goose

goose is a free open source AI agent that runs on your computer. It can inspect and edit files, execute shell commands, run tests, research topics, analyse data, and automate work through a desktop app, terminal interface, or API. Block originally developed the project. It now sits within the Agentic AI Foundation at the Linux Foundation and uses the Apache 2.0 licence.

You are not tied to one model company. Goose works with more than 15 providers, including OpenAI, Anthropic, Google, Ollama, OpenRouter, Azure, and Amazon Bedrock. It can use API keys, supported ChatGPT, Claude, and Gemini subscriptions, or models running locally. The built in local inference option downloads a compatible GGUF model and runs it directly through llama.cpp, so private files can stay on your machine and the agent can work without an internet connection.

Its MCP support gives the agent access to external tools and data. More than 70 extensions are documented, covering services such as GitHub, Google Drive, databases, browsers, and developer tools. Goose Desktop can also display MCP Apps inside a conversation, including forms, buttons, charts, and other interactive interfaces.

Longer jobs can be stored as YAML Recipes with prompts, parameters, extensions, and smaller subrecipes. Goose can delegate parts of a task to parallel subagents, while the main session remains available for direction and review. Recipes can be shared with a team or run without the desktop interface in automated jobs and CI pipelines.

Security controls include approval modes, tool permissions, sandboxing, prompt injection detection, and an adversary reviewer. Those controls deserve attention because MCP extensions may receive access to files, accounts, and shell commands. Anonymous usage metrics are optional, according to the project documentation. When you choose a cloud model, the provider's own data retention rules still apply.

Goose suits developers who want an agent they can inspect, configure, and run locally. It is also capable of research, writing, and file processing, although setting up providers and extensions takes more work than opening a hosted chatbot. Local models remove API fees and can keep work offline, but speed, context length, and tool use depend on your hardware and the model you choose.

Features

Desktop, CLI, and API: Runs on macOS, Windows, and Linux, with interfaces for interactive work and automation.
Model choice: Connects to OpenAI, Anthropic, Google, Ollama, OpenRouter, Azure, Amazon Bedrock, and other providers.
Local inference: Runs compatible GGUF models in process through llama.cpp without a separate model server.
MCP extensions: Connects to more than 70 documented tools and data sources, with support for additional MCP servers.
MCP Apps: Displays interactive forms, controls, and visualisations inside Goose Desktop.
Reusable Recipes: Saves prompts, parameters, tools, and subrecipes as portable YAML workflows.
Parallel subagents: Delegates separate tasks to independent agents and reports their tool activity in the main session.
Developer access: Reads and writes files, executes commands, runs tests, and works with repositories on the local machine.
Automation support: Runs recipes without the desktop interface and fits into scripts or CI pipelines.
Security controls: Includes permissions, approval modes, sandboxing, prompt injection checks, and an adversary reviewer.
Open source licence: Released under Apache 2.0 and governed through the Linux Foundation's Agentic AI Foundation.

<h3>Alternatives to goose</h3>

<ul>

<li><a href="https://aitoolsdirectory.com/tool/opencode">OpenCode</a>: Open-source coding agent with terminal, desktop, and IDE interfaces, broad model support, permissions, skills, and local inference.</li>

<li><a href="https://aitoolsdirectory.com/tool/cline">Cline</a>: Open-source coding agent for VS Code, JetBrains, and the command line that edits files and runs commands with approval controls.</li>

<li><a href="https://aitoolsdirectory.com/tool/antigravity-cli">Antigravity CLI</a>: Terminal coding agent with parallel subagents, sandboxing, plugins, MCP support, and multi-step task execution.</li>

<li><a href="https://aitoolsdirectory.com/tool/openai-codex">OpenAI Codex</a>: Runs coding tasks in parallel cloud sandboxes, executes tests, edits repositories, and prepares pull requests for review.</li>

<li><a href="https://aitoolsdirectory.com/tool/hermes-agent">Hermes Agent</a>: Self-hosted autonomous agent that learns from completed tasks and saves reusable skills on your server.</li>

</ul>