Skip to main content
PromptQuorum
Home/Power Local LLM/Goose Review 2026: Block's Open-Source AI Agent, Now Under the Linux Foundation
Local AI Agents & Tool Use

Goose Review 2026: Block's Open-Source AI Agent, Now Under the Linux Foundation

Β·10 min readΒ·By Hans Kuepper Β· Founder of PromptQuorum, multi-model AI dispatch tool Β· PromptQuorum
Read in:
πŸ‡ΊπŸ‡ΈenπŸ‡©πŸ‡ͺdeπŸ‡«πŸ‡·frπŸ‡―πŸ‡΅jaπŸ‡¨πŸ‡³zhπŸ‡ͺπŸ‡ΈesπŸ‡§πŸ‡·ptπŸ‡ΈπŸ‡¦arπŸ‡°πŸ‡·ko

Goose is a free, open-source, local-first AI agent β€” desktop app, CLI, and API β€” built by Block and publicly launched January 28, 2025, then donated in December 2025 to the Agentic AI Foundation (AAIF) at the Linux Foundation. It is Apache 2.0-licensed, confirmed via the GitHub API on September 12, 2026, and its repository (github.com/aaif-goose/goose) shows 54,156 stars as of that date. Goose itself has no license fee β€” connect it to Anthropic, OpenAI, Google, a local Ollama model, or one of dozens of other supported LLM providers using your own API key, and extend it with Model Context Protocol (MCP) tools.

Goose (goose-docs.ai) is a free, open-source, local-first AI agent β€” available as a desktop app, a command-line tool, and an API β€” for coding, automation, and workflows. Block (the company behind Square, Cash App, Afterpay, and Tidal) built and publicly launched it on January 28, 2025, then donated it in December 2025 as a founding project of the Agentic AI Foundation (AAIF) at the Linux Foundation, alongside Anthropic's Model Context Protocol (MCP) and OpenAI's AGENTS.md. The code and governance formally moved to the AAIF on April 7, 2026, with the canonical repository now at github.com/aaif-goose/goose (the old block/goose URL redirects there). This review covers what goose does, how to install it, its pricing model, and how it compares to other open-source coding agents.

Goose Review 2026: Block's Open-Source AI Agent, Now Under the Linux Foundation

Key Takeaways

  • Goose is free and open-source; Apache 2.0-licensed per the GitHub API, checked September 12, 2026
  • Built by Block (Square, Cash App, Afterpay, Tidal) and publicly launched January 28, 2025
  • Donated by Block in December 2025 as a founding project of the Agentic AI Foundation (AAIF) at the Linux Foundation, alongside Anthropic's MCP and OpenAI's AGENTS.md
  • Code and governance formally moved to the AAIF on April 7, 2026; canonical repo is github.com/aaif-goose/goose (the old block/goose URL redirects there)
  • GitHub repository shows 54,156 stars as of September 12, 2026, verified via the GitHub API
  • Available as a native desktop app (macOS, Linux, Windows), a CLI, and an API
  • Connects to dozens of LLM providers with your own API key (Anthropic, OpenAI, Google, OpenRouter, Azure, Amazon Bedrock, and more), or a local Ollama model at no cost
  • Extensible via the Model Context Protocol (MCP), the same open standard Anthropic contributed to the AAIF

πŸ“ In One Sentence

Goose is a free, open-source, Apache 2.0-licensed, local-first AI agent β€” desktop app, CLI, and API β€” originally built by Block and now governed by the Agentic AI Foundation at the Linux Foundation.

πŸ’¬ In Plain Terms

Goose is a program that can read and edit files, run shell commands, and call outside tools on your behalf to finish a coding or automation task, instead of just answering questions in a chat window. You point it at an LLM (a local Ollama model, or a cloud API like Anthropic or OpenAI), and it carries out multi-step work β€” running tests, editing code, calling APIs β€” until the task is done.

πŸ“ŒNote: This review is the deep-dive companion to goose's entry in the Local LLM Software Directory β€” see that page for how goose compares at a glance to dozens of other local AI tools.

What Is Goose?

Goose is an open-source, local-first AI agent β€” a desktop app, command-line tool, and API β€” that reads and edits files, runs shell commands, and calls external tools to complete coding, automation, and workflow tasks. Its GitHub repository describes it as "an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM." Unlike a chat assistant that only suggests code, goose is designed to act: it can run in a loop, executing commands and checking results, until a task is finished.

  • Core function: an agent harness that combines an LLM you choose, a set of built-in and extensible tools, and Model Context Protocol (MCP) integrations to carry out multi-step coding and automation tasks
  • Interfaces: a native desktop app (macOS, Linux, Windows), a terminal CLI, and a programmatic API β€” all built on the same underlying agent engine
  • Local-first design: goose itself runs on your own machine; the LLM it calls can also be local (via Ollama) or a cloud API of your choice
  • Developer: originally built by Block, the company behind Square, Cash App, Afterpay, and Tidal; now governed by the Agentic AI Foundation (AAIF) at the Linux Foundation
  • Canonical repository: github.com/aaif-goose/goose, Apache 2.0-licensed per the GitHub API

From Block to the Linux Foundation: A Short History

Goose's ownership structure changed twice within about 14 months of its public launch β€” first as an internal Block project made public, then as a donation to neutral open-source governance. The timeline below is built from Block's and the Linux Foundation's own announcements.

  1. 1
    January 28, 2025 β€” Public launch
    Why it matters: Block publicly released goose as an open-source, local-first AI agent, after building and using it internally.
  2. 2
    December 9, 2025 β€” Agentic AI Foundation announced
    Why it matters: The [Linux Foundation announced the AAIF](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation) with goose (from Block) as a founding project, alongside Anthropic's Model Context Protocol (MCP) and OpenAI's AGENTS.md β€” moving governance of a widely used agent project out of a single company's hands.
  3. 3
    April 7, 2026 β€” Repository and docs move to the AAIF
    Why it matters: Goose's canonical GitHub repository relocated from `block/goose` to [github.com/aaif-goose/goose](https://github.com/aaif-goose/goose), and its documentation moved to [goose-docs.ai](https://goose-docs.ai); the old URLs redirect, but this is the point where the AAIF, not Block alone, formally controls the project's governance.

What Can You Do With Goose?

Goose's feature set centers on acting on a task rather than only describing how to do it, per goose-docs.ai and the GitHub README.

  • Agentic task execution β€” goose reads and edits files, runs shell commands, and checks its own results in a loop until a coding or automation task is complete, rather than only returning a suggested diff
  • Bring-your-own-LLM β€” connect goose to your choice of LLM provider (Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Amazon Bedrock, and dozens more listed in the official provider docs) using your own API key, or point it at a local Ollama model for no cost
  • Model Context Protocol (MCP) extensions β€” goose connects to external tools, databases, and services (GitHub, Google Drive, browsers, custom internal tools) through MCP, the open standard Anthropic also contributed to the AAIF; goose maintains a directory of documented extensions at goose-docs.ai/extensions
  • Recipes β€” reusable, shareable YAML task definitions that let a team package a repeatable workflow (a code migration, a test run, a release checklist) so it can be re-run by anyone with goose installed
  • Three interfaces on one engine β€” the desktop app, CLI, and API all run the same underlying agent, so a workflow built in one interface is not locked to it
  • Self-hosted and auditable β€” because the code is Apache 2.0-licensed and now Linux-Foundation-governed, teams can audit, fork, and self-host goose without depending on a single vendor's continued goodwill

Usage Examples: Three Ways to Use Goose

These are concrete workflows built from goose's documented desktop, CLI, and extension features above β€” not hypothetical use cases.

Goose Pricing: Free Software, Bring Your Own Key

Goose itself has no license fee. Its GitHub repository is Apache 2.0-licensed, confirmed via the GitHub API, with no feature gated behind payment from the project itself. What you pay depends entirely on which LLM you connect goose to.

Goose software (desktop, CLI, API)

Cost:
Free
Notes:
Apache 2.0-licensed, no goose-side fee for any interface or feature

Local model via Ollama

Cost:
Free
Notes:
No API key needed; cost is only your own hardware and electricity

Cloud provider (Anthropic, OpenAI, Google, and others)

Cost:
Per-provider API pricing
Notes:
Billed directly by the provider you choose, at that provider's own rates β€” not set or marked up by goose

Because goose is bring-your-own-key, there is no single "goose price" beyond the free software itself β€” check your chosen provider's own pricing page before running a large or repeated task, especially with a frontier cloud model.

Goose vs. OpenHands

Goose and OpenHands are both free, open-source, self-hostable agent platforms that go beyond chat-only code suggestions, and both get compared as the natural starting point for teams that want an autonomous coding agent they fully control. The clearest differences are governance model and primary sandboxing approach.

Governance

Goose:
Linux Foundation (Agentic AI Foundation), donated by Block
OpenHands:
All Hands AI, a venture-backed startup

License

Goose:
Apache 2.0
OpenHands:
MIT

Interfaces

Goose:
Desktop app (macOS, Linux, Windows), CLI, API
OpenHands:
Web GUI, terminal UI, CLI

Sandboxing

Goose:
Runs directly against your machine or environment by default, with configurable permission prompts
OpenHands:
Runs agents inside a sandboxed Docker container by default

LLM model

Goose:
Bring your own key across dozens of providers, or a local Ollama model
OpenHands:
Bring your own key, or Claude Code, Codex, and Gemini CLI as agent backends

Extensibility

Goose:
Model Context Protocol (MCP) extensions, plus reusable YAML "recipes"
OpenHands:
Custom agent configuration and microagents

If you want an agent governed by a neutral, vendor-independent open-source foundation with a lighter default sandbox, goose's model fits. If you want an agent that isolates every run inside a Docker container by default, evaluate OpenHands directly β€” see the full OpenHands review for details. Both are free for self-hosted use; verify current features on each project's own site before deciding.

Who Should Use Goose?

Whether goose fits depends on whether you want an agent you fully control, self-host, and can point at any LLM you choose.

Competitors and Alternatives

Goose is most often compared to OpenHands, Cline, Aider, and Open Interpreter β€” other open-source tools that turn an LLM into an agent capable of acting on a codebase rather than only chatting about it.

OpenHands

Best known for:
Sandboxed, Docker-isolated autonomous coding agent platform
Link:
OpenHands review
Articles about OpenHands (3)

Also mentioned in:

Open Interpreter

Best known for:
Lets an LLM run code locally to complete general (not just coding) tasks
Link:
Open Interpreter review
Articles about Open Interpreter (2)

Also mentioned in:

This list reflects tools commonly compared to goose, not an independent PromptQuorum ranking β€” see the Local LLM Software Directory for the full catalog, including goose's own directory entry, and verify each tool's current pricing and feature set before choosing.

Common Mistakes When Evaluating Goose

Most confusion about goose comes from conflating it with a hosted, single-model chat product, or missing the December 2025/April 2026 governance change.

Frequently Asked Questions

What is goose?

Goose (goose-docs.ai, source at github.com/aaif-goose/goose) is a free, open-source, local-first AI agent β€” a desktop app, CLI, and API β€” for coding, automation, and workflow tasks, originally built by Block and now governed by the Agentic AI Foundation at the Linux Foundation.

Is goose free?

The goose software itself is free and Apache 2.0-licensed, with no feature gated behind payment. Cost depends on which LLM you connect it to: a local Ollama model is free to run, while a cloud provider (Anthropic, OpenAI, Google, and others) bills per that provider's own API pricing.

Is goose open source? What license does it use?

Yes. Goose's GitHub repository is licensed under Apache 2.0, confirmed via the GitHub API (license.spdx_id: "Apache-2.0") on September 12, 2026.

Who owns or governs goose now?

Block built and publicly launched goose on January 28, 2025. In December 2025, Block donated it as a founding project of the Agentic AI Foundation (AAIF) at the Linux Foundation, alongside Anthropic's Model Context Protocol (MCP) and OpenAI's AGENTS.md. The repository and governance formally moved to the AAIF on April 7, 2026.

What platforms does goose support?

Goose's desktop app is available for macOS (Apple Silicon and Intel), Linux (DEB, RPM, Flatpak), and Windows; its CLI installs on all three, including via Homebrew or a PowerShell script on Windows, per the official installation docs.

Does goose require a specific LLM?

No. Goose is bring-your-own-LLM: connect it to Anthropic, OpenAI, Google, OpenRouter, Azure, Amazon Bedrock, and dozens of other providers listed in its provider docs, or point it at a fully local model through Ollama.

What is the Model Context Protocol (MCP), and how does goose use it?

MCP is an open standard (contributed to the AAIF by Anthropic) for connecting an AI agent to external tools and data sources. Goose uses MCP for its extension system, letting it connect to services like GitHub, databases, and browsers through a documented directory of extensions at goose-docs.ai/extensions.

How many GitHub stars does goose have?

Goose's repository (github.com/aaif-goose/goose) showed 54,156 stars as of September 12, 2026, verified directly via the GitHub API. Check the repository directly for a current count, since it changes daily.

What are goose "recipes"?

Recipes are reusable, shareable YAML task definitions that let a team package a repeatable goose workflow β€” such as a code migration or a test run β€” so it can be re-run consistently by anyone with goose installed.

How does goose compare to OpenHands?

Both are free, open-source, self-hostable agent platforms. The main differences are governance (goose: Linux Foundation via the AAIF; OpenHands: All Hands AI, a startup) and default sandboxing (OpenHands isolates runs in Docker by default; goose runs closer to your own machine unless configured otherwise). See the full comparison above.

Sources

← Back to Power Local LLM