cursor-agent as a new CLI provider option.
    • Persists OAuth credentials to the system keyring for MCP integrations, eliminating repeated auth prompts.
    • Supports per-recipe configuration to limit which tools are available during a recipe run.
    • Improves model selection UX in the CLI.
    • Speeds up extension loading by parallelizing the load process.

    +3 more show less

    • Skips fetching pricing information when GOOSE_CLI_SHOW_COST is false, reducing unnecessary network calls.
    • Supports loading recipe deeplinks in a single window when the app is closed.
    • Integrates token counting reliability improvements with summarization.

    8 details in Everything

52. v1.3.0 Aug 11, 2025 · release notes

    Goose v1.3.0 adds OAuth CLI support, file completion, session description editing, client disconnect detection, and Linux build improvements.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.3.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.3.0

    • Adds OAuth provider abstraction for CLI configuration, enabling authenticated provider setup from the command line.
    • Adds file path tab-completion in the CLI for faster input.
    • Enables editing of session descriptions directly in the UI.
    • Detects client disconnects and automatically cancels in-flight tool calls.
    • Adds StreamableHTTP OAuth support via RMCP.

    +7 more show less

    • Bumps auto-compact threshold to 80%, reducing premature context compactioncontext compactionCondensing an AI conversation or working history to free space in the model's context window. A summary preserves useful information, but details can be lost..
    • Token counting in auto-compact now uses provider metadata for more accurate limits.
    • Adds GPT-5 context limit support.
    • Suppresses ANSI escape codes automatically when output is piped.
    • Adds comprehensive Linux build support with cross-compilation in bundle workflow.
    • Adds additional telemetry metrics with UI implementation.
    • Window title now reflects the current active operation.

    12 details in Everything

53. v1.2.0 Aug 6, 2025 · release notes

    Goose v1.2.0 adds parallel sub-recipe execution, LiteLLM provider, recipe retry logic, wakelock, and more.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.2.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.2.0

    • Adds experimental support for running sub-recipes multiple times in parallel for concurrent task execution.
    • Adds LiteLLM provider with automatic prompt caching support.
    • Adds recipe retry logic with configurable success criteria so recipes can self-heal on failure.
    • Adds wakelock feature to prevent the system from sleeping while Goose is working.
    • Adds inline Python extension support.

    +20 more show less

    • Adds GitHub Copilot provider ability to fetch supported models.
    • Adds OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. as an out-of-the-box provider experience for new Goose installations.
    • Adds ability for users to view and edit their non-secret configuration values.
    • Adds support for configuring the hints filename(s) via configuration.
    • Adds WaitingForUserInput state to the agent loop.
    • Adds an option to disable random thinking messages.
    • Adds an option to summarize the chat when an error is triggered (autocompact + one-shot summarization).
    • Adds environment variable to set Ollama request timeout.
    • Adds subagent independent extension manager and per-subagent turn count with a simpler agent loop.
    • Adds consolidated subagent execution for dynamic tasks.
    • Adds recipe install warning before accepting untrusted recipes.
    • Adds file picker for scheduling recipes that defaults to the recipe directory.
    • Adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. UI integration.
    • Adds mouse back navigation button support on the Settings screen.
    • Adds extension name tooltips on tool banners.
    • Adds goose info -v for stable verbose info output.
    • Updates Groq models for better tool-callingtool-callingA model's ability to answer with a structured request to run a named function or API, which the application executes and returns. It is how agents take actions. support.
    • Adds HTTP 429-style retry handling for Anthropic API HTTP 529 overloaded responses in the GCP Vertex AI provider.
    • Enables streaming for the Anthropic first-party provider.
    • Recipe runs now prompt for missing extension secrets instead of silently failing.

    25 details in Everything

54. v1.1.4 Jul 24, 2025 · release notes

    Goose v1.1.4 adds chat summarization on error and enhanced loading animations with thinking icons and flying bird.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.1.4 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.1.4

    • Adds option to summarize the chat session when an error is triggered.
    • Enhances loading states with thinking icons and a flying bird animation.

    2 details in Everything

55. v1.1.0 Jul 16, 2025 · release notes

    Goose v1.1.0 ships a redesigned desktop UI with sidebar and settings tabs, new recipe list command, glob/grep file search tools, and Windows CLI installer.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.1.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.1.0

    └──▷ TRY IT
    List all available recipes to quickly discover and reference them before running one.
    copy

    $ goose recipe list

    Clear the persisted session history mid-session to start fresh without restarting Goose.
    copy

    $ /clear

    • Adds recipe list subcommand to enumerate available recipes from the CLI.
    • New /clear command in CLI sessions clears the persisted session file.
    • Adds glob search and grep tools to improve file search capabilities.
    • Implements OpenAI streaming support for faster, real-time LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. responses.
    • Complete redesign of the Goose desktop UI with a sidebar and settings tabs.

    +1 more show less

    • Adds download_cli.ps1 PowerShell installer script for Windows users.

    6 details in Everything

56. v1.0.36 Jul 15, 2025 · release notes

    Goose v1.0.36 adds streaming LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. responses, cost estimation, OpenRouterOpenRouterA unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. model discovery, vi edit mode, and new provider/model CLI flags.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.36 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.36

    └──▷ TRY IT
    Pick a specific provider and model on the fly without touching your config — useful when testing a new model in a one-off session.
    copy

    $ goose run --provider openai --model gpt-4o <session-or-prompt>

    • Streams LLM responses in real time instead of waiting for full completion.
    • Adds per-provider cost estimation in the Goose CLI so practitioners can compare spend before committing to a provider.
    • Fetches OpenRouter's supported model list dynamically inside goose configure.
    • Exposes AZURE_OPENAI_API_KEY as a visible, configurable parameter in the config system.
    • Adds --provider and --model CLI options to the goose run command for one-off provider/model selection.

    +5 more show less

    • Adds a Streamable HTTPStreamable HTTPAn MCP transport protocol that lets a client send JSON-RPC requests over HTTP POST while the server streams responses back using Server-Sent Events, replacing the older stdio and SSE transports for remote tool communication. CLI flag for HTTP-based streaming transport.
    • Supports vi edit mode via rustyline configuration for CLI line editing.
    • Allows Ollama to be used with non-tool (chat-only) models.
    • Adds structured outputstructured outputForcing a model's response to match a given format or JSON schema, so software can parse it reliably instead of hoping the text is well formed. support in Goose CLI and Goose Desktop.
    • Alphabetizes extensions in the UI for easier navigation.

    10 details in Everything

57. v1.0.32 Jul 3, 2025 · release notes

    Goose v1.0.32 adds OAuth 2.0OAuth 2.0An open authorization framework (IETF RFC 6749) that lets a service grant limited API access to a third party without sharing credentials, giving cyber tools a standard way to authenticate against external platforms. for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface., streamable-HTTP transport, JSON schema recipe validation, and sub-recipes from GitHub.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.32 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.32

    └──▷ TRY IT
    Validate a recipe file against its JSON schema to catch structural errors before running it.
    copy

    $ goose recipe validate <path-to-recipe.yaml>

    • Adds fuzzy file search functionality for navigating files.
    • Adds JSON schema validation to the goose recipe validate CLI command.
    • Introduces max_turns setting for the agent to cap autonomous runs without user input.
    • Enables structured outputstructured outputForcing a model's response to match a given format or JSON schema, so software can parse it reliably instead of hoping the text is well formed. support in recipes, usable from both CLI and GUIGUIA graphical user interface, controls such as windows, buttons and menus. It lets people use software visually rather than entering commands..
    • GUI now renders structured output defined in recipes.

    +10 more show less

    • Enables running sub-recipes sourced directly from GitHub.
    • Adds native OAuth 2OAuth 2An open authorization framework (RFC 6749) that lets a service grant limited access to a user's resources on another service without exposing credentials, giving cyber tools a standard way to authenticate against third-party APIs..0 authentication support to the MCP client.
    • Adds streamable-HTTP transport support across backend, desktop, and CLI.
    • Supports recipe parameters in the Goose desktop app.
    • Adds a close button (X) to toast notifications.
    • Adds support for the Escape key to dismiss the settings menu.
    • Adds Playwright MCP server to the extensions list.
    • Adds /extension path for extension installation.
    • Improves UX for saving recipes.
    • Prioritizes path suffix when truncating long paths in the desktop header.

    15 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !GitHub Copilot Provider has been temporarily removed; any setup relying on it will no longer function.
58. v1.0.31 Jul 1, 2025 · release notes

    Goose v1.0.31 adds sub-recipe chaining from the CLI, async token counting, and Claude 4 model support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.31 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.31

    • Supports passing additional sub-recipes via the command line for composable recipe workflows.
    • Adds Claude 4 models as available provider options.
    • Implements async token counter with network resilience and performance optimizations.
    • Allows use of /dev/null for no-session mode, enabling fully stateless runs.

    4 details in Everything

59. v1.0.30 Jun 27, 2025 · release notes

    Goose v1.0.30 adds voice dictation, subagents, CLI providers for Claude/Gemini, cost tracking, and broader cron schedule support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.30 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.30

    └──▷ TRY IT
    Suppress all non-essential output when running a Goose session in CI or scripted pipelines.
    copy

    $ goose run --quiet --instructions 'Summarize the repo structure'

    Resume a session by its ID using the new short alias instead of --name.
    copy

    $ goose session --resume --id my-session-id

    • Adds voice dictation support via OpenAI Whisper and ElevenLabs integration.
    • Introduces subagents capability, enabling Goose to orchestrate multi-agent workflows.
    • New CLI providers for Claude Code and Gemini, expanding model backend options.
    • Adds comprehensive cost tracking display for LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. usage, with a flag to toggle visibility.
    • Wider cron schedule support for the built-in scheduler.

    +21 more show less

    • Adds optional view_range parameter to the text editor tool and text editor insert tool.
    • New sub recipe tools for composing and reusing recipe components.
    • Lists Groq-supported models from the CLI during model configuration.
    • Lists Databricks-supported models with fuzzy search during model configuration.
    • Adds --id alias for --name parameter in goose session --resume command.
    • Adds --quiet / -q flag to goose run for suppressing output.
    • Adds scheduler type setting for controlling how scheduled tasks are dispatched.
    • Adds ability to create a new directory directly from the working directory selection UI.
    • Drag-and-drop support for opening sessions from Finder in the desktop UI.
    • Adds chain-of-thoughtchain-of-thoughtIntermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens. panel displayed above assistant messages in the UI.
    • Supports optional fast edit models for accelerated editing operations.
    • Adds lead-worker model selection and real-time model display in the GUIGUIA graphical user interface, controls such as windows, buttons and menus. It lets people use software visually rather than entering commands..
    • Updates vector tool strategy to read vector DB path from an environment variable.
    • Updates Google Gemini models to the latest available versions.
    • Adds xAI Provider support for Grok models.
    • Platform tool now allows Goose to manage its own schedule autonomously.
    • Recipe Library feature for discovering and reusing recipes.
    • Adds a setting for the quit confirmation dialog.
    • Richer tool call UI messages for better visibility into agent actions.
    • Improves config file editing with recovery fallback mechanisms.
    • Native Windows CLI build support.

    26 details in Everything

60. v1.0.29 Jun 18, 2025 · release notes

    Goose v1.0.29 adds a system prompt parameter to run, a /clear command, alphabetized extensions, and a confirmation dialog for unsaved changes.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.29 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.29

    └──▷ TRY IT
    Inject a custom system prompt when running a Goose recipe or session non-interactively, useful for enforcing persona or constraints in CI pipelines.
    copy

    $ goose run --system "You are a security-focused code reviewer. Flag any use of eval() or exec()." --recipe review.yaml

    Clear accumulated context during a long session to reduce token usage or start a fresh task without restarting Goose.
    copy

    $ /clear

    • Adds --system (system prompt) parameter to the goose run command, enabling custom system-level instructions per run.
    • Adds /clear command to clear the Goose context mid-session.
    • Alphabetizes extensions listing in the Goose CLI for easier navigation.
    • Adds confirmation dialog for unsaved changes when closing the extension modal in the UI.
    • Adds a Help & Feedback section in App Settings.

    +2 more show less

    • Adds devcontainer support for development environments.
    • Ensures a newline is written at the end of file writes.

    7 details in Everything

61. v1.0.28 Jun 14, 2025 · release notes

    Goose v1.0.28 adds SageMaker TGI support, an LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. tool router, recipe settings, and desktop auto-update.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.28 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.28

    • Adds SageMaker TGI as a supported provider backend.
    • New LLM tool selector for the tool router, enabling AI-driven routing of tool calls.
    • Tool router now includes extension names in vector DB and search tool metadata.
    • Goose recipes now support configurable settings.
    • Adds auto-update functionality to the Goose desktop app, gated by an UPDATES_ENABLED flag.

    +2 more show less

    • CLI no longer halts startup when one or more MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. extensions fail to load.
    • Temporal disabled by default, with dynamic port selection when enabled.

    7 details in Everything

62. v1.0.26 Jun 9, 2025 · release notes

    Goose v1.0.26 adds session export, a temporal scheduler, lead/worker models, Snowflake provider, and Gemini 2.5 support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.26 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.26

    └──▷ TRY IT
    Set a custom context management strategy to control how Goose handles long conversations.
    copy

    $ goose config set GOOSE_CONTEXT_STRATEGY summarize

    Load recipes from a custom directory by setting GOOSE_RECIPE_PATH before running.
    copy

    $ GOOSE_RECIPE_PATH=/team/recipes goose run --recipe my-recipe

    • New export CLI command converts sessions to Markdown for sharing or archiving.
    • Adds a temporal scheduler with schedule creation, editing, pause/unpause, live tracking, and task cancellation for scheduled jobs.
    • New lead/worker model architecture enables multi-model agent workflows.
    • New goose web command provides a local web-based terminal alternative.
    • Adds Snowflake as a new LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider.

    +22 more show less

    • Adds support for Gemini 2.5 Flash Preview and Pro Preview models.
    • Adds Claude 4 to the Vertex AI provider dropdown with context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. limit support.
    • New GOOSE_CONTEXT_STRATEGY CLI config setting controls context management behavior.
    • CLI now shows active context lengthcontext lengthThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. during sessions.
    • New system prompt override support in goose-llm.
    • Supports deep link-based schedule creation with comprehensive extension support.
    • Adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server notification message handling.
    • Implements LanceDB vector-based tool selection for improved tool routing.
    • Adds MCP router skeleton for future tool routing.
    • Developer MCP extension falls back to .gitignore when no .gooseignore is present.
    • Adds screenshot paste support in the desktop UI.
    • Adds interactive session deletion from the CLI.
    • Adds retries with exponential backoff for the Databricks provider.
    • Supports configurable tool_params_max_length for large tool response handling.
    • New GOOSE_RECIPE_PATH environment variable for discovering recipes in custom paths.
    • GitHub Copilot provider gains streaming support (enabling gpt-4.1 and claude models).
    • Adds Speech MCP extension to the extensions directory.
    • Allows viewing and editing existing recipes in the Desktop UI.
    • CLI hint shown when input is empty (press Enter or Ctrl-J).
    • Google Drive MCP gains label support.
    • Desktop UI supports drag-and-drop file input.
    • Adds recipe_dir configuration for recipe discovery.

    27 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Tool router environment variable names now require a goose prefix (e.g., previously unprefixed vars are renamed).
63. v1.0.24 May 15, 2025 · release notes

    Goose v1.0.24 adds Venice.ai LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. support, tool-loop protection, directory tracking, and recipe parameters

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.24 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.24

    └──▷ USE IT
    Pin the planner to a specific model and provider without using environment variables
    yaml copy

    GOOSE_PLANNER_MODEL: gpt-4o
    GOOSE_PLANNER_PROVIDER: openai

    • Adds Venice.ai as a private open-source LLM provider option
    • Adds tool repetition monitoring to detect and prevent infinite tool-call loops
    • Adds goose run flag to execute sessions with no persistence to disk
    • Adds directory tracking so projects can be resumed by location
    • Adds configurable parameters in recipe files, enabling parameterized automation workflows

    +6 more show less

    • Adds recipe explanation display in the UI so users can see what a recipe does before running it
    • Supports setting GOOSE_PLANNER_MODEL and GOOSE_PLANNER_PROVIDER in config.yaml
    • Adds token usage progress info alert in the UI
    • Defaults to concise tools mode in the GUIGUIA graphical user interface, controls such as windows, buttons and menus. It lets people use software visually rather than entering commands. for reduced noise
    • Adds providerConfig parameter to exposed goose-llm library functions
    • Auto-generates Kotlin bindings for goose-llm via uniffi-rs proc macros

    11 details in Everything

64. v1.0.23 May 8, 2025 · release notes

    Goose v1.0.23 adds GitHub Copilot provider, a /summarize slash command, session search, and full --resume message loading.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.23 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.23

    └──▷ TRY IT
    Trigger an in-session context summarization to free up token space without ending the session.
    copy

    $ /summarize

    • Adds GitHub Copilot as a supported LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider.
    • New /summarize slash command lets users trigger context summarization directly from the CLI.
    • Adds a summarize-on-command button for context management in the UI.
    • Loads all previous messages when --resume is passed, enabling full session continuity.
    • Adds search to the sessions list for faster session navigation.

    +6 more show less

    • Adds a menu item to check Goose Desktop's version on Linux.
    • Expanded ToolCall options for greater tool invocation flexibility.
    • Removes outdated truncation of tool descriptions in OpenAI and Databricks providers, enabling richer tool context.
    • Adds model token limits fallback matching backend logic for more reliable context handling.
    • Adds recipe welcome message support.
    • Allowlist blocks with SSESSEServer-Sent Events, a browser API and HTTP-based protocol where a server pushes a stream of text events to a client over a single long-lived connection, giving tools a lightweight alternative to WebSockets for one-way real-time updates. downgraded to warning level for less disruptive handling.

    11 details in Everything

65. v1.0.22 May 6, 2025 · release notes

    Goose v1.0.22 adds GitHub-based recipe retrieval, file drag-and-drop, a global hotkey, runtime metrics, and Mac keyboard shortcuts.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.22 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.22

    └──▷ TRY IT
    Run a recipe in headless/unattended mode without supplying an interactive prompt, suitable for CI pipelines.
    copy

    $ goose run --recipe <recipe>.yaml --headless

    Run benchmarks and write results to a custom directory for easier artifact collection in CI.
    copy

    $ goose bench --output-dir <path>

    • Supports configurable retrieval of recipes via GitHub, enabling teams to share and version-control recipe libraries centrally.
    • Enables drag-and-drop of files directly into the Goose desktop window for faster context attachment.
    • Adds a global hotkey to bring up Goose from anywhere on the desktop.
    • Exposes runtime metrics in completion responses, giving visibility into model execution performance.
    • Adds Mac keyboard shortcuts to the desktop UI.

    +6 more show less

    • Allows recipes to run in headless mode without requiring a prompt, enabling unattended automation.
    • Adds Azure credential chain logging to aid in diagnosing Azure auth configuration issues.
    • Adds a configurable output directory for benchmark runs.
    • Combines create and reply comment tools into a single unified tool.
    • Stores global and local chat history in localStorage for persistence across sessions.
    • Handles context summarization feedback directly in the UI.

    11 details in Everything

66. v1.0.21 Apr 29, 2025 · release notes

    Goose v1.0.21 adds context-limit options, session ordering, model-switching without reinit, and spellcheck in the desktop app.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.21 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.21

    • Adds ascending/descending ordering support for goose session list.
    • Enables switching or adding a model without reinitializing extensions, reducing disruption mid-session.
    • Presents interactive options to the user when context lengthcontext lengthThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. is exceeded, with a new context management modal.
    • Adds full spellcheck and autocorrection support for editable content in the desktop app.
    • Adds context limits for the latest Gemini and GPT models.

    +7 more show less

    • Adds new Google model support in the provider configuration.
    • Adds a goose-llm crate for use by the Goose service.
    • Adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. router extension disable capability.
    • Reduces the number of Google Drive tools exposed, narrowing the tool surface.
    • Adds bottom bar text truncation with tooltips in the UI.
    • Adds session deletion preview showing the to-be-removed session before confirming.
    • Unifies the permission flow for enabling extensions.

    12 details in Everything

67. v1.0.20 Apr 23, 2025 · release notes

    Goose v1.0.20 adds session removal, a GUIGUIA graphical user interface, controls such as windows, buttons and menus. It lets people use software visually rather than entering commands. deeplink extension warning, and a bumped tool limit with updated alerts.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.20 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.20

    • Adds ability to remove sessions via new remove session capacity.
    • Adds GUI warning when installing an extension from a deeplink.
    • Bumps tool limit and updates associated alerts.

    3 details in Everything

68. v1.0.19 Apr 19, 2025 · release notes

    Goose v1.0.19 adds o4-mini and GPT-4.1 support, persistent CLI themes, custom OpenAI headers, and an FFI for the Rust library.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.19 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.19

    • Adds support for the o4-mini model.
    • Adds per-model prompts for better GPT-4.1 support.
    • Supports OPENAI_CUSTOM_HEADERS being set in the Goose config file.
    • Builds a prototype FFI for the Goose Rust library, enabling external language bindings.
    • GOOSE_CLI_THEME now persists to the config file and applies to future sessions automatically.

    +2 more show less

    • Adds a permissions page in the bottom menu bar of the chat UI.
    • Adds token and tools alerts in the chat bottom menu bar.

    7 details in Everything

69. v1.0.18 Apr 16, 2025 · release notes

    Goose v1.0.18 adds tool-level permission control, recipes/custom agents, SSESSEServer-Sent Events, a browser API and HTTP-based protocol where a server pushes a stream of text events to a client over a single long-lived connection, giving tools a lightweight alternative to WebSockets for one-way real-time updates. extensions, file-based secrets, and GCP Vertex AI support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.18 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.18

    └──▷ TRY IT
    Set model temperature without modifying config — useful for scripting creative or deterministic runs.
    copy

    $ GOOSE_TEMPERATURE=0.2 goose session

    Point Goose at a self-hosted or proxy Anthropic endpoint instead of the default API.
    copy

    $ ANTHROPIC_HOST=https://my-anthropic-proxy.internal goose session

    • Adds tool-level permission control in both CLI and UI, letting practitioners approve or deny individual tools before execution.
    • Introduces recipes — custom Goose agent configurations that can be shared via URL deeplinks.
    • Adds a recipe editor UI for building and sharing custom agent configurations.
    • Supports SSE (Server-Sent Events) extensions via scheme URI, enabling remote extension connectivity.
    • Adds support for generic GCP Vertex AI Claude and Gemini models as a provider.

    +19 more show less

    • Adds ANTHROPIC_HOST configuration option for Anthropic in CLI and UI, enabling custom endpoint targeting.
    • Adds a temperature environment variable for controlling model temperature without code changes.
    • Supports file-based secrets for extension configuration.
    • Adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. router extension discovery and install tool, enabling in-session extension installation.
    • Enables Settings V2 UI with migration logic from Settings V1.
    • Parallelizes extension startup with toast-based error reporting for faster initialization.
    • Adds --with-remote-extension flag for launching Goose with a remote extension.
    • Adds exponential backoff to the Bedrock provider for improved reliability.
    • Supports NO_COLOR environment variable when PrettyPrinter is used.
    • Adds permission field to the list tools API response.
    • Enables frontend tools execution from the UI.
    • Adds full URL extraction support for Google Drive.
    • Auto-initializes goose config.yaml on first run.
    • Enables spellcheck in the chat input.
    • Adds a system theme option to the UI.
    • Adds tooltips throughout the UI.
    • Adds debounce for search and chat text inputs to improve typing responsiveness in large sessions.
    • Accumulates tokens across a session file for accurate total token usage reporting.
    • Adds Google Drive URI verification and example instructions.

    24 details in Everything

70. v1.0.17 Apr 1, 2025 · release notes

    Goose v1.0.17 adds session sharing, chat search, prompt library, VSCode server support, and input/output token tracking.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.17 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.17

    └──▷ HOW TO FIND IT
    Insert a new line mid-prompt without submitting — useful when composing multi-line instructions in the chat UI.
    📍Option + Enter
    • Adds session sharing from the UI, with pre-configuration via GOOSE_BASE_URL_SHARE environment variable and additional metadata in the shared session view.
    • Adds search functionality to the chat view.
    • Adds a Prompt Library to the UI.
    • Adds the VSCode server to the extensions list.
    • Adds input and output token tracking to SessionMetadata.

    +10 more show less

    • Adds tool annotations for built-in tools.
    • Adds a named feature flag system for Settings V2.
    • Adds a timeout field to the Settings V2 modal.
    • Adds an additive entry in the Settings V2 model selector.
    • Enables Option+Enter keyboard shortcut to insert a new line in chat input.
    • Adds env-var editor when adding or editing extensions.
    • Adds an allowlist option for the goosed daemon.
    • Adds copy button for extensions and makes extension builtins available at any time.
    • Adds Playwright end-to-end testing setup.
    • Turns the Goose entrypoint into a library function for programmatic use.

    15 details in Everything

71. v1.0.16 Mar 26, 2025 · release notes

    Goose v1.0.16 adds PDF reading, shareable bots, a /plan command, Databricks+Claude 3.7, custom OpenAI headers, and Java/JDK MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.16 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.16

    └──▷ TRY IT
    Generate a structured plan for a complex task before letting Goose execute it — useful for long-horizon or multi-step agentic sessions.
    copy

    $ goose /plan "Set up a Python project with CI, linting, and tests"

    • Adds --plan command in CLI to invoke a reasoner with a plan system prompt.
    • Adds PDF reader capability, enabling Goose to ingest PDF files directly.
    • Supports custom HTTP headers for the OpenAI provider.
    • Exports Azure API version as a configurable parameter.
    • Adds Databricks model format with support for Claude 3.7 with extended thinkingextended thinkingIntermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens..

    +16 more show less

    • Allows setting the OpenAI request timeout from config.
    • Adds retry logic for the Google provider.
    • Adds Java/JDK support for MCP servers running via extensions.
    • Adds a noop tool to the tool shim for testing/passthrough scenarios.
    • Introduces shareable Goose bots for sharing configured assistant profiles.
    • Improves extension security via an environment variable denylist.
    • Supports deep-link extension installs directly from settings.
    • Expands Google Drive integration with comments, replies, folders, shortcuts, and file moves.
    • Adds read and write support for Google Sheets.
    • Adds a description field to the extension configuration and modal.
    • Enables extensions to be turned on automatically at startup.
    • Adds a models dropdown to the UI for switching models inline.
    • Adds a new configure provider flow in the UI.
    • Adds in-app log viewer to the desktop UI.
    • Introduces new toast notifications for system events.
    • Surfaces StdioProcessError messages in the UI for better MCP debugging.

    21 details in Everything

72. v1.0.15 Mar 18, 2025 · release notes

    Goose v1.0.15 adds Google Drive write/comment support, AWS Bedrock in desktop, and a new CLI workflow builder.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.15 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.15

    • Adds Google Drive write tools and comment-read capability, enabling agents to read comments and write files in Drive.
    • Adds image resizing logic for Google Drive with Content::Image support for richer document handling.
    • Adds PKCEPKCEProof Key for Code Exchange, an OAuth safeguard that binds an authorization code to the client that requested it. An intercepted code cannot be redeemed without the client's matching verifier.-based OAuth2OAuth2An open authorization framework (RFC 6749) that lets a service grant limited access to a user's resources on another service without exposing credentials, giving cyber tools a standard way to authenticate against third-party APIs. flow for Google Drive with generic token storage, improving authentication security.
    • Unifies Google Drive read/write scope across all commands so a single token covers all operations.
    • Adds basic AWS Bedrock support to the Goose desktop app as a new model provider.

    +3 more show less

    • Introduces a CLI workflow build command for constructing and running agent workflows from the command line.
    • Adds a new extensions modal in the UI with auto-start on add and reorganized extensions settings.
    • Adds default metrics for core evaluations to standardize eval measurement.

    8 details in Everything

73. v1.0.14 Mar 13, 2025 · release notes

    Goose v1.0.14 adds session listing, write-approve mode, Google Sheets, Ollama tool shim, and stdin CLI support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.14 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.14

    └──▷ TRY IT
    List all past Goose sessions to review or resume previous work.
    copy

    $ goose session list

    Pipe a prompt directly into Goose from another command without an interactive session.
    copy

    $ echo 'Summarize the files in this directory' | goose -i -

    • Adds session list command to the CLI for viewing past sessions.
    • Adds Google Sheets support in the Google Drive built-in MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server.
    • Adds Ollama tool shim, enabling tool-call support for Ollama-backed models.
    • Supports stdin input in the CLI via -i - or with no arguments for piped/scripted workflows.
    • Moves Google Drive credentials into the system keychain with an optional fallback.

    +2 more show less

    • Extensions now read from config and load in the background with a visible pending state.
    • Adds a copy-to-clipboard button for error messages in toast notifications.

    7 details in Everything

74. v1.0.13 Mar 12, 2025 · release notes

    Goose v1.0.13 adds smart-approve mode, memory condensation, PDF/Office file support, a --debug flag, and a full benchmarking framework.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.13 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.13

    └──▷ TRY IT
    Capture verbose agent activity to diagnose tool call failures or unexpected model behaviour during a session.
    copy

    $ goose session --debug

    Run a one-shot task with full debug output to trace exactly what the agent does at each step.
    copy

    $ goose run --debug --instructions 'Summarise the quarterly report' report.pdf

    • Enables smart-approve mode by default, allowing the agent to automatically approve low-risk tool calls without user interruption.
    • Adds memory condensation for longer-context sessions, keeping conversations coherent beyond normal context limits.
    • Supports reading PDFs natively, plus .doc/.xls and simple HTML files, broadening the range of documents Goose can process.
    • Adds --debug flag to goose session and goose run for verbose diagnostic output during CLI sessions.
    • Implements a tool permission store for persistent tracking of approved/denied tool permissions across sessions.

    +12 more show less

    • Introduces the Goose Bench framework for functional and regression testing of agent behaviour, including parallel processing in approve mode and eval result writing to an eval directory.
    • Adds an image tool to the developer MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface., enabling Goose to capture and process screenshots (including macOS screenshots).
    • Adds Ctrl/Cmd + ↑/↓ keyboard shortcut to navigate message history in the CLI.
    • Supports Databricks OAuth refresh tokensrefresh tokensA credential used to obtain a new access token without signing in again. It is sent to the authorization service rather than used directly to call the protected API. via the Databricks API integration.
    • Adds copy message content action to the UI for quick clipboard access to any message.
    • Adds mode completion (tab-complete for modes) in the CLI.
    • Updates config endpoints for use with providers, improving provider configuration flexibility.
    • Enables building Goose inside a Docker container for portable, reproducible deployments.
    • Adds auto-including directories in binary/bench working-dir for the bench framework.
    • Splits required_extensions in bench configuration into builtin and external categories for finer control.
    • Adds UI setting configuration panel for managing settings directly from the desktop interface.
    • Retains session state through view changes in the desktop UI.

    17 details in Everything

75. v1.0.12 Mar 4, 2025 · release notes

    Goose v1.0.12 adds global prompt templates at startup, form injection for provider modals, and extension management in Settings v2.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.12 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.12

    • Loads global prompts at startup using the minijinja templating engine, enabling reusable prompt templates across sessions.
    • Introduces form injection for provider modals in the UI, streamlining provider configuration.
    • Adds extension add and edit functionality in the Settings v2 UI.

    3 details in Everything

76. v1.0.11 Mar 4, 2025 · release notes

    Goose v1.0.11 adds GCP Vertex AI, Claude 3.7 Sonnet with extended thinkingextended thinkingIntermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens., sessions API, .gooseignore support, and tab-based slash command completion.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.11 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.11

    └──▷ TRY IT
    Use tab completion for slash commands and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. prompts in an interactive Goose CLI session.
    copy

    $ goose
    # Inside the session, type '/' then press Tab to browse available slash commands and MCP prompts

    • Adds GCP Vertex AI as a supported provider platform.
    • Adds Claude 3.7 Sonnet with extended thinking via Anthropic provider.
    • New sessions API enabling viewing and resuming previous sessions.
    • Adds .gooseignore support to restrict Goose's access to specified files or directories.
    • Adds tab-based slash command completion and prompt info completion in the CLI.

    +16 more show less

    • Adds MCP prompt support via slash commands in the CLI.
    • Implements global command history storage.
    • Supports customizing extension timeout.
    • Supports arbitrary path for sessions in the CLI.
    • Adds goose mode to the chat window with full UI support.
    • Adds context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. limit for OpenAI reasoning models.
    • Adds ability for users to turn off smart approve.
    • Builds native Intel Mac app.
    • Adds view and edit support for .goosehints file in the desktop UI.
    • Moves configure goosehints into the MoreMenu for easier access.
    • Adds extensions install link generator.
    • Adds a tutorial built-in extension for the app.
    • Improves CLI default behavior when no command is provided.
    • Installs Python 3.10 in the uvx shim before uv for better compatibility.
    • Keyboard shortcut Command/Ctrl + Comma navigates to Settings view in the desktop UI.
    • Improves the allow-tool UI.

    21 details in Everything

77. v1.0.10 Feb 26, 2025 · release notes

    Goose v1.0.10 adds goose update, interactive post-run mode, permission prompts before tool calls, and a configurable OpenAI base path.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.10 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.10

    └──▷ TRY IT
    Keep your Goose CLI up to date without reinstalling manually.
    copy

    $ goose update

    • New goose update command lets users update the Goose CLI to the latest version in place.
    • Interactive mode after run: Goose can drop into an interactive session at the end of a non-interactive run.
    • Permission prompt before each tool call, giving users explicit control over tool execution.
    • Read-only tools now skip the approval prompt automatically, reducing friction for safe operations.
    • Supports setting a custom OpenAI base path, enabling use of OpenAI-compatible endpoints.

    +2 more show less

    • Settings v2 scaffolding and ConfigProvider integration lay groundwork for a new configuration UI.
    • Alpha providers grid refactor in the UI.

    7 details in Everything

78. v1.0.9 Feb 21, 2025 · release notes

    Goose v1.0.9 adds an experiment manager for feature flags and consolidates settings config.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.9 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.9

    • Adds an experiment manager to control whether individual features are enabled.
    • Consolidates Goose settings into a unified config.
    • Adds a default value for the OpenAI host configuration.

    3 details in Everything

79. v1.0.8 Feb 20, 2025 · release notes

    Goose v1.0.8 adds a new info command, customizable system prompt templates, OpenAI org/project support, reasoning effort control, and configurable OpenAI host.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.8 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.8

    └──▷ TRY IT
    Quickly inspect which config file and directories Goose is using on your machine.
    copy

    $ goose info

    • New goose info command displays all directories in use and current configuration at a glance.
    • Adds a fully customizable system prompt template mode, giving users complete control over the system prompt.
    • Supports configuring a custom OpenAI host, enabling use with OpenAI-compatible endpoints.
    • Adds organization and project fields for the OpenAI provider configuration.
    • Supports OpenAI reasoning effort configuration for O1/O3 models.

    +2 more show less

    • Google Drive search tool gains corpora and pageSize parameters for finer-grained search control.
    • Propagates external_model_message upward on errors, surfacing upstream model error details.

    7 details in Everything

80. v1.0.7 Feb 14, 2025 · release notes

    Goose v1.0.7 adds Linux computer control support and verbose tool output.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.7 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.7

    • Adds Linux computer control capability, enabling automated desktop interaction on Linux systems.
    • Introduces verbose tool output mode for more detailed visibility into tool execution.

    2 details in Everything

81. v1.0.6 Feb 13, 2025 · release notes

    Goose v1.0.6 adds configurable tool output, image handling, extension management, VertexAI/Claude support, and experimental Windows CLI.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.6 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.6

    • Adds goose configure flow for adjusting the amount of tool output shown in sessions.
    • Adds support for handling images mentioned in messages.
    • Enables removing installed extensions via the CLI.
    • Adds server-side endpoints for config management via goose-server.
    • Supports extending the system prompt with custom content.

    +6 more show less

    • System prompt now includes the current date and response formatting instructions.
    • Simplifies CLI session management.
    • Ports MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. prompts into the developer extension.
    • Follows XDGXDGA set of specifications maintained by freedesktop.org that standardise where Linux applications store config, data, and cache files, letting tools respect user-defined paths instead of hardcoding locations. spec on Linux/macOS and uses Windows known folders for config and log paths.
    • Adds experimental Windows support for the Goose CLI.
    • Supports modifying AZURE_OPENAI_API_VERSION via configuration.

    11 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Config and log paths on Linux/macOS now follow the XDG spec; existing setups storing config or logs in non-XDG locations may need to migrate files to the new paths.
82. v1.0.5 Feb 7, 2025 · release notes

    Goose v1.0.5 adds AWS Bedrock provider, updated Gemini models, multi-value CLI extension options, and visible session aliases.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.5 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.5

    • Adds AWS Bedrock as a new LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider.
    • Updates available Gemini models.
    • Supports multiple values in a single CLI option when adding extensions.
    • Makes the CLI session command alias visible in help output.
    • Replaces the prompt for Ollama smaller models for improved compatibility.

    +2 more show less

    • Respects the terminal emulator's base colors for better theme integration.
    • Adds GitHub project links on extension cards in the UI.

    7 details in Everything

83. v1.0.4 Jan 31, 2025 · release notes

    Goose v1.0.4 adds Azure OpenAI as a provider and configures Electron to open links in new windows.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.4 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.4

    • Adds Azure OpenAI as a supported LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider.
    • Configures Electron app to open external links in a new window instead of in-app.

    2 details in Everything

84. v1.0.3 Jan 31, 2025 · release notes

    Goose v1.0.3 adds a CONFIGURE=false install option and Ollama host configuration support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.3 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v1.0.3

    └──▷ TRY IT
    Install Goose non-interactively in CI or automated environments without being prompted to configure a provider.
    copy

    $ CONFIGURE=false curl -fsSL https://github.com/block/goose/releases/latest/download/install.sh | bash

    • Adds CONFIGURE=false option to the install script to skip interactive configuration during automated or scripted installs.
    • Supports setting a custom Ollama host, enabling Goose to connect to remote or non-default Ollama instances.
    • Updates the UI to expose Ollama host configuration directly in the interface.

    3 details in Everything

85. v0.9.9 Nov 5, 2024 · release notes

    Goose v0.9.9 adds a browser toolkit, tracing support, a license checker, and defaults to the synopsis toolkit.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.9.9 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v0.9.9

    • Adds a browser toolkit, enabling Goose to interact with web browsers as a tool.
    • Adds tracing option to the run command, grouping traces under the session name for observability.
    • Adds a license checker for auditing dependency licenses.
    • Defaults Goose to the synopsis toolkit out of the box.
    • Adds support for optional parameters in JSON schema conversion in exchange, broadening tool-call compatibility.

    +2 more show less

    • Adds a list moderators capability.
    • Includes the new Anthropic model in recommended config.

    7 details in Everything

86. v0.9.6 Oct 18, 2024 · release notes

    Goose v0.9.6 adds web browsing, session resumption, and a synopsis-based core loop.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.9.6 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v0.9.6

    • Adds web browsing capability to the core Goose loop.
    • Enables resuming a previous session with a dedicated run flag.
    • Introduces a synopsis approach for the core Goose loop to improve context handling.

    3 details in Everything

87. v0.9.5 Oct 15, 2024 · release notes

    Goose v0.9.5 adds Google vision support and updates the default Ollama model to qwen2.5.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.9.5 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v0.9.5

    • Adds vision support for Google provider, enabling image-based interactions.
    • Updates the default Ollama model from mistral-nemo to qwen2.5.

    2 details in Everything

88. v0.9.4 Oct 10, 2024 · release notes

    Goose v0.9.4 adds Groq and o1 model support, a new run command, provider listing, keychain API key storage, and global goosehints.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.9.4 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v0.9.4

    └──▷ TRY IT
    Check which LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. providers are configured and available before starting a session.
    copy

    $ goose providers list

    • Adds Groq as a supported LLM provider.
    • Supports OpenAI o1-preview and o1-mini deep thinking reasoner models.
    • New goose run command for non-interactive session execution.
    • New goose providers list command to enumerate available providers.
    • Saves API keys to the system keychain automatically for users.

    +3 more show less

    • Adds session management guards to protect active sessions.
    • Adds support for a global user-level goosehints file for persistent instructions.
    • Adds local Langfuse tracing option for observability.

    8 details in Everything

89. v0.9.3 Sep 26, 2024 · release notes

    Goose v0.9.3 adds shell completions, Jira toolkit, sandbox mode, markdown plans, and session auto-save.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.9.3 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v0.9.3

    └──▷ TRY IT
    Generate and install shell completions so you can tab-complete Goose subcommands and flags.
    copy

    $ goose shell-completions

    • Adds shell-completions subcommand for generating shell tab-completion scripts.
    • Introduces a Jira toolkit for interacting with Jira from within Goose sessions.
    • Supports running Goose inside a Docker-style sandbox for isolated execution.
    • Enables support for markdown-formatted plans.
    • Auto-saves sessions before the next user input to preserve context.

    +3 more show less

    • Tracks cost and token usage in the log file for session auditing.
    • Adds --version option to display the current Goose version.
    • Gives commands the ability to execute logic, expanding extensibility of the command system.

    8 details in Everything

90. v0.9.0 Sep 10, 2024 · release notes

    Goose v0.9.0 adds Jinja-templated goosehints, per-repo hints, toolkit listing, and regex-based dangerous command detection.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.9.0 https://github.com/block/goose.git
    # already have the repo? check out this version:
    $ git checkout v0.9.0

    • Supports Jinja templating in goosehints files for dynamic hint configuration.
    • Adds per-repository hints so projects can ship their own Goose context alongside code.
    • New command to show all available toolkits at a glance.
    • Adds regex-based checks to detect and flag potentially dangerous shell commands.

    4 details in Everything