Goose Documentation — Open-Source, Extensible AI Agent
by Block / Agentic AI Foundation (Linux Foundation)
Run a local, model-agnostic coding agent and turn your workflows into portable YAML recipes you can review in a pull request.
Overview
Goose is a general-purpose AI agent written in Rust and released under Apache 2.0 by Block; in 2026 its governance moved to the Agentic AI Foundation at the Linux Foundation, so it is community-governed rather than controlled by a single vendor. The repository carries roughly 53.6k stars and 6.1k forks. The documentation is organised as Quickstart, Guides, Tutorials, Extensions and Recipes, and the Guides tree is where the substance sits: folders for Managing Sessions, Context Engineering, Recipes, Managing Tools, Security and MCP Apps, alongside standalone pages on CLI Commands, CLI Providers, ACP Providers, Configuration Files, Environment Variables, Tool Shim, MCP Sampling, MCP Elicitation, MCP Roots, LLM Rate Limits, Multi-Model Config, Extension Allowlist, Remote Server, Roaming Agents, Codebase Analysis, Offline Docs and Custom Distributions. Goose is deliberately model-agnostic: provider pages cover Anthropic, OpenAI, Google, Azure, Bedrock, Ollama, OpenRouter and more than fifteen others, and the Agent Client Protocol pages let you drive it from an existing subscription rather than raw API keys. Capabilities arrive as MCP extensions — over seventy documented, with thousands more reachable through third-party MCP servers — and Recipes are portable YAML files bundling instructions, required extensions, parameters and subrecipes, which is what makes an agent workflow reviewable, shareable and runnable in CI. Practitioners on Hacker News praised the editor-agnostic design but flagged real caveats worth knowing before you start: 'local' describes the agent, not the model; output quality varies sharply with the backend model you choose; and it edits files aggressively enough that you should work on a clean branch.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- ~1 hour for the quickstart and first recipe; ongoing reference thereafter
- Provider
- Block / Agentic AI Foundation (Linux Foundation)
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Install and run Goose as a desktop app, a CLI, or an embeddable API on macOS, Linux or Windows
- ✓Configure any of 15+ LLM providers, including a fully local Ollama backend, through config files and environment variables
- ✓Add and manage capabilities as MCP extensions, and use the tool shim when a model lacks native tool calling
- ✓Write portable YAML Recipes that bundle instructions, required extensions, parameters and subrecipes for reuse in CI
- ✓Apply the context-engineering guides to keep long multi-step sessions inside a model's context window
- ✓Constrain blast radius with the security guide, the extension allowlist and explicit tool management
- ✓Drive Goose from an existing chat subscription using Agent Client Protocol providers instead of raw API keys
- ✓Point the agent at an unfamiliar codebase and run remote or roaming agents against a shared server
Highlights
- •Governed by the Agentic AI Foundation at the Linux Foundation — vendor-neutral rather than one company's roadmap
- •Recipes are the real differentiator: a YAML file turns an ad-hoc agent session into a reviewable, shareable, CI-runnable artifact
- •Written in Rust and runs entirely on your machine; desktop, CLI and API all share the same engine
- •Genuinely model-agnostic — 15+ documented providers, and an offline path through Ollama for on-prem or air-gapped work
- •Apache 2.0 with ~53.6k GitHub stars and a 249-point Hacker News launch thread containing unusually candid practitioner critique
Who It’s For
Best For
- ✓Engineers who want a coding agent decoupled from any single editor or model vendor
- ✓Platform teams standardising repeatable agent workflows across an organisation via version-controlled YAML
- ✓Developers building or testing their own MCP extensions against a real, open agent host
- ✓Teams needing a local-first agent that can run against self-hosted or on-premises models
Prerequisites
- •Comfort with a terminal, environment variables and editing YAML configuration files
- •An LLM provider API key or a local Ollama install — Goose ships no model of its own, and per-token costs are yours
- •Basic familiarity with the Model Context Protocol helps before writing custom extensions
FAQ
What is Goose Documentation — Open-Source, Extensible AI Agent?
Goose is a general-purpose open-source AI agent that runs on your own machine as a desktop app, CLI, or embeddable API. It is written for engineers who want a coding and automation agent that is not welded to one editor or one model vendor. Working through these docs you learn to wire up any of 15+ LLM providers (including a fully local Ollama setup), add capabilities as MCP extensions, constrain what the agent may execute, and capture multi-step workflows as portable YAML Recipes you can share with a team or run in CI.
Is Goose Documentation — Open-Source, Extensible AI Agent free?
Goose Documentation — Open-Source, Extensible AI Agent is free to access.
What level is Goose Documentation — Open-Source, Extensible AI Agent for?
Goose Documentation — Open-Source, Extensible AI Agent is aimed at a intermediate audience. Recommended background: Comfort with a terminal, environment variables and editing YAML configuration files, An LLM provider API key or a local Ollama install — Goose ships no model of its own, and per-token costs are yours, Basic familiarity with the Model Context Protocol helps before writing custom extensions.
How long does Goose Documentation — Open-Source, Extensible AI Agent take?
Expect roughly ~1 hour for the quickstart and first recipe; ongoing reference thereafter. Most learners work through it at their own pace.
What will I learn from Goose Documentation — Open-Source, Extensible AI Agent?
You'll learn: Install and run Goose as a desktop app, a CLI, or an embeddable API on macOS, Linux or Windows; Configure any of 15+ LLM providers, including a fully local Ollama backend, through config files and environment variables; Add and manage capabilities as MCP extensions, and use the tool shim when a model lacks native tool calling; Write portable YAML Recipes that bundle instructions, required extensions, parameters and subrecipes for reuse in CI; Apply the context-engineering guides to keep long multi-step sessions inside a model's context window; Constrain blast radius with the security guide, the extension allowlist and explicit tool management; Drive Goose from an existing chat subscription using Agent Client Protocol providers instead of raw API keys; Point the agent at an unfamiliar codebase and run remote or roaming agents against a shared server.
Topics
Sources
This page was written from 3 sources, 2 on domains other than goose-docs.ai.