Goose is the Agentic AI Foundation's Apache-2.0 open source AI agent that started at Block and has now graduated to the Linux Foundation's AAIF. With 46,000+ GitHub stars, native desktop apps for macOS, Linux, and Windows, a full CLI, an embeddable API, and a Rust core, it is one of the most fully-shipped open agents available and a serious alternative to closed coding assistants for developers who want LLM-agnostic local control. ## What Goose Actually Does Goose runs as a general-purpose agent on your machine. It is not just a code suggestion engine; it can install dependencies, execute shell commands, edit files, run tests, query databases, fetch from APIs, drive a browser, and chain those actions into multi-step workflows. The marketing line of going beyond code suggestions is technically accurate: Goose treats the developer's entire local environment as a tool surface, not just the editor buffer. ## Rust Core, Cross-Platform Surface The codebase is roughly 63% Rust and 30% TypeScript, with Rust handling the agent runtime, tool execution, and provider dispatch, and TypeScript powering the desktop UI. The Rust core is what makes Goose viable as a CLI on Linux servers and as a native app on macOS and Windows without a heavyweight Electron-only path. Performance and portability are real, not aspirational, and the binary footprint is small enough to embed in other tooling. ## 70+ Extensions via MCP Goose connects to over 70 extensions through the Model Context Protocol (MCP), the open standard for letting agents talk to tools, data sources, and services. MCP servers for databases, GitHub, Google Drive, browsers, Slack, filesystems, and dozens of other targets are all reachable from Goose out of the box. This matters because MCP is the layer where Goose's tool surface scales with the broader ecosystem rather than depending on Block-specific integrations. ## 15+ LLM Providers, Bring Your Own Subscription Provider support covers Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Bedrock, and more than a dozen others. Users can authenticate with API keys or attach existing Claude, ChatGPT, or Gemini subscriptions through ACP, which lowers the operating cost dramatically for anyone who already pays for a frontier model subscription. Local models via Ollama are first-class, so Goose works fully offline against a quantized DeepSeek or Llama if needed. ## Desktop, CLI, and API Goose ships three usage surfaces from one codebase. The desktop app gives a chat-style interface with rich tool output rendering for developers who want a visible inspector. The CLI is the path for terminal-native workflows and scripting, including unattended sessions. The API lets Goose be embedded into other applications, custom developer tools, or CI systems where an agent needs to act on behalf of the platform itself. All three share the same Rust runtime, so capability is consistent across them. ## Custom Distributions Organizations can build custom Goose distributions with preconfigured providers, branded UI, locked-down extension lists, and policy hooks. This is the path Block and other adopters have used to ship internal agent platforms on top of Goose without forking the core, and it is one of the reasons Goose has attracted enterprise interest as part of AAIF. ## Why Goose Stands Out Most open agents are either thin LLM wrappers without serious tool execution or research prototypes without a sustainable shipping cadence. Goose is unusual because it is a real product (desktop app, installers, signed binaries, accessibility support) that also happens to be open source, runs in Rust, ships with first-class MCP, and now has Linux Foundation governance behind it. The combination is rare and is what is driving its star growth. ## Limitations The Rust toolchain makes contribution friction higher than for Python-based agents, and the codebase is large enough that orientation takes time. Provider quality is uneven; the agent is only as good as the underlying model, and using a weak local model with heavy tool use produces brittle behavior. MCP is powerful but its security model puts responsibility on the user to vet which extensions are trusted with shell or filesystem access. Apache-2.0 licensing keeps the project permissively reusable, but the governance move to AAIF means external contributors are now negotiating with a foundation rather than a single company, which has both upsides (neutrality) and downsides (slower decisions).