Goose
Also known as: codename goose, Block goose, aaif-goose
Vendor neutral open source AI agent under the Linux Foundation's Agentic AI Foundation, running locally as desktop app, CLI or API, with MCP extensions, recipes and subagents, a scheduler, and 30+ model providers on your own keys.
Goose, originally released as codename goose, is an open source, general purpose AI agent that runs on the user's own machine. Created by Block's open-source team and now governed as a vendor-neutral project under the Agentic AI Foundation at the Linux Foundation, it is built in Rust and distributed under a permissive Apache license. Although its first and best-known use is software engineering, Goose is designed to take on broader work too, including research, writing, automation, and data analysis.
Given a high level goal, Goose breaks the task into steps and carries them out. It reads and writes files, runs code and tests, installs dependencies, executes terminal commands, refines its own output and calls external APIs, iterating until the job is done. It works in the developer's actual environment and can build, run and fix a project end to end.
Goose runs locally, so code and data stay on the user's machine. People use it through a native desktop app for macOS, Linux and Windows, a full command line interface for terminal and scripting workflows, or an API for embedding it in their own tools.
Goose works with many model providers, including Anthropic, OpenAI, Google and local models through Ollama. Its extensions use the Model Context Protocol, and Goose was one of the first public MCP clients. MCP lets the agent connect to the systems where data and tools live, from content repositories to business applications and development environments. It can discover new systems on the fly, and the community keeps expanding what it can do by building new integrations.
Because it is open source, anyone can extend Goose, build a custom interface, or ship their own preconfigured distribution.
Vendor details
Canonical URL
https://goose-docs.ai
Category
Coding agent
Funding status
Vendor neutral open source project, Apache 2.0, written in Rust. Created inside Block, open sourced as codename goose in January 2025, contributed to the Linux Foundation's Agentic AI Foundation as a founding project in December 2025 alongside Anthropic's MCP and OpenAI's AGENTS.md, with repository and governance formally transferred from Block to the AAIF in April 2026. Roughly 52,000 GitHub stars and over 500 contributors; Block remains an active maintainer. AAIF platinum members include AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI.
Company status
independent
Use cases & customers
Target customers
Deployment options
Integrations
Extensions are MCP servers, with a catalog of more than 70 covering SaaS applications, data sources and development environments, and the agent can discover new systems on the fly. Goose was one of the first public MCP clients. Every tool is individually configurable and can be disabled, including native ones. A goose-server REST API with an OpenAPI specification and generated TypeScript SDK supports embedding. A Provider trait allows custom LLM integrations across 30+ providers, and white labeled distributions with preconfigured providers and branding are a supported build target.
In practice
You want an autonomous agent but your proprietary code can't leave your laptop. Goose runs locally, reading and writing files, running tests, and executing commands in your own environment, so your data never leaves your machine.
You'd rather delegate a whole task than get line-by-line suggestions. You give Goose a high-level goal and it breaks it into steps, writing the code, running it, catching bugs, and fixing them until it's done.
You don't want to be locked to one model or one set of tools. Goose works with providers like Anthropic, OpenAI, and local models through Ollama, and connects to your systems through the Model Context Protocol.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
12.0 / 14 capabilities · 86%
| Integrations & Tool Calling | Full |
|---|---|
|
The agent reads and writes files, executes shell commands, builds and runs software, runs tests and calls external APIs in the developer's own environment, with more than 70 MCP extensions connecting SaaS applications, data sources and development environments; tools are individually configurable and can be turned off, including native ones, and MCP Apps provide a UI proxy surface. Sourcegoose-docs.ai extensions docs and github.com/aaif-goose/gooseread 2026-08-30 |
|
| Workflow Orchestration | Full |
|
Recipes package instructions, prompt, extensions, parameters and sub_recipes into a reusable shareable unit defined by a YAML or JSON schema, with template variables filled at runtime. Larger workflows are composed from sub-recipes, subagents run sub-recipes in isolated execution contexts and can run concurrently, and a scheduler runs recipes on cron or set frequencies unattended. Sourcegoose-docs.ai/docs/guides/recipes and goose-docs.ai scheduling docsread 2026-08-30 |
|
| Knowledge Grounding & RAG | Partial |
|
The agent works in the developer's actual environment, reading files and running commands directly, and reaches external content repositories, business applications and data sources through MCP extensions, with the ability to discover new systems on the fly. Project guidance follows the AGENTS.md standard, an AAIF sibling project. Goose does not describe codebase indexing, embeddings or a retrieval architecture. Sourcegoose-docs.ai and aaif.io/projects/gooseread 2026-08-30 |
|
| Human Oversight & Guardrails | Full |
|
Permission modes set how much goose may do alone. Manual Approval asks for confirmation before using any tool or extension, Smart Approval approves low risk actions and flags the rest for approval, Chat Only allows no tool use, and granular per tool permissions refine both. Completely Autonomous is the default, though, so approval applies only once a stricter mode is chosen. Sourcegoose-docs.ai/docs/guides/managing-tools/goose-permissionsread 2026-09-29 |
|
| Security, Identity & Governance | Partial |
|
Goose secures work through its architecture and configuration. The agent runs locally under the user's own credentials with bring your own keys, tools can be disabled one by one and scoped per recipe and subagent, and code need never leave the machine. There is no hosted service, no accounts and no organization layer, so there is no SSO, RBAC, audit logging or attestation, and none is claimed. The project is vendor neutral under the Agentic AI Foundation, which publishes a project health score for it. Sourcegoose-docs.ai and aaif.io/projects/gooseread 2026-08-30 |
|
| Observability & Auditability | Full |
|
Every session is kept as a record in a local SQLite database holding session metadata and the full conversation of messages and tool interactions, beside command history and system log files, all stored on the user's machine and never sent to external servers. Past sessions are browsable and resumable in the CLI and desktop app, so the user holds the execution history, though there is no organization level export. Sourcegoose-docs.ai/docs/guides/logsread 2026-09-29 |
|
| Memory & State Persistence | Partial |
|
The built in Memory extension stores what the user teaches goose (commands, snippets, preferences) as files per project or globally, loads them into every session and exposes remember, retrieve and remove tools, and sessions are resumable. The store is written at the user's instruction. Sourcegoose-docs.ai/docs/mcp/memory-mcp and docs/guides/logsread 2026-09-29 |
|
| Deployment & Data Residency | Full |
|
Goose runs entirely on the developer's own machine as an Apache 2.0 licensed Rust binary, as a desktop app for macOS, Linux and Windows or a CLI, so code and data stay local. With an Ollama or LM Studio provider and a locally hosted model it operates fully offline with no API keys and no external network calls, which suits air gapped environments, and an Inference Mesh supports local models. Sourcegoose-docs.ai and github.com/aaif-goose/gooseread 2026-08-30 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
A Recipe Cookbook publishes ready to use recipes for common development scenarios, shareable by URL and launchable with a single click, alongside an extension catalog of more than 70 MCP extensions installable into the agent and a skills system for role based specialization. Custom slash commands can be bound to recipes in any chat session. Sourcegoose-docs.ai/docs/guides/recipes and goose-docs.ai extensions catalogread 2026-08-30 |
|
| Triggers & Channel Coverage | Full |
|
Invocation spans a native desktop application on macOS, Linux and Windows, a full command line interface for terminal and scripting workflows, and a REST API with an OpenAPI specification and generated TypeScript SDK for embedding; a scheduler runs recipes at a set frequency or on a custom cron schedule unattended, and a gateway provides remote access. Sourcegoose-docs.ai scheduling and CLI docs, and the goose-server API referenceread 2026-08-30 |
|
| Model Flexibility & Routing | Full |
|
More than 30 LLM providers are supported, including Anthropic, OpenAI, Google, Groq, Mistral, Cohere, Amazon Bedrock and local inference through Ollama and LM Studio, configurable per session and per recipe. A Provider trait allows custom provider integrations and an Inference Mesh serves local models. The project is bring your own key throughout, and multi model use lets different models review each other's work. Sourcegoose-docs.ai provider docs and goose-docs.ai/docs/guidesread 2026-08-30 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
Goose was one of the first public MCP clients, and its Extensions are MCP servers, with a catalog of more than 70 and the ability to discover new systems on the fly. A goose-server REST API with an OpenAPI specification and generated TypeScript SDK supports embedding, custom MCP servers and tools can be built, a Provider trait allows custom LLM integrations, and white labeled distributions with preconfigured providers and branding are a supported build target. The whole project is Apache 2.0 in Rust at github.com/aaif-goose/goose. Sourcegoose-docs.ai extensions and API docsread 2026-08-30 |
|
| Testing, Debugging & Optimization | Full |
|
The repository carries an evals/harbor directory for running goose against benchmark tasks, so a customer can evaluate the agent's behavior with the project's own harness, and recipes support structured response schemas and retries for deterministic checks. Sourcegithub.com/aaif-goose/gooseread 2026-09-29 |
|
| Browser & Computer Use | Partial |
|
The built in Computer Controller extension automates everyday computer tasks, controlling applications and system settings through macOS UI automation via the third party Peekaboo CLI and scraping the web. Interface control comes from that outside engine, and browser control otherwise comes from catalog extensions such as Playwright, Selenium and Browserbase that the user installs. Sourcegoose-docs.ai/docs/mcp/computer-controller-mcpread 2026-09-29 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Subagents that failed or were canceled can now be resumed in Goose v1.56.0 instead of started over. The UI links to each subagent's session, which opens as view only, so users can see what that subagent did. A new setting caps how many subagents run at the same time.
Bears on: Workflow orchestration
View sourceRelease 1.55.0 of goose binds subagents to their parent's hooks, goose mode and tool approvals, so work the agent delegates runs under the same checks as the parent session. The agent also stops inferring or steering while a tool approval is pending. The same release removes background subagents and the orchestrator and deletes the legacy agent loop.
Bears on: Workflow orchestration
View sourceGoose version 1.45.0 introduces several new capabilities, including support for the latest Gemini models and a configurable GOOSE_DOCS_ROOT environment variable for air-gapped documentation access. The release also adds structured summary output, the ability to disable built-in skills, and resolves a security vulnerability by patching the Nostr dependency.
Bears on: Security / enterprise
View sourcePricing
Free and open source.
No subscription. The only cost is model usage, billed by the provider behind the user's own API key, or nothing for a local model.
What is public
Goose is an open source AI agent, created at Block and now governed under the Agentic AI Foundation, that runs locally and connects to the LLM provider the user picks. It is free to use. Users supply and pay for the model through their own API key, or run a local model.
Billing mechanics
No vendor subscription. Cost is whatever the chosen model provider charges for the API usage Goose generates, or zero for a local model.
Variable cost rationale
Goose itself is free and meters nothing. Model usage on the user's own key is pay as you go, so spend tracks the provider's API rates and how much Goose runs. Running a local model brings the model cost to zero.
Sales call required
No, self serve available
Free / trial
Free and open source under Apache 2.0, running locally
Lowest paid plan
No paid plan
Key ambiguities
Total cost depends entirely on which model Goose is pointed at and how much it runs. There is no Goose price.
Missing data
No vendor pricing exists. Model provider API spend is the only cost, and it differs for every user.
Related vendors
- Cognition — Maker of Devin, an autonomous AI software engineer
- 10Web — Agentic website platform whose AI agents build, host and manage…
- AgentUI — Managed AI app builder for operations teams that generates and hosts…
- Aider — Open source, model agnostic terminal coding agent that edits your…
- Amp — Amp is a coding agent from Amp Frontier Corporation that runs each…
- Anthropic Claude Code — Anthropic's agentic coding system across terminal, desktop, IDE, web…
Alternatives to Goose
The closest documented capability profiles to Goose among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Cline12.0 / 14Fuller documented coverage on Browser & Computer Use
- JetBrains AI12.0 / 14Fuller documented coverage on Browser & Computer Use
- Capy12.5 / 14Fuller documented coverage on Security, Identity & Governance and Browser & Computer Use
- Cursor12.5 / 14Fuller documented coverage on Security, Identity & Governance and Browser & Computer Use
- Google Antigravity12.5 / 14Fuller documented coverage on Security, Identity & Governance and Browser & Computer Use
- OpenHands13.5 / 14Fuller documented coverage on Security, Identity & Governance and Memory & State Persistence
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded