Open-source AI agent goose released

Summary

Goose is a free, open-source AI agent that can be used as a desktop application, command-line interface, or API. Users can seamlessly switch between more than 15 large language model (LLM) providers and automate tasks through MCP tool integrations, sub-agents, and PC control plugins. The GIGAZINE review provides a step-by-step guide to installation on Windows, demonstrating file editing, browser control, and recurring prompt 'recipes'.

Why it matters

In a market where tools like Claude Cowork and ChatGPT Work lock users into a single model, Goose's provider-agnostic, open-source architecture lowers both cost and vendor risk barriers for teams that do not want to rely on proprietary subscriptions. Its MCP-based extensibility and sub-agent orchestration position it as a potential building block for custom AI workflows, moving beyond the scope of a single-purpose chatbot.

Highlights

• Supports over 15 LLM providers; can orchestrate Claude Code and Codex in a single unified environment via ACP
• MCP plugins (e.g., 'Computer Controller') enable the agent to drive local browsers and operating systems; third-party MCP servers like Context7 can be plugged in
• Sub-agent mode distributes work across multiple models; the API/SDK layer allows the agent to be embedded in other applications

Critical take

The article acknowledges that the 'recipe' (recurring prompt) feature was not functioning correctly at the time of writing. This undermines the promise of reliable scheduled automation for a tool that heavily relies on this feature. From a broader perspective, the review reads like a shallow feature list with basic manual testing. There is no benchmarking or discussion of error handling. Additionally, the PC control plugin is presented without any mention of the security or permission model implications of an LLM driving a desktop environment.