Updated on
Goose: reviews and analysis
Block's open agent, extensible and built to automate beyond code.
Affiliate link · no extra cost to you
Goose is Block's open source agent with an ambition wider than code: automating whole tasks by connecting to your tools through extensions. It runs locally, with the model of your choice, and its extension system is among the most mature in the open ecosystem. As a pure programming assistant the specialists outpace it: its ground is broad automation.
Best for: Teams wanting to automate complete workflows, not just write code.
What the internet says
The community rates it as a general agent rather than a code-only one, with more than seventy extensions and fifteen model providers, including running it entirely against a local model. The handover to a Linux Foundation body removes the single-company risk, and the criticism left is about rough edges rather than fundamentals.
What the web repeats in favour
- Free under Apache-2.0 with no paid tier: you pay only the usage of the provider you choose
- It can run entirely against a local model, with no keys and no calls off your machine, which works for sensitive code
- More than seventy MCP extensions and fifteen model providers, with a desktop app and a command line
What the web repeats against
- Being a general-purpose agent it is not as tuned for code work as the specialised ones, and it shows on long tasks
- Initial setup asks you to pick provider, model and extensions, so getting started costs more than with a closed tool
- With no in-house subscription there is no spending cap: the real cost is set by whichever model you connect
Sweep sources: GitHub · Communities · Communities · Review sites
Pros / Cons
Pros
- Mature, open extension system
- Runs locally with your chosen model
- Backed by a large company
Cons
- Less specialized in pure coding than rivals
- Needs configuration to perform
TLDR: Goose is Block’s open source agent with an ambition wider than code: automating whole tasks by connecting to your tools through extensions. It runs locally, with the model you choose, and its extension system is among the most mature in the open ecosystem. As a pure coding assistant the specialists outpace it: its ground is broad automation.
What Goose is and how it works
Goose comes out of Block (the company behind Square and Cash App), which built it to automate its own engineers’ work and released it as open source. That provenance explains its character: it is not a code-writing assistant but a task-executing machine, of which code is one part.
Its architecture revolves around extensions: connectors giving it access to specific tools (the browser, internal APIs, whatever services your flow uses), built on the open standard the industry has adopted for connecting agents to tools. With the right extensions, the task “pull the errors from the monitoring dashboard, find the guilty commit and prepare the fix” is a commission, not a script.
It runs locally (desktop or CLI), with the model you choose and your key, keeping data control on your machine.
What it is like day to day
The value appears when automating flows that cross tools: the engineering errands touching three systems that nobody enjoys. With each system’s extension configured, the agent chains them with enough judgment for routine work.
As a pure programming assistant (editing a codebase with fine judgment), specialists like Claude Code perform better: Goose delivers without it being its home ground. The fair comparison is not against them but against the scripts you would write by hand to automate: against those, a configurable agent gains in flexibility what it loses in predictability.
The initial investment is real: choosing a model, configuring extensions and tuning permissions is an afternoon of work, and the return arrives afterwards, with every automated flow that stops being done by hand.
Pricing and plans
Free and open source, no plans. The cost is the consumption of whichever model you connect, with the local option for zero cost and data at home. For volume automations, pairing with economical models keeps the bill in cents.
Who it is for (and who it is not for)
Goose is for teams and developers wanting to automate complete flows that cross tools, with local control and no provider lock-in. A large company’s backing gives it a stability other open projects envy.
It is not for anyone after the best pure coding assistant, nor for anyone wanting results without configuration: here the power is paid for in setup.
Goose alternatives
OpenCode is the code-focused open option with a more polished interface. Claude Code is the commercial specialist when the work is the codebase. Aider covers the manual, cheap approach. They are in Goose alternatives and the ranking in the best AI coding tools.
Frequently Asked Questions
How does it differ from a normal coding agent?
In ambition: coding agents edit repositories, Goose automates tasks that cross tools (browser, APIs, services) through extensions. Code is part of its ground, not all of it.
Which models does it support?
The usual providers’ with your key, plus local models for zero cost and data at home. Model choice per task is yours, with the usual pattern of cheap for the mechanical.
What does it cost?
The tool nothing: it is open source backed by Block. You pay the consumption of whichever model you use, which with economical or local ones stays in cents or at zero.
Is it hard to set up?
It needs real setup: model, extensions and permissions. That is an afternoon’s work for a technical profile, and the investment returns with every flow that stops being done by hand.
Best alternatives to Goose
Guides that use Goose
How to automate your daily work with free CLI agents
Guide to automating with free agents: Gemini CLI and Qwen Code in scripts and scheduled tasks…
How to automate HR paperwork with AI: letters, certificates and templates
Guide to people administration with AI: templates with variables, batch drafting with review and…
How to digitize invoices and receipts with AI and never type a figure again
Guide to digitizing accounting documents with AI: capture on the spot, extraction with Dext or…