Goose is an open-source artificial intelligence agent originally developed by Block. It operates locally as a desktop application or command-line interface and can perform software development and other computer-based tasks using large language models. The agent is model-independent, supporting models from multiple providers, including Anthropic, OpenAI, Google, Amazon Bedrock, and OpenRouter, as well as locally hosted models through Ollama.
Goose is designed to give users control over their preferred language model while providing the agent with access to local development tools and external services. It functions as an agent harness, enabling automated execution of complex, multi-step tasks on a user's computer.
History
Goose was developed at Block and publicly introduced as codename goose in January 2025. The initial release focused on software engineering tasks, allowing users to select their preferred language model while giving the agent access to local development tools and external services. The project was released as open-source software from the outset.
In December 2025, Block contributed Goose as a founding project of the Linux Foundation's Agentic AI Foundation (AAIF), alongside Anthropic's Model Context Protocol (MCP) and OpenAI's AGENTS.md. This move aimed to foster collaboration and standardization within the emerging field of agentic artificial intelligence. The project's repository and governance were formally transferred from Block to the AAIF in April 2026.
Features
Goose can read and modify files, execute shell commands, build and run software, perform tests, and interact with external services through its agent harness. This capability makes it suitable for a wide range of automation tasks beyond simple code generation, including system administration and data processing.
The agent is available as a native desktop application for macOS, Linux, and Windows, as well as through a command-line interface and an application programming interface (API). This multi-interface approach allows Goose to be integrated into various workflows, from interactive use to fully automated pipelines.
Goose supports the Model Context Protocol, allowing the agent to connect to external tools, databases, APIs, browsers, and other services through MCP extensions. This extensibility is a core design principle, enabling users to tailor the agent's capabilities to their specific needs.
Model and Provider Support
A defining characteristic of Goose is its model independence. It supports multiple AI providers, including Anthropic, OpenAI, Google, Microsoft Azure, Amazon Bedrock, and OpenRouter. Additionally, users can run models locally through Ollama, providing options for privacy, cost control, and offline operation.
Goose can also operate as an Agent Client Protocol (ACP) server. This allows it to integrate with compatible development environments, including Visual Studio Code, Zed, and JetBrains products, bringing its automation capabilities directly into popular coding interfaces.
Ecosystem and Governance
As a founding project of the Linux Foundation's Agentic AI Foundation, Goose is positioned within a broader effort to create open standards for AI agents. Its governance is now community-driven under the AAIF, with contributions from various organizations and individuals. The project's open-source nature encourages community development and transparency, aligning with other initiatives in the open-source AI space.
The development of Goose reflects broader trends in Generative AI and large language models, where agents are increasingly used to bridge the gap between conversational AI and actionable computer tasks. Its architecture, leveraging Multi-Head Attention and other Transformer (architecture)-based techniques, is typical of modern AI systems, though Goose itself abstracts these complexities from the end user.
See Also
- List of AI-assisted software development tools
- Lists of open-source artificial intelligence software
External Links
- Official website
- Goose on GitHub