Goose

Goose is an open-source, general-purpose AI agent that runs entirely on your machine. Originally built by Block and written in Rust, it goes beyond code suggestions: it executes shell commands, edits files, runs tests, and orchestrates multi-step tasks autonomously. Because it is on-machine, your code and credentials stay local — Goose only talks to whichever model provider you point it at.

It is aimed at developers and power users who want an agent they fully control rather than a hosted black box. You bring your own LLM and extend its capabilities through the Model Context Protocol, so the same agent can write code, query a database, or automate a research workflow depending on the extensions you connect.

Highlights

• MCP-native — connect to 70+ documented extensions over the Model Context Protocol to add tools, data sources, and integrations to the agent.
• Model-agnostic — works with 15+ providers including Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, and Bedrock, including local models.
• Two interfaces — a native desktop app for macOS, Linux, and Windows, plus a full CLI for terminal and headless/scripted runs.
• Beyond code — runs commands, edits files, executes code, and chains multi-step workflows for research, automation, and data analysis, not just coding.
• On-machine and private — the agent runs locally; only model calls leave your machine, so credentials and source stay under your control.
• Built in Rust — fast startup and a single portable binary, with an API for embedding Goose into your own tools.

In an AI-assisted workflow

Goose fits where you want autonomy without giving up control of the model or the machine. Configure a provider once, then drive it from the terminal for scripted, repeatable tasks:

goose configure        # pick a provider and model
goose session          # start an interactive agent session
goose run -t "Add a /health endpoint to the Express app in src/ and write a test"

Connect MCP extensions (a database, a browser, your issue tracker) and the same agent can reach across tools to complete a task end to end, then hand you the diffs.

Good to know

Goose is free and open source under the Apache-2.0 license. You supply your own model API key (or run a local model), so usage cost depends on the provider and model you choose. The desktop app and CLI are available for macOS, Linux, and Windows.