Goose Review: Block’s Free Open-Source AI Agent for Everything Local

What Goose Is

Goose is an open-source AI agent that runs on your machine as a desktop app, a terminal CLI, or an embeddable API. Built in Rust and originally open-sourced by Block, it was contributed to the Agentic AI Foundation (AAIF) in April 2026, which signals long-term community stewardship. Based on published project data (30,000–50,000+ GitHub stars) and third-party reviews, Goose’s pitch is breadth: it’s not just a coding assistant but an agent environment for local work.

Point Goose at a task and it can edit files, install packages, execute shell commands, run test suites, and chain multi-step workflows — in code, but also in research, writing, data analysis, and system automation. Extensibility comes through MCP (Model Context Protocol): Goose ships extension servers for developer tools, computer control, and memory, and you can add your own. It supports a huge provider list — OpenAI, Anthropic, Google/Gemini, Ollama, OpenRouter, AWS Bedrock, Azure, xAI, and more — plus it can delegate to external CLI agents like Claude Code or Codex.

What’s Actually Free

Goose is free and open source under Apache 2.0 — the desktop app, CLI, and API all cost nothing, with no subscription tier and no usage metering by Goose. Your only cost is the model:

• $0 with local models — Ollama plus a code-capable model (e.g. Qwen3-Coder) gives you a completely free, fully local coding agent: no API keys, no cloud, everything on your machine.
• Pay-per-token otherwise — bring any provider key; Goose takes no cut.
• Free provider tiers — several supported providers offer free API allowances you can plug in.

Install is a one-liner (Homebrew or the install script), and the desktop app makes it approachable for developers who don’t live in the terminal.

Key Features

• Desktop app + CLI + API — one agent surface across GUI, terminal, and embeddable use, with shared config.
• Provider-agnostic — 15+ LLM providers plus local models; swap freely.
• MCP extensions — built-in servers for dev tools, computer control, and memory; add your own.
• Beyond-code actions — install dependencies, run commands, execute tests, automate local workflows.
• Recipes — shareable, reusable task definitions for common multi-step jobs.
• Session management — resume, export, and inspect past sessions with token totals.
• Delegation — hand off to external CLI agents (Claude Code, Codex, Cursor Agent) when useful.
• Rust-built — fast, single-binary distribution with modest resource use.

Pros and Cons

Pros

• Completely free and open source with corporate-grade backing (Block, now AAIF).
• Local-first: a genuinely private $0 setup is achievable with Ollama.
• Goes beyond code — the same agent automates surrounding workflows.
• Desktop app lowers the barrier vs. terminal-only agents.
• MCP extensibility is best-in-class for a free tool.

Cons

• Broader than coding-only tools — less focused if you just want completions.
• Newer and rougher around the edges than Aider for pure multi-file refactoring, per some reviews.
• The Block-to-AAIF transition adds mild uncertainty about future governance.
• Agent quality still depends on the model you choose; local models lag frontier ones.
• More setup and provider decisions than a bundled commercial tool.

Who It’s Best For

Goose is best for developers who want one free, local, extensible agent for coding plus the automation around it — running tests, managing environments, chaining tools via MCP. It’s also a strong pick for privacy-sensitive work and for regions with poor API access, since the local-model path needs nothing from the cloud. It’s a weaker fit if you want a narrow, polished coding-only assistant, or a fully managed cloud product with support SLAs.

FAQ

Is Goose really free?

Yes — Apache 2.0 open source, no subscription, no usage metering. You pay only for model API calls, or $0 with local models.

Who makes Goose?

Goose was built and open-sourced by Block (Square/Cash App’s parent company) and contributed to the Agentic AI Foundation in April 2026, where it’s now community-stewarded.

How is Goose different from Aider?

Aider is a git-native terminal pair programmer focused on code editing. Goose is a broader local agent environment — desktop app plus CLI — that handles code plus surrounding workflows, with deeper MCP extensibility.

Can Goose run fully offline?

Yes — pair it with Ollama and a local model for a fully offline, $0 setup. Quality depends on the local model’s capability.

Does Goose work on Windows?

Goose ships for macOS, Windows, and Linux via the desktop app and CLI installers.

Bottom Line

Goose is the free agent for people who think bigger than autocomplete: local-first, provider-agnostic, extensible through MCP, and backed by serious open-source stewardship. Based on specs, pricing, and early reviews, pair it with a local model for a private $0 setup, or with a frontier API key when you need maximum brainpower — either way, the software itself never charges you.