Goose
Goose can run system commands, install dependencies and run tests, with an open choice of underlying model. It aims to automate development tasks end to end rather than limiting itself to editor suggestions. The project, initially run by Block, has since been transferred to a new GitHub organization (`aaif-goose`).
Key advantages
- Open source, active community
- Goes beyond suggestions: actually executes the actions
- Compatible with several model providers
Who it's for, and why
Developers who want to delegate entire tasks (installing, running, testing), not just code generation.
- Automated installation and configuration of an environment
- Running and debugging tests
- Automating multi-step development tasks
Getting started
- 1Follow the up-to-date install guide on the project's repository: install commands changed along with the organization transfer, so it's better to check the currently documented version than a fixed command.
- 2Configure a model provider (Anthropic, OpenAI, Google...) via `goose configure`.
- 3Start a session: `goose session start`.
Things to watch out for
- The project changed GitHub organization (transferred to `aaif-goose/goose`): check that any third-party documentation or link points to the current repository before relying on it.
- Can run system commands and install dependencies autonomously: supervision is recommended on a production environment, not just a disposable dev machine.
Frequently asked questions
Is Goose still a Block (Square) project?
The repository was transferred to a new organization (`aaif-goose`): check the current origin and governance before relying on it for anything critical.
Which models are supported?
Several providers: Anthropic, OpenAI, Google and others depending on configuration, with no lock-in to one.
How is it different from a simple code-completion assistant?
Goose actually runs commands (installing, testing) rather than just proposing text for manual approval.
Similar tools
Worth exploring too
Claude Code
Anthropic's command-line agent, able to read, write and execute code directly inside a real project.
Vercel MCP
Vercel's MCP server: check deployments, logs and projects directly from an AI agent.
Playwright
Microsoft's browser testing and automation framework, used here to capture real screenshots.