Submit an agent

goose

Open-source, extensible AI agent for desktop and command line that works with any LLM.

FreeFree planOpen-source local AI agent for coding and general tasks

No user reviews yet. Be the first to rate it.

Visit website
goose website screenshot
goose website

Quick Facts

DeveloperAgentic AI Foundation (Linux Foundation); created by Block
Websitegoose-docs.ai
Agent typeOpen-source local AI agent for coding and general tasks
Best forDevelopers and technical teams who want a free, local agent they can extend with MCP servers and point at any model.
Not ideal forNon-technical users wanting a hosted product with vendor support, or teams that cannot manage their own model access.
Pricing modelFree, open source (Apache 2.0); pay for model usage
Starting priceFree
Pricing detailsgoose itself is free under the Apache 2.0 license. Users pay their LLM provider, sign in with a ChatGPT subscription, use a free model tier or run local models.
Free planYes
ChannelsDesktop app, CLI, API, JetBrains plugin
What it does on its ownInstalls, executes, edits and tests on its own using tools and extensions; recipes package repeatable multi-step workflows.
IntegrationsAnthropic, OpenAI, Google Gemini, Ollama, Mistral, xAI, Azure OpenAI, OpenRouter, AWS Bedrock, MCP servers, JetBrains
PlatformsmacOS, Linux, Windows
Agent launched2025
Details last verified29 October 2026
CategoriesCoding & Software Development, Free AI Agents, General-Purpose AI Agents, Open-Source AI Agents
User ratingNo reviews yet

goose costs nothing beyond the model it runs on, and it now belongs to the Agentic AI Foundation, a Linux Foundation body, rather than to Block, the company behind Square and Cash App that created it. It was one of the foundation’s three founding projects in December 2025, alongside the Model Context Protocol and AGENTS.md.

Block built goose as an internal coding agent, rolled it out to all employees and released it under the Apache 2.0 license. It runs on a user’s own machine, not in a vendor’s cloud.

How It Works

goose comes as a desktop app for macOS and a command-line tool for macOS, Linux and Windows, plus an API and a JetBrains plugin. On first launch it asks for a model provider. Options include an API key from Anthropic, OpenAI, Google or others, a ChatGPT subscription, routers such as OpenRouter, or local models through Ollama; more than 15 providers are documented. Its docs say it works best with Claude models because it leans heavily on tool calling.

Given a task, goose plans the steps and acts: installing packages, running commands, editing files and testing the result. Extensions built on the Model Context Protocol give it new abilities, with more than 70 first-party ones and access to thousands of public MCP servers. Recipes, written in YAML, bundle prompts, extensions, parameters and error handling into shareable workflows that a team can rerun.

Pricing

goose is free to download and use. Costs depend on the model: pay-per-token API usage, an existing ChatGPT subscription, a free tier such as Google Gemini’s, or no fee at all with a local model on your own hardware. Heavy tool use can burn through tokens quickly on paid APIs, so watch usage.

What to Watch Out For

An agent that can run shell commands on your machine needs guardrails, so review its permission settings and be careful with untrusted MCP servers and prompts. Results vary a lot by model, and cheaper or local models may struggle with long, tool-heavy tasks. There is no vendor support contract; help comes from documentation and the open-source community.

Recent News

In May 2026, The New Stack examined why Block handed goose to the Linux Foundation, reporting that Block’s ownership of the trademarks had held back enterprise adoption and that neutral governance was meant to fix it.

Features

  • Desktop app and command-line interface
  • Works with 15+ LLM providers or local models
  • 70+ first-party MCP extensions plus the wider MCP ecosystem
  • YAML recipes for shareable, repeatable workflows
  • Runs code, edits files and executes commands locally
  • ChatGPT subscription sign-in as a provider option

Pros and Cons

Pros

  • Free and Apache 2.0 licensed
  • Neutral governance under the Linux Foundation's Agentic AI Foundation
  • Model-agnostic, including local models
  • Large extension ecosystem through MCP
  • Runs on macOS, Linux and Windows

Cons

  • Quality depends on the model
  • docs say it works best with Claude models
  • Requires setting up providers and API keys
  • No paid vendor support
  • Shell access means permissions need careful setup

Alternatives

User Reviews

No reviews yet. Be the first to rate goose.

Write a Review