An open-source AI agent released by Block (formerly Square) in January 2025. It runs on your own machine and autonomously handles engineering work such as editing code, debugging, and deploying. Its defining trait is that it is not tied to any single AI vendor: you can pick a model from more than 15 providers, including Anthropic, OpenAI, Google, and Ollama (local models). Goose has since been donated to the Agentic AI Foundation (AAIF) under the Linux Foundation and is developed by the community. It is licensed under Apache 2.0, and the software itself costs nothing to use.
Key Features
- Three ways to use it: desktop app, CLI, and API: Native apps for macOS, Linux, and Windows, plus a CLI for terminal-based workflows and an API for embedding Goose in other systems. It is written in Rust and runs lightly
- Model-agnostic by design: Supports 15+ providers including Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, and Bedrock. Switching models is a matter of swapping API keys, and with Ollama you can keep everything on local models
- Extensions via MCP: Goose supports the Model Context Protocol (MCP) standard, with 70+ extensions connecting to databases, GitHub, Google Drive, browsers, and more. Your own MCP servers plug in as-is
- Recipes for reusable workflows: Save frequently used tasks as YAML “recipes” that can be shared across a team and re-run later, so you stop rewriting the same instructions
- Subagents for parallel work: Independent tasks can be handed to subagents and processed concurrently
- Security features: Prompt injection detection, execution permission controls, sandbox mode, and an adversary reviewer that double-checks what the agent is doing
- Editor integration (ACP): Goose can act as an Agent Client Protocol server, so it can be called from VS Code, Zed, and JetBrains IDEs
Pricing
| Plan | Price | What’s included |
|---|---|---|
| Goose itself | $0 (Apache 2.0) | Desktop app, CLI, API, and all extensions. No feature limits |
| LLM usage | Pay-as-you-go per provider | You bring your own API key from Anthropic, OpenAI, and so on |
| Using local models | $0 | Running a local LLM through Ollama incurs no API charges |
Pricing is current as of August 2026. Please check the official site for the latest information.
Pros & Cons
✅ Pros
- The software is free and the source is public, so there is no vendor lock-in
- You can freely choose models and tune the balance between cost and quality yourself
- It is built to run locally; combined with Ollama, your code never has to leave your machine
- Strong MCP support means existing MCP servers can be reused directly
- Recipes turn routine work into a shared asset the whole team can run
⚠️ Cons
- You need to supply your own LLM API key, so initial setup takes more effort than a commercial service
- With pay-as-you-go models, costs are hard to predict when the agent runs for a long time
- Output quality depends heavily on the model you choose; cheaper models may not behave as expected
- Because it executes commands autonomously, using it without reviewing permission settings can lead to unintended changes
- There is no official support desk — documentation and the community (Discord) are what you rely on
Comparison with Similar Services
| Item | Goose | Claude Code | Cline | OpenHands |
|---|---|---|---|---|
| Provider | Created by Block, now AAIF (Linux Foundation) | Anthropic | Cline Bot | All Hands AI |
| License | Open source (Apache 2.0) | Proprietary | Open source | Open source |
| Form factor | Desktop app / CLI / API | CLI / desktop | VS Code extension | Web UI / CLI |
| Models | Choose from 15+ providers | Claude models | Multiple providers | Multiple providers |
| Cost of the tool | Free | Subscription or API usage | Free (LLM billed separately) | Free (LLM billed separately) |
Who Is It For
- Developers who want to swap models freely rather than depend on one AI vendor
- Teams that want to try an AI agent with local models, keeping source code inside the company
- People who already build and run their own MCP servers and want an agent that makes use of them
- Anyone who wants to systematize routine maintenance or debugging steps as “recipes”
- Those who value open source and the ability to read the code themselves
Summary
Goose combines the freedom to choose your model with the reassurance of running on your own machine, all as open source. The tool itself is free; the only cost is the API usage of whichever LLM you pick. Because MCP extensions and recipes let you grow it around your own environment and workflows, it is a strong option for anyone who finds commercial coding agents limiting, or who cannot send code outside their organization. Install the desktop app and start with a small task using an API key you already have — or Ollama.