goose

Free

goose is an open-source AI agent that runs locally on your machine. You bring your own API keys or connect existing subscriptions, and it writes code, runs commands, and edits files from your desktop or terminal.

It is built in Rust and now lives under the Linux Foundation's Agentic AI Foundation. The Apache-2.0 license matters if you need independence from a single vendor.

It sits in AI Agents work—autonomous helpers that act across tools rather than just suggesting the next line.

Open it if you want local control and provider choice. Skip it if you want a turnkey cloud service with one bill.

It helps when these jobs are in front of you.

• Developers wanting local AI execution: Your code stays on your machine with your own hardware. You trade cloud convenience for privacy.
• Users with existing LLM subscriptions via multiple providers: You connect Claude, OpenAI, Google, or Ollama without buying another subscription.
• Those needing cross-platform desktop AI tooling: You run the same agent on macOS, Linux, and Windows with the same recipes.
• Teams sharing repeatable workflows through recipes: You export YAML workflows with parameters and sub-recipes that teammates import with one link.
• Engineers who value version-controlled automation over one-off prompting: You diff recipe changes, revert mistakes, and treat prompts like code.
• Users wanting ambient AI assistance without leaving their terminal: The CLI feels native alongside git and build commands.
• Developers building the same project types repeatedly: A recipe scaffolds the boilerplate so you start from working code, not blank files.
• Privacy-conscious users who can accept slower iteration: Local Ollama models keep everything offline, though results lag behind Claude Sonnet.

These are the parts you notice after a few tries.

• Recipes as shareable workflows: You write a YAML file with parameters, sub-recipes, and pinned MCP servers. Then you share a link that imports with one click.
• Provider flexibility without switching tools: You jump from Claude to OpenAI to a local model without learning a new interface each time.
• ACP bridge for existing subscriptions: You run Goose on top of Claude Code or ChatGPT Plus you already pay for, adding recipes and 70+ extensions.
• Named sessions that persist: You close your terminal and reopen to the same context, no state lost.
• MCP extensions for real tools: You connect GitHub, Google Drive, browsers, and databases through an open standard, not locked APIs.

Setup takes a few decisions before the first prompt.

1. Install the desktop app or CLI for your platform.
2. Choose your LLM provider and paste your API key.
3. Enable or install MCP extensions for the tools you need.
4. Start a session and describe the task in plain language.
5. Review permissions before Goose runs commands, then iterate.

Skip it when the fit is wrong.

• Skip it if you want zero configuration: You pick providers, manage credentials, and audit permissions yourself. There is no bundled model.
• Skip it if you need the fastest iteration: Local models on normal hardware lag behind cloud Claude. Privacy costs speed.
• Skip it if you want guaranteed safety: The default autonomous mode grants broad permissions. You must review before it acts.
• Skip it if ACP resumability matters: Sessions bridged from Claude Code cannot be resumed or forked mid-work.