goose
Open source AI agent with a desktop app, CLI and API, built in Rust
About goose
goose is an open source AI agent that runs on your own machine instead of in somebody else's cloud. It ships three ways into the same engine: a native desktop app for macOS, Linux and Windows, a full CLI for terminal work, and an API for embedding it in something else. The whole thing is written in Rust, and the project describes itself as general purpose rather than code specific, so research, writing, automation and data analysis are explicitly in scope alongside programming.
The part worth understanding before you compare it to anything else is governance. goose moved to the Agentic AI Foundation at the Linux Foundation, which puts the project under vendor-neutral, community governance rather than one company's roadmap. That move also changed the addresses: the repository is now at github.com/aaif-goose/goose and the docs live at goose-docs.ai, with the old Block URLs redirecting. The licence is Apache-2.0, there are over 600 contributors, and the project counts more than 70 documented MCP extensions. It is model agnostic across 15 or more providers, including Ollama for local models, and through the Agent Client Protocol it can run on a Claude, ChatGPT or Gemini subscription you already pay for rather than a separate API bill.
You install the desktop app or run a one-line CLI installer, then point goose at a model provider using an API key, a local Ollama instance, or an existing chat subscription over ACP. From there you describe what you want done and the agent installs, executes, edits and tests rather than returning suggestions for you to apply. Tools reach it through the Model Context Protocol, so databases, APIs, browsers, GitHub and Google Drive connect through the same open standard rather than bespoke integrations. Work you repeat gets captured as a Recipe, a portable YAML file holding instructions, extensions, parameters and subrecipes, which can be shared with a team or run in CI. Because goose also acts as an ACP server, Zed, JetBrains and VS Code can drive it from inside the editor.
- •Three Surfaces, One Engine - A native desktop app for macOS, Linux and Windows, a full CLI for terminal workflows, and an API to embed the agent in your own software
- •Model Agnostic - Works with 15+ providers including Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure and Bedrock, or through an existing Claude, ChatGPT or Gemini subscription via ACP
- •MCP Extensions - More than 70 documented extensions connecting databases, APIs, browsers, GitHub and Drive through the open Model Context Protocol
- •Recipes - Workflows captured as portable YAML with instructions, extensions, parameters and subrecipes, so the same run works on a teammate's machine or in CI
- •Subagents - Independent agents spawned to handle code review, research or file processing in parallel while the main conversation stays focused
- •Security Controls - Prompt injection detection, per-tool permission controls, a sandbox mode, and an adversary reviewer that watches for unsafe actions
- •Vendor-Neutral Governance - Apache-2.0 under the Agentic AI Foundation at the Linux Foundation, with 600+ contributors
Developers who want an agent on their own hardware, with their own model keys, and no vendor in the middle of the work. The ACP support matters most if you already pay for Claude or ChatGPT and would rather not open a second metered account, and the Ollama support matters most if the code cannot leave the machine at all. Teams get the clearest value from Recipes, which turn an ad-hoc prompt sequence into something reviewable and runnable in CI. It is a weaker fit if you want a focused coding assistant with an opinionated setup, because goose deliberately spans research, writing and automation as well as code, and in the narrow coding case it competes with tools built for nothing else. It is also a poor fit if you need a published support contract: there is no commercial tier to buy.
Pricing
No pricing page found
- Open Source (Apache-2.0)$0/mo
Checked 2026-10-23
