Goose Open source

What it is

Goose is an open-source (Apache-2.0) AI agent that runs on your own machine as a desktop app for macOS, Linux and Windows, as a CLI, or through an API for embedding. It was created at Block and has since moved to the Agentic AI Foundation (AAIF) at the Linux Foundation for vendor-neutral governance; the code now lives at aaif-goose/goose and the docs at goose-docs.ai. Goose goes beyond code suggestions: it can install packages, run commands, edit files and test the result, and it is also used for non-coding automation.

The README's CLI install is curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash; the desktop app is a separate download.

Key features

• Works with 15+ LLM providers, including Anthropic, OpenAI, Google and local models through Ollama
• Recipes: portable YAML workflow files you can share with a team or run in CI
• Subagents for parallel work such as code review and research
• Security features including prompt-injection detection, permission controls and a sandbox mode
• Custom distributions if you want to ship a branded or preconfigured build

Skills, subagents and MCP

Goose is MCP-first: its extensions are MCP servers, and the project lists more than 70 integrations for databases, APIs, browsers and cloud services. The documentation also covers skills for adding custom capabilities and subagents for delegating tasks. The repository itself uses an AGENTS.md file for contributor instructions.

Pricing

Goose is free. You pay only for the model provider you choose, or nothing if you run a local model.

Getting started

After installing, you choose a model provider and add the extensions you need from the desktop app or the CLI configuration. Because recipes are plain YAML, a team can check them into a repository and run the same workflow locally or in a pipeline, which makes Goose useful for repeatable chores such as dependency updates, reports or code reviews.

Who it is for

People who want an open, provider-neutral agent with a desktop interface, teams that want to share repeatable workflows as recipes, and anyone building around MCP servers.

Verdict

Goose is a solid choice for an MCP-centred, general-purpose agent with good governance behind it. It is less specialised for large refactors than coding-only agents, and results depend heavily on the model you connect.