Goose
Local-first AI agent — offers three interfaces: desktop, CLI, and API.
- Local-first AI agent: Operates directly on the user's machine through three interfaces: a desktop app (macOS/Linux/Windows), a CLI, and an API. Performs code editing, shell execution, testing, and multi-step workflows autonomously without cloud dependency.
- Supports 15+ LLM providers: Including Anthropic, OpenAI, Google Gemini, Ollama, OpenRouter, Azure, Bedrock, Groq, xAI, Alibaba (Qwen/DashScope), Databricks, and Scaleway. Optimized for the Claude 4 model but also supports local models (Ollama, LM Studio). Existing ChatGPT/Claude subscriptions can be utilized via ACP (Agent Client Protocol).
- 70+ MCP extensions: Connects to over 70 extensions, including databases, APIs, browsers, GitHub, and Google Drive, through Anthropic's Model Context Protocol standard. Operates a community Extensions Marketplace.
- YAML-based recipe system: Captures workflows as YAML recipes, structured into instructions, extensions, parameters, and subrecipes, enabling reusability and sharing.
- Sub-agent parallel processing: Employs a Rust-based, high-performance architecture to run sub-agents in parallel, efficiently handling complex multi-tasking.
- Security features: Includes prompt injection detection and an adversarial reviewer mode.
- MCP Apps: Features MCP apps that render interactive UIs within the desktop environment.
💻 System Requirements
Not required (LLM inference is delegated to an external API). When using a local model (Ollama), 8-24GB of VRAM is required, depending on the model size. Local inference is possible on non-NVIDIA GPUs with Linux Vulkan support.
Desktop app ~160MB (varies by platform), approximately 500MB in total including the CLI. Local models are stored separately.
⚡ Installation
4-1. Quick Start
# macOS — Homebrew (CLI)
brew install block-goose-cli
# macOS — Homebrew (Desktop App)
brew install --cask block-goose
# Linux / macOS / WSL — curl script
curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash
4-2. Detailed Installation
# Windows — PowerShell (Native)
Invoke-WebRequest -Uri "https://raw.githubusercontent.com/aaif-goose/goose/main/download_cli.ps1" -OutFile "download_cli.ps1"
.\download_cli.ps1
# Windows PATH configuration (Git Bash/MSYS2)
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
# Linux — DEB package (Ubuntu/Debian)
sudo dpkg -i goose_*.deb
# Provider configuration (automatically guided during the first run, subsequent reconfigurations)
goose configure
# Start a session
goose session
# Update
goose update
The desktop app can be directly downloaded from the GitHub Releases page for macOS (Apple Silicon/Intel), Linux (DEB/RPM/Flatpak), and Windows ZIP. Windows users require Git Bash, MSYS2, or PowerShell, and Linux installation is also possible via WSL.
🧬 Bio Use Cases
Refactoring the Entire Codebase
In a large code repository, search for specific patterns, analyze dependencies, and then autonomously perform refactoring across multiple files. By connecting to GitHub PR generation via MCP, the entire cycle of code review, modification, commit, and PR can be completed in a single session.
Multi-Provider Cost Optimization Workflow
Define a multi-model strategy using YAML recipes to separate simple code generation (using free tiers like Groq/Google Gemini), complex architecture design (using Claude 4 Opus), and code review (using GPT-5), thereby controlling monthly API costs while maintaining quality.
Automated Data Pipeline Construction
Through MCP expansion, perform database queries, write data transformation scripts, execute tests, and validate results in a single Goose session. Saving it as a recipe transforms it into a repeatable, automated pipeline.
FAQ
What is Goose?
Local-first AI agent: Operates directly on the user's machine through three interfaces: a desktop app (macOS/Linux/Windows), a CLI, and an API. Performs code editing, shell execution, testing, and multi-step workflows autonomously without cloud dependency. Supports 15+ LLM providers: Including Anthropic, OpenAI, Google Gemini, Ollama, OpenRouter, Azure, Bedrock, Groq, xAI, Alibaba (Qwen/DashScope), Databricks, and Scaleway. Optimized for the Claude 4 model but also supports local models (Ollama, LM Studio). Existing ChatGPT/Claude subscriptions can be utilized via ACP (Agent Client Protocol). 70+ MCP extensions: Connects to over 70 extensions, including databases, APIs, browsers, GitHub, and Google Drive, through Anthropic's Model Context Protocol standard. Operates a community Extensions Marketplace. YAML-based recipe system: Captures workflows as YAML recipes, structured into instructions, extensions, parameters, and subrecipes, enabling reusability and sharing. Sub-agent parallel processing: Employs a Rust-based, high-performance architecture to run sub-agents in parallel, efficiently handling complex multi-tasking. Security features: Includes prompt injection detection and an adversarial reviewer mode. MCP Apps: Features MCP apps that render interactive UIs within the desktop environment.
When should I use Goose?
Local-first AI agent — offers three interfaces: desktop, CLI, and API.
What is a biomedical use case for Goose?
Refactoring the Entire Codebase: In a large code repository, search for specific patterns, analyze dependencies, and then autonomously perform refactoring across multiple files. By connecting to GitHub PR generation via MCP, the entire cycle of code review, modification, commit, and PR can be completed in a single session.
📝 Update Notes
No update notes yet.
🧪 Related Code of Life
No related Code of Life posts yet.