Goose
an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM
Last verified:
What is Goose?
Goose is an open-source, general-purpose AI agent that runs locally (desktop app or CLI) and connects to language model providers to perform tasks such as research, writing, automation, data analysis, and developer workflows. It exposes an interactive agent loop that lets the model request and execute tool calls (run commands, edit files, open apps) via a modular extension system, enabling the agent to act on the user’s machine while returning structured results. Key features include a cross-platform desktop app (macOS, Linux, Windows), a CLI with session and configuration commands, an Extensions system for adding capabilities (Apps, Docs, MCP servers, integrations), and support for the Model Context Protocol (MCP) to connect to external data sources and servers. Goose is aimed at developers, researchers, and power users who want an extensible local agent that can integrate with custom tools, run code and shell commands, manipulate files, and connect to self-hosted or third-party LLM providers.
Goose pricing
Pricing model: Freemium
The website and project materials present Goose as an open-source agent (no license fee for the software itself) and provide downloads and CLI install scripts; however, running inference requires connecting to an LLM provider (self-hosted models or third-party APIs) which may have its own pricing. Goose itself is distributed freely and does not list paid tiers for the agent software; costs come from the chosen model provider or any hosted MCP services you connect to.
Goose pros
- Open-source agent you can audit and extend
- Runs locally via desktop app or CLI for privacy and control
- Cross-platform support: macOS, Linux, Windows
- Extension system lets you add Apps, Docs, and custom tools
- Model Context Protocol (MCP) support for external data interoperability
- Can execute shell commands and manage files from agent tool calls
- Interactive loop that returns tool-call results back to the model
- Ability to run arbitrary HTML apps in standalone windows via Apps extension
- Session and configuration commands for reproducible workflows
- Designed to interoperate with any LLM provider you choose
- Token/context management via context revision to reduce irrelevant info
- CLI-first workflow enables automation and scripting
- Prebuilt integrations (Docs MCP server) for serving documentation to the agent
- Supports streaming outputs when used with compatible providers
- GitHub-backed project with public repo and documentation for contributors
Goose cons
- Requires technical setup for CLI or desktop installation
- Local execution means resource limits depend on the user’s machine
- Not a managed hosted LLM service — needs provider or self-hosted models
- Some extensions (MCP servers) require additional configuration
- User must manage API keys and provider connections securely
- Potential security risk if misconfigured tools can run arbitrary commands
- Less plug-and-play for non-technical end users compared with SaaS agents
- Feature parity/UX depends on community-driven extensions and maintenance
Frequently asked questions about Goose
How do I install Goose on my machine?
You can download the cross-platform desktop application for macOS, Linux, or Windows from the Goose website or install the CLI via the provided install script; the CLI installation is documented with a curl command that downloads and runs the installer, and further configuration is available via the goose configure command.
Do I need to provide an LLM provider to use Goose?
Yes — Goose is an agent interface and extensions platform; it connects to language model providers (self-hosted or third-party) to perform inference, so you must configure a provider or local model to run tasks and completions.
What are Extensions and how do I use them?
Extensions are modular components that add capabilities (for example, Apps, Docs, or MCP servers); you enable or configure them through the app sidebar or CLI, and they expose tools the agent can call during its interactive loop to perform actions like opening HTML apps, serving docs, or accessing external data sources.
What is the Model Context Protocol (MCP) and why does Goose use it?
MCP is an open standard for interoperability between AI agents and data sources; Goose uses MCP to connect to MCP servers or systems so the agent can access external data and services in a structured way, improving tool integration and context delivery to the model.
Can Goose run shell commands and edit files?
Yes — through its extensions and tool-call mechanism, Goose can execute shell commands, read and write files, and run programmatic actions as part of the agent’s tool calls, returning results to the model for iterative workflows.
Is Goose suitable for non-developers?
Goose targets developers, researchers, and power users; non-technical users may find setup and configuration (provider setup, extensions, CLI usage) challenging compared with hosted SaaS agent products, though the desktop app and documented guides lower the barrier somewhat.
How does Goose handle context and token limits?
Goose implements context revision in its interactive loop to remove old or irrelevant information, helping manage token usage and keep the model focused on the most important context, which mitigates but does not eliminate provider token limits.
Where can I find documentation and guides for Goose?
Comprehensive documentation, guides for CLI commands, extension usage, and architecture details are available on the Goose docs site and the project’s GitHub Pages, covering installation, configuration, and developer-focused guides.
Can I create custom HTML apps within Goose?
Yes — the Apps extension allows you to create, manage, and launch custom HTML apps in standalone windows so the agent can present or interact with custom UI components and tooling locally.
Is Goose actively maintained and open for contributions?
Goose is published as an open-source project with a public GitHub repository and documentation, encouraging community contributions, issue reports, and extension development to expand the agent’s capabilities.