Goose is an open-source, extensible AI agent framework designed to run natively on user machines (desktop app, CLI, and API). Built in Rust for performance and portability, it serves as a general-purpose agentic platform for code development, research, writing, automation, and data analysis. Unlike single-purpose AI tools, Goose implements a sophisticated architecture centered around protocol-based interoperability, particularly the Model Context Protocol (MCP) and Agent Client Protocol (ACP).

Core Technical Architecture

1. Multi-Interface Architecture

Desktop Application: Native cross-platform application for macOS, Linux, and Windows with rich UI capabilities including MCP Apps (interactive UIs rendered directly in the desktop interface)
Command-Line Interface (CLI): Full-featured terminal interface supporting scripting, automation, and integration into developer workflows
API Layer: Embeddable API enabling integration into custom applications, CI/CD pipelines, and enterprise systems

2. Protocol-Based Extensibility Framework

Model Context Protocol (MCP) Integration: Goose implements one of the deepest MCP integrations in the ecosystem, supporting over 70+ extensions that connect to databases, APIs, browsers, GitHub, Google Drive, and other external services
Agent Client Protocol (ACP) Support: Functions as both an ACP server (enabling integration with editors like Zed, JetBrains, and VS Code) and supports ACP providers (Claude Code, Codex, Amp, Pi) to leverage existing AI subscriptions
Extension Architecture: Built-in extension manager with security scanning for malware detection before activation

3. Context Management System

Session Management: Persistent context-aware sessions that maintain conversation history and state across interactions
Memory Extension: Persistent knowledge storage system with local (project-specific) and global (user-wide) scope, supporting category-based organization and tag-based retrieval
Context Engineering: Comprehensive system for building background knowledge through .goosehints files, skills, persistent instructions, hooks, and prompt templates

4. Workflow Automation Capabilities

Recipes: Reusable YAML-based workflow configurations that package extensions, prompts, and settings together for consistent, shareable automation
Subagents: Isolated, parallelizable agent instances that can handle focused tasks while keeping the main conversation clean, supporting both sequential and concurrent execution patterns
Slash Commands: Custom shortcuts for quickly executing reusable instructions in any chat session

5. Security Architecture

Adversary Mode: Independent agent reviewer that monitors tool calls for potentially harmful actions
Prompt Injection Detection: Built-in protection against malicious command injection
Permission Controls: Configurable permission modes (autonomous, manual approval, smart approval) with granular tool-level access control
Sandbox Mode: Isolated execution environment for enhanced security

Primary Use Cases

Development & Engineering

Code generation, refactoring, and review with developer extensions
API development following standardized patterns (RFC 7807, JWT authentication, etc.)
Automated testing and quality assurance workflows
CI/CD pipeline integration via recipes and CLI

Research & Analysis

Multi-source information gathering with browser and database extensions
Data analysis and visualization with Auto Visualiser extension
Academic research assistance with citation management and literature synthesis

Content Creation & Documentation

Technical documentation generation with context-aware formatting
Marketing copy creation with brand voice consistency
Report generation with automated data extraction and visualization

Enterprise Automation

Internal tool integration via custom MCP extensions
Security compliance automation with adversary mode monitoring
Team knowledge management through shared recipes and memory systems

Goose

What is Goose?

Categories & Use Cases

Product Demos