Key takeaways
- 48K+ GitHub stars and foundation governance — Block donated goose to the Linux Foundation's Agentic AI Foundation (AAIF), where it anchors the agentic stack alongside MCP and AGENTS.md
- Local-first and MCP-native — runs on your machine as a CLI or desktop app, connects to 70+ extensions, and goes beyond code suggestions to install, execute, edit, and test
- Free and model-agnostic — Apache 2.0, bring your own keys for 15+ providers including local models via Ollama; no paid tier exists
FAQ
What is Goose?
Goose is an open-source, extensible AI agent that runs locally as a CLI or desktop app, going beyond code suggestions to install dependencies, execute commands, edit files, and run tests with any LLM.
How much does Goose cost?
Goose is free and open source (Apache 2.0). You bring your own API keys for providers like Anthropic, OpenAI, or Google, or run local models via Ollama — you pay only for inference.
Who maintains Goose?
Goose was created by Block (the company behind Square and Cash App) and was donated to the Agentic AI Foundation (AAIF) at the Linux Foundation, where it is an anchor project alongside MCP and AGENTS.md.
How is Goose different from Cline?
Cline lives inside VS Code with GUI approvals for every action; Goose is editor-independent — a standalone CLI and desktop app — and is governed by a neutral foundation rather than a venture-backed company.
Executive Summary
Goose is an open-source, extensible AI agent that runs locally and goes beyond code suggestions — it installs dependencies, executes commands, edits files, and runs tests with any LLM.[1] Created by Block in 2024 and written in Rust, it has grown to 48,000+ GitHub stars and 5,100+ forks as of June 2026, with daily commit activity.[1]
Its defining 2026 development is governance: Block donated goose to the Agentic AI Foundation (AAIF), the Linux Foundation body formed in December 2025 with Anthropic, Block, and OpenAI as founding contributors.[2] Goose is one of AAIF's three anchor projects alongside the Model Context Protocol and AGENTS.md, and the repository completed its move from block/goose to the aaif-goose organization on April 7, 2026.[3] That makes goose the only major coding agent governed by a neutral foundation rather than a vendor.
| Attribute | Value |
|---|---|
| Creator | Block (donated to AAIF / Linux Foundation)[2] |
| Founded | 2024 (repo created August 2024)[1] |
| Funding | N/A — foundation-governed open source; AAIF backed by platinum members including AWS, Google, Microsoft, Anthropic, OpenAI, and Block[2] |
| GitHub Stars | 48,000+ (June 2026)[1] |
| License | Apache 2.0[1] |
Product Overview
Goose is a local-first agent: you run it on your own machine as a terminal CLI or a native desktop app, point it at any LLM provider, and it automates engineering tasks end-to-end — from writing code to executing and testing it.[4] Extensibility is the core design bet: goose is MCP-native, connecting to 70+ extensions covering databases, APIs, browsers, GitHub, Google Drive, and more.[4]
Beyond interactive use, goose supports "recipes" — workflows captured as portable YAML configurations with parameters and subrecipes, shareable across teams and runnable in CI/CD — plus parallel subagents for independent tasks.[4]
Key Capabilities
| Capability | Description |
|---|---|
| Autonomous Execution | Installs, executes, edits, and tests — not just code suggestions[1] |
| MCP Extensions | 70+ extensions via the Model Context Protocol; build your own[4] |
| Recipes | Portable YAML workflows with parameters and subrecipes for team sharing and CI/CD[4] |
| Subagents | Parallel independent subagents for concurrent tasks[4] |
| Security Controls | Prompt injection detection, tool permission controls, sandbox mode, adversary reviewer[4] |
| Any LLM | 15+ providers, including local models via Ollama[4] |
Product Surfaces
| Surface | Description | Availability |
|---|---|---|
| Desktop App | Native GUI for macOS, Linux, and Windows[4] | GA |
| CLI | Terminal interface, single-command install[4] | GA |
| API | Embed the goose agent in other applications[4] | GA |
Technical Architecture
Goose is written in Rust and runs entirely on your machine — there is no hosted service in the loop; your code and prompts go directly to whichever LLM provider you configure.[1] The CLI installs with a single curl command, and the desktop app ships installers for all three major platforms.[4]
Installation:
curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash
Key Technical Details
| Aspect | Detail |
|---|---|
| Deployment | Local (CLI, desktop app, embeddable API)[4] |
| Model(s) | Any — Anthropic, OpenAI, Google, Azure, Bedrock, OpenRouter, Ollama/local (15+ providers)[4] |
| Integrations | 70+ MCP extensions; Agent Client Protocol (ACP) for leveraging existing subscriptions[4] |
| Open Source | Yes (Apache 2.0)[1] |
MCP-native by lineage: goose now sits in the same foundation as MCP itself — AAIF anchors MCP (10,000+ published servers), goose, and AGENTS.md (adopted by 60,000+ projects) as a coherent open agentic stack.[2]
Strengths
- Neutral governance — Foundation-owned under the Linux Foundation's AAIF, with Anthropic, Block, and OpenAI as founding contributors; no single vendor controls the roadmap[2]
- Strong adoption — 48K+ stars, 5,100+ forks, and daily pushes as of June 2026[1]
- Local-first privacy — Runs on your machine; code goes only to the LLM provider you choose[4]
- True model agnosticism — 15+ providers including fully local models via Ollama; no lock-in[4]
- MCP-native extensibility — 70+ extensions, and first-class alignment with the protocol's home foundation[4]
- Team automation — Recipes turn agent workflows into shareable, parameterized YAML runnable in CI/CD[4]
- Genuinely free — Apache 2.0 with no paid tier; you pay only for inference[4]
Cautions
- Harness quality criticized — Early HN reception (249 points, 68 comments) included reports of unwanted file edits across projects, a non-functional ComputerController extension for many users, and unclear documentation[5]
- Setup friction — Skeptics found configuration fiddly compared to single-purpose tools, and questioned whether prompt-composition overhead beats doing simple tasks manually[5]
- Token costs are yours — BYOK-only means agentic loops bill directly against your API key; HN users flagged per-action costs discouraging side-project iteration[5]
- Differentiation questions — Commenters questioned what goose offers over Cursor, Cline, and Aider; the foundation move answers this on governance, not necessarily on product[5]
- No enterprise tier — Community support only; no SLAs, SSO, or vendor accountability — a feature of foundation governance, but a gap for procurement-driven buyers
- Transition turbulence — The April 2026 repo and docs migration left some technical issues being resolved via Discord[3]
Pricing & Licensing
| Tier | Price | Includes |
|---|---|---|
| Goose | Free | CLI, desktop app, API, all extensions (Apache 2.0)[4] |
| LLM API | Varies | Pay your chosen provider directly (BYOK), or run local models for free |
Licensing model: Free and open source (Apache 2.0), governed by the Agentic AI Foundation at the Linux Foundation — no commercial tier, no subscription.[4]
Hidden costs: Agentic loops with frontier models consume tokens quickly; HN users cited per-action costs as a real consideration for heavy iteration.[5] Local models via Ollama eliminate API spend at the cost of capability.
Competitive Positioning
Direct Competitors
| Competitor | Differentiation |
|---|---|
| Cline | VS Code-native with GUI approval for every action; goose is editor-independent (CLI + desktop) and foundation-governed |
| Aider | Terminal-only pair programmer in maintenance mode; goose is multi-surface and actively developed with daily pushes[1] |
| OpenHands | Platform-style autonomous agent with a hosted cloud; goose is local-first with no hosted service |
| Claude Code | Anthropic-native CLI from a model vendor; goose is model-agnostic and vendor-neutral |
| Codex CLI / Gemini CLI | First-party open-source CLIs from OpenAI and Google; goose is the neutral, any-model alternative under shared governance |
When to Choose Goose Over Alternatives
- Choose Goose when: You want a local-first, model-agnostic agent with desktop and CLI surfaces, MCP extensibility, and a roadmap no single vendor controls
- Choose Cline when: You live in VS Code and want explicit GUI approval for every file change and command
- Choose Aider when: You want the most minimal terminal workflow with git-native commits and accept a slower release pace
- Choose OpenHands when: You want cloud-hosted autonomous agents rather than a local tool
- Choose Claude Code when: You're committed to Claude models and want a model vendor's first-party tooling
Ideal Customer Profile
Best fit:
- Developers who want one agent across terminal and desktop without committing to an editor
- Teams standardizing on MCP who want an agent from the protocol's own foundation
- Organizations with open-source governance requirements (foundation-owned, Apache 2.0)
- Privacy-sensitive users running local models via Ollama
- Teams encoding repeatable workflows as shareable recipes in CI/CD
Poor fit:
- Buyers needing enterprise SLAs, SSO, or a vendor to call
- Developers who want zero-config polish — community reports describe setup friction[5]
- Teams wanting cloud-hosted background agents rather than a local tool
Viability Assessment
| Factor | Assessment |
|---|---|
| Financial Health | Strong — No burn risk; AAIF is backed by platinum members AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, and OpenAI[2] |
| Market Position | Contender — 48K+ stars puts it among the most-starred open coding agents, behind Cline (63K+) and ahead of Aider (~46K)[1] |
| Innovation Pace | Active — Daily pushes as of June 2026; subagents, recipes, and security controls shipped[1] |
| Community/Ecosystem | Strong — 5,100+ forks, 70+ extensions, and now a shared foundation with MCP's 10,000+ server ecosystem[2] |
| Long-term Outlook | Positive — Foundation governance removes single-vendor risk; the open question is product velocity against well-funded first-party agents |
Goose's move to AAIF is the strongest structural position of any open coding agent: it cannot be acquired, pivoted, or abandoned by a single company, and it shares a roof with MCP and AGENTS.md — the standards layer of the agentic ecosystem.[2] The risk is the inverse of its strength: foundations guarantee continuity, not velocity, and goose competes with agents whose vendors ship weekly with frontier-model co-design.
Bottom Line
Goose is the vendor-neutral choice among AI coding agents: local-first, genuinely free, model-agnostic, and now governed by the same Linux Foundation body that stewards MCP and AGENTS.md. Its 48K+ stars and daily release cadence show real momentum, while early criticisms of harness polish and setup friction are the honest counterweight.[5]
Recommended for: Developers and teams who want a foundation-governed, any-model agent with CLI and desktop surfaces, MCP extensibility, and shareable recipe workflows — especially those with open-source or data-locality requirements.
Not recommended for: Enterprises that need SLAs and vendor support, or developers who want the most polished out-of-box experience and don't mind vendor lock-in to get it.
Outlook: The AAIF donation makes goose the default "commons" agent of the MCP era. If foundation governance attracts contributions from its platinum-member backers rather than slowing decisions, goose is positioned to outlast venture-backed rivals; watch whether product velocity holds now that Block no longer owns the roadmap.
Research by Ry Walker Research • methodology