MILabs
Tools

goose AI agent: self-hosted setup for small business

How goose, an open-source self-hosted AI agent from Block, fits small business work and how MILabs sets it up.

Updated 2026-02-16

What goose is

goose is an open-source AI agent built by Block. It is self-hosted, which means it runs on a computer the business controls rather than only inside a vendor's cloud. A user gives goose a request in plain language, and the agent works out a sequence of steps and carries them out with the tools it has been given. Those tools are added as extensions, and many connect through the Model Context Protocol, so the agent can read files, query data sources or call other software. goose can connect to several language model providers, and the business chooses which one suits each job. Because the software runs on the business's own infrastructure, the owner decides where data is stored and who can reach the agent. Small teams tend to use goose for repeatable technical and office tasks, where a person reviews the output before it leaves the business. It suits work that is clearly defined, involves files or systems the business already holds, and has someone available to look after it.

Block: official site

goose in plain language

An AI agent is a program that takes a goal, decides on a sequence of actions, and carries them out, rather than producing a single reply to a single question. A chatbot answers. An agent reads a file, looks at the result, chooses the next step and keeps going until the task is finished. goose follows that pattern. Underneath it sits a language model, which reads text and generates instructions. goose hands that model the available tools, receives the requests it makes, runs them, and feeds the results back in.

The location of that model matters. A model can be hosted by a provider over the internet, or it can run on a machine the business owns. goose can be pointed at either kind, so one agent may send some text to an outside service and keep other text on local hardware. That choice shapes both privacy and running cost.

Tools are the other half. Each capability is an extension with its own permissions, and an agent is only as capable as the extensions it has been given. Reading one folder and writing to one calendar are separate grants, and each one deserves its own decision.

What businesses use it for

  • Shaping documents drawn from the business's own folders into clean drafts
  • Searching internal data sources and turning results into summaries for staff
  • Repeating the same checks across spreadsheets, forms and exported reports
  • Tidying routine files and records so a person can review them quickly
  • Helping developers and IT staff with scripts, configuration and tests

Who it suits

  • Owners who want an agent running on hardware the business controls
  • Technical teams at ease maintaining software and reviewing its extensions
  • Office teams with repetitive file and system work that a person checks

How MILabs sets it up

Setup begins with a conversation with the owner about the work goose should take on and which deployment fits it. MILabs then chooses with the business the computer or server that will host goose, and connects the systems the business already uses, such as its file store, accounting software or booking calendar, through extensions. Permissions are set so the agent reads only what each task needs, and any write access stays limited and logged where the setup allows it. Before handover, the owner and MILabs agree in writing who looks after goose and who approves changes to its extensions. Keeping goose current, reviewing its access and checking its output rests with that named person.

Builds it fits

Industries it fits

goose in practice

In a working business, goose usually begins as one narrow job rather than a general assistant. A team might point it at a single shared folder of supplier invoices and ask it to pull out the supplier name, date, total and tax line into a list for someone to check. The request is written in ordinary language, and the agent works through the files one at a time, opening each and noting anything it could not read cleanly.

The review step is part of the job. Output lands in a holding location, and a person decides what moves into the accounting system. Errors show up clearly this way, since a misread date or a missing line stands out against the source document. Over time, the team learns which kinds of file need extra attention, and the instructions given to the agent are refined to match.

Access stays tight. The agent might have read access to one folder and write access to a single output location, with no ability to delete or send anything. Each new task gets its own decision about what it may touch, and a short written note records what was granted and why.

good questions, straight answers

Frequently asked questions

What are the limits of goose?
goose gets its strength from the setup around it. It runs on hardware the business chooses, so its speed and reach depend on that machine and the language model connected to it. Each extension reaches only the systems it has been granted, which keeps every task scoped. Teams that put a person in the review step tend to get the most reliable output from the agent.
Who looks after goose after setup?
The business does, and MILabs puts that in writing before handover. The agreement names a person who keeps goose updated, reviews the installed extensions, checks permissions regularly and decides how output issues are handled. The owner keeps authority over access to data, and the team that inherits the system receives notes on how it was configured. Responsibility stays clear and easy to find for everyone on the team.
What should I check on the vendor's site before using goose?
The project's documentation at block.github.io/goose sets out the current feature set, supported interfaces and installation steps. Its release notes show what each version adds. Those pages are the clearest guide to what goose can do today, so they make a good reference for the conversation. The easiest first step is to say hello through the enquiry form, and we'll take it from there.
How do I know whether goose is paying off?
Compare the work goose handles with what the team did before. Pick two or three tasks, such as sorting incoming documents or preparing a weekly summary, and record how long they took and how many needed correction before the agent was used. Then measure again after the same tasks have run through goose for a while. Judge the result against the business's own records, which tell the clearest story.
Which industries use goose?
Businesses that handle sensitive records and want their tools to stay on their own hardware are the usual fit. Examples include accounting and bookkeeping firms, legal services, insurance brokers, IT managed service providers, engineering consultancies and local councils. Each sector brings its own data and its own rules, which shapes how the work is designed. The owner and the team that will run it decide which tasks come first.
let's make something happen

Workflow automation: how we can help

Choose how you want to work on this with MILabs. Both options start with an enquiry.

Discovery is free, no obligation from the first chat.

Let us build it for you

We build it and look after it.

Build it yourself, we'll help

You build it with our direction, teaching and a weekly check-in.