goose
A general-purpose open source AI agent that goes beyond code — desktop, CLI, and API
Evidence: AGENTS.md requires contributors to follow issue workflow, but does not explicitly require user confirmation or least privilege; however, documentation mentions ACP protocol supports requestPermission for user confirmation of sensitive operations, and Docker image runs as non-root user by default, reflecting least privilege. Data flow transparency: telemetry (PostHog) is mentioned but data flow details are not provided. Sensitive data handling: secrets stored in system keyring or file, but security measures not detailed. Dependency security: Cargo.toml pins some dependency versions, but no vulnerability scanning mentioned. External effects: goose can execute commands and modify files, but permission controls not explicit. Rollback: no rollback mechanism mentioned. Source attribution: PRs must link issues, but code provenance not explicit. Deductions: lack of explicit user confirmation mechanism, insufficient data flow transparency, dependency security measures not stated, no rollback mechanism.
Evidence: AGENTS.md and CONTRIBUTING.md provide detailed development workflow, testing, linting, etc., showing good internal consistency. Dependency availability: build dependencies and installation steps are listed, but availability guarantees not mentioned. Failure messages: no error handling or failure notification guidance. Deductions: dependency availability and failure messages lack concrete evidence.
Evidence: Documentation clearly describes multiple use scenarios (CLI, desktop, Docker, custom distributions), broad audience. Capability boundaries: extension mechanisms (MCP, recipes) and custom providers are described, but limits not explicit. Trigger precision: commands and configuration are detailed, but trigger conditions not covered. Environment fit: multi-platform build guides (Linux, WSL, Termux) and Docker support provided. Deductions: trigger precision and some capability boundaries not explicit.
Evidence: Information architecture is clear with README, CONTRIBUTING, GOVERNANCE docs. Install notes are detailed covering multiple platforms. Naming stability: project name and commands consistent. Examples and FAQ: quickstart and tutorial links provided, but no FAQ. Known limitations: some limitations mentioned (e.g., musl build disables features), but not systematically listed. License: Apache-2.0, clear. Versioning: version number in Cargo.toml, but no changelog. Maintenance responsibility: GOVERNANCE.md defines maintainer roles and processes. Deductions: missing FAQ and changelog, known limitations not comprehensive.
Evidence: Output usability: clear command and configuration examples enable quick start. Marginal value: goose offers multiple features and extensions, high value. Cost-benefit: open source and free, but requires LLM API keys, cost depends on usage. Deductions: no specific performance or efficiency data.
Evidence: Claims in documentation (e.g., supports 15+ providers, 70+ extensions) lack specific sources or verification methods. Cross-source corroboration: links to external docs and community, but no independent verification. Fact-inference separation: documentation does not clearly distinguish facts from inferences. Deductions: lack of traceable verification evidence.
- The AGENTS.md is primarily aimed at contributors, not end users, so trust-related details (e.g., permissions, data flow) may not be fully reflected.
- Telemetry (PostHog) mentioned in documentation may involve data collection; users should be aware and control it.
- Dependency security does not mention vulnerability scanning; users are advised to check dependency security themselves.
What does this agent do, and when should you use it?
goose is a general-purpose AI agent that runs on your machine, useful for more than just coding — also research, writing, automation, and data analysis. It ships as a native desktop app (macOS/Linux/Windows), a full CLI, and an embeddable API, built in Rust for performance and portability. goose works with 15+ model providers (Anthropic, OpenAI, Google, Ollama, Bedrock, and more) and connects to 70+ extensions via the Model Context Protocol, and is now part of the Linux Foundation's Agentic AI Foundation (AAIF).
Executes tasks locally — install, execute, edit, test — and can use your existing Claude, ChatGPT, or Gemini subscription (via ACP) or a direct API key from any provider; connects to extensions through MCP for extra capabilities (e.g. accessing specific tools or services), and supports building custom distributions with preconfigured providers, extensions, and branding.
- Handle writing or research work beyond coding, from a local terminal or the desktop app
- Reuse an existing Claude/ChatGPT/Gemini subscription instead of buying separate API credits
- Automation scenarios that need MCP extensions to connect specific tools (databases, internal systems, etc.)
- Build a branded, preconfigured internal distribution on top of the same base
What are this agent's strengths and limitations?
- General-purpose scope covers writing, research, and automation beyond coding
- Governed under a Linux Foundation project, offering more transparent governance and longer-term maintenance assurance than a single-company project
- Can reuse existing Claude/ChatGPT/Gemini subscription usage instead of forcing separate API billing
- Supports 70+ MCP extensions and a custom-distribution mechanism suited to internal team redistribution
- General-purpose positioning means it may go less deep on any single scenario (e.g. pure coding) than a specialized tool
- The real security boundary depends heavily on which extensions you enable, and needs per-extension review rather than a single overall judgment
- This review did not verify actual runtime behavior under any specific combination of extensions
How do you install or deploy this agent?
Download the desktop app from the official docs (goose-docs.ai's Getting Started page), or install the CLI directly: curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash.
How do you use this agent?
After installing, launch the desktop app or run the CLI, configure a model provider (API key or existing subscription), optionally add MCP extensions, then describe your task in natural language, e.g. "Summarize this log file and produce a list of issues."How does this agent compare with similar options?
Compared to coding-focused terminal agents like Codex CLI or opencode, goose's key differentiator is generality — it isn't limited to code tasks, and natively supports reusing a Claude/ChatGPT/Gemini subscription instead of forcing a separate API. If you need a local agent that spans more than just coding, or want to build an internally distributed variant on the same base, goose fits better.