Jack Dorsey's Goose: The Open-Source AI Agent

Block's open-source AI agent Goose is taking over developer workflows. Local-first, model-agnostic, MCP-ready — here's what it is and why it's exploding.

What is Goose?

Goose is an open-source, on-machine AI agent built by Block (Jack Dorsey''s company, the parent of Square and Cash App). Unlike chat-only assistants, Goose is designed to execute — it edits files, runs shell commands, calls APIs, and orchestrates multi-step work directly on your computer.

It was released publicly in early 2025 under the Apache 2.0 license and has quickly become one of the most-starred AI agent projects on GitHub. The pitch is simple: the power of Claude Code or Cursor''s agent mode, but model-agnostic, local-first, and fully open.

Why Goose is Different

Most AI tools lock you into one provider. Goose flips that:

Bring your own model. Plug in Anthropic Claude, OpenAI GPT-5, Google Gemini, or run fully local with Ollama.

Extensible via MCP. Goose was one of the first major agents to adopt the Model Context Protocol, meaning it can talk to any MCP server — GitHub, Slack, Postgres, Figma, you name it.

Runs anywhere. CLI, desktop app, or embedded in your own product.

Recipes. Reusable, shareable agent workflows defined in a single YAML file. Think of them as npm packages for agent behavior.

How Goose Has Grown

When Block open-sourced Goose in January 2025, it had a few thousand GitHub stars. Within months it crossed 15,000 stars, and as of April 2026 the repo has surged past [43,000+ GitHub stars](https://github.com/block/goose) — making it one of the most-starred AI agent projects in the world and home to a thriving community of contributors building extensions.

A few drivers behind the surge:

1. Jack Dorsey''s public advocacy. Dorsey has been vocal on X (@jack) about open-source AI being the only safe path forward, and he uses Goose daily. His posts routinely send waves of new users to the project.

2. The MCP wave. As Anthropic''s Model Context Protocol gained traction in late 2025, Goose was already there — making it the easiest agent to wire into the new MCP ecosystem.

3. Local-first appeal. Engineers tired of sending proprietary code to hosted SaaS agents found Goose''s "runs on your laptop with Ollama" story refreshing.

4. Block engineering at scale. Block''s own thousands of engineers use Goose internally, which means real production hardening — not a weekend project.

What People Are Building With It

Code refactors and migrations across massive monorepos

Data pipelines that read from Postgres, transform, and post to Slack

On-call assistants that triage alerts and open PRs autonomously

Custom internal copilots built on Goose Recipes, deployed company-wide

Why It Matters for Service Businesses

At 88 Labs AI, we watch the agent ecosystem closely because the underlying tech directly shapes what we can deploy for clients. Goose represents a meaningful shift: agent infrastructure is becoming a commodity. The differentiator is no longer "do you have AI" — it''s how thoughtfully it''s wired into your operations.

For roofers, HVAC contractors, real estate teams, and the other service businesses we work with, frameworks like Goose mean faster builds, lower lock-in, and agents that can be handed off and maintained without a vendor relationship.

The Bottom Line

Goose is doing for AI agents what Linux did for servers: making the powerful default the open one. Whether you''re a solo dev experimenting with local models or a CTO planning your next automation initiative, it''s worth your weekend.

Repo: github.com/block/goose