Goose: Dorsey’s local Claude rival

Goose is an open-source artificial intelligence agent from Block that runs on your own machine instead of a vendor’s cloud. Block announced “codename goose” on January 28, 2025, and the project is now positioned as a desktop app, command line tool, and application programming interface for coding and other tasks. (block.xyz) (goose-docs.ai) An artificial intelligence agent is software that does work after you give it a goal, not just text autocomplete. Block said Goose can read and write files, run code and tests, install dependencies, and connect to outside tools through Anthropic’s Model Context Protocol, which is a standard way to let models use apps and data sources. (block.xyz) (goose-docs.ai) The local part means the software itself runs on a Mac, Windows, or Linux computer you control. Goose’s documentation says it works with more than 15 model providers, including Anthropic, OpenAI, Google, Ollama, Azure, Bedrock, and OpenRouter, so users can bring their own model instead of being locked to one service. (goose-docs.ai) That puts Goose next to Anthropic’s Claude Code, which is also an agentic coding tool that can read a codebase, edit files, and run commands. Anthropic’s pricing page says Claude Code is bundled into paid Claude plans including Pro at $17 a month billed annually or $20 monthly, with higher-priced Max, Team, and Enterprise tiers above that. (anthropic.com) (claude.com) The sharper distinction is where the work runs and who controls the stack. Anthropic’s documentation says Claude Code on the web runs on Anthropic-managed cloud infrastructure, while Goose is marketed as “on-machine” software under the Apache 2.0 open-source license. (code.claude.com) (goose-docs.ai) (block.xyz) Block is also tying Goose to a broader open-standards push. The Goose site says the project is now part of the Agentic AI Foundation at the Linux Foundation, and the Linux Foundation said on December 9, 2025 that the new group launched with founding contributions including Anthropic’s Model Context Protocol, Block’s Goose, and OpenAI’s AGENTS.md. (goose-docs.ai) (linuxfoundation.org) Block says Goose has moved beyond an internal experiment. In a March 2026 post announcing a Goose grant program, the company said the tool had grown to thousands of community members, been adopted by companies including Databricks, and was saving Block teams 50 percent to 75 percent of development time, with 60 percent of Block’s workforce using it weekly. (block.xyz) The project’s public footprint has also grown. Goose’s documentation site says the project has more than 38,000 GitHub stars, more than 400 contributors, and more than 70 documented Model Context Protocol extensions for services including GitHub, Google Drive, browsers, and databases. (goose-docs.ai) Dorsey’s pitch lands in a market where developers are weighing convenience against cost, control, and lock-in. Goose does not remove model costs if a user connects paid providers, but it does let people run the agent layer locally, inspect the code, and swap models and tools without committing to one subscription. (goose-docs.ai) (block.xyz) So the story is less about a single new chatbot than about who owns the coding assistant sitting on a developer’s laptop. Goose is Block’s bet that an open, local agent can compete with subscription-first tools by giving users the software itself, not just access to a hosted service. (block.xyz) (goose-docs.ai)