Goose (Block)

Goose was created inside Block (Square/Cash App's parent company) and open-sourced from there. In December 2025 the Linux Foundation announced the Agentic AI Foundation (AAIF), with Anthropic's MCP, OpenAI's AGENTS.md and Block's Goose as founding contributions; the actual repository move happened in April 2026, from github.com/block/goose to github.com/aaif-goose/goose, with docs moving to goose-docs.ai. The old URLs (block.github.io/goose, github.com/block/goose) still return HTTP 200 but are now client-side redirects to the new domain — worth knowing if you have anything bookmarked or scripted against the old paths. The New Stack reported the reason wasn't just optics: Manik Surtani, AAIF's CTO and former head of open source at Block, said Block's retention of the Goose trademark had been "hampering enterprise adoption" — companies were wary of building on a project one vendor could still unilaterally control. Under AAIF it's now a Series of LF Projects, LLC, governed by a documented MAINTAINERS/GOVERNANCE process rather than one company's roadmap.

Goose is Apache-2.0 licensed and free to install (desktop app for macOS/Linux/Windows, or CLI). It has no hosted inference and no subscription — Goose's own docs state plainly that it's "free and open source... but not all supported LLM Providers provide a free tier," meaning your real cost is whatever you pay Anthropic, OpenAI, Google, Groq, a local Ollama model, or one of 15+ other supported providers directly. Apache-2.0 also means you can embed or fork it commercially without a source-disclosure obligation, unlike some agent projects with source-available or non-commercial terms.

Goose supports four execution modes: Autonomous (modifies files, runs commands and can delete files without asking), Manual Approval (confirms every tool call), Smart Approval (an LLM-based classifier auto-approves what it judges low-risk and flags the rest), and Chat-only (no tool access at all). The documentation is explicit that Autonomous is the default. The project's own SECURITY.md acknowledges the risk directly, warning that "developer agents have the ability to run code and take actions on your computer" and pose a different risk profile than a chat-only LLM. This is a real trade-off, not a defect: Smart or Manual mode is one config change away (goose configure, /mode mid-session, or GOOSE_MODE env var), but a first-time user who skips that step is running an agent with standing permission to edit or delete files unattended.

As of today, the repo has 53,063 stars, 6,046 forks and 244 open issues (github.com/aaif-goose/goose, live API). It was created in August 2024 and is under continuous development — the latest tagged release is v1.46.0 (published 2026-08-12), and pull requests are still being merged same-day (e.g. PR #10940 merged 2026-08-20). Goose's own docs page cites "500+ contributors," a figure we could not independently verify against the API in this pass (GitHub's unauthenticated rate limit was exhausted) — treat that specific number as vendor-stated rather than confirmed.

Goose ships 8 built-in extensions (data visualization, JS execution, computer control, developer tools, memory, subagent delegation, persistent instructions, a tutorial system) plus 50+ community MCP extensions covering things like GitHub, MongoDB, Figma and various media-generation services, per goose-docs.ai's extensions page. It uses the Model Context Protocol, so it can also reach any general-purpose MCP server beyond that curated list.

There is no published enterprise support tier, SLA, or paid support plan — Goose's docs point only to GitHub Issues and a Discord server. That's consistent with its status as a Linux Foundation project rather than a vendor-backed product; if you need a contracted support relationship, one is not currently on offer from AAIF or Block.