OpenCode setup for small business teams: uses and fit
How small businesses use OpenCode, a coding agent run on their own machines, and how MILabs sets it up.
Updated 2026-07-21
What OpenCode is
OpenCode is a coding agent made by Anomaly. A developer runs it from a terminal, points it at a code repository, and asks it to read files, explain code, propose changes and run commands in that project. It can be pointed at different model providers, depending on the setup, and because it is self-hosted it can run on hardware the team controls. For a business, the value lies in the work it takes off a development team's list: reading unfamiliar code, drafting changes, writing tests and tidying up old scripts. The developer still reviews every change before it is merged. It suits teams that already keep their software in version control and have someone who can read the output critically. This page describes what a business is likely to use it for, and the specifics come from the first conversation.
OpenCode in plain language
A coding agent is a program that works inside a software project by reading its files, proposing edits and running commands, while a person decides what gets accepted. Agents in this group differ mainly in where they run, which language models they can call, and how much control the team keeps over the environment. OpenCode can be hosted on hardware a team controls, which shapes the questions a business settles before it starts.
The first question is where the code goes. When the agent runs on a laptop or server the business controls, project files are read from that machine. Whether anything leaves it depends on the model the agent calls. A hosted model receives the prompts and any code included with them, and the terms governing that data belong to the model provider.
The second question is how far the output can be trusted without review. An agent can propose a change that reads well and still miss an edge case, call a helper that does not exist in the codebase, or edit a test so that it passes. The practical effect is that the agent moves effort rather than removing it: less time on a first draft, more time reading one closely.
What businesses use it for
- Explaining inherited code so a new developer can get up to speed quickly
- Drafting small changes and bug fixes for a developer to review before release
- Writing and updating tests for features that already exist
- Tidying old scripts, moving legacy configuration and reducing technical debt
- Documenting how internal tools work so the rest of the team can follow them
Who it suits
- Engineering teams whose software already lives in version control
- Technical founders looking for support on everyday code tasks
- IT managed services providers who maintain client codebases
How MILabs sets it up
MILabs starts with the owner. Together you agree which repositories the tool may read, which model provider it will call, and whether any code or prompts may leave your own systems. The tool is then installed on the machine or server the team chooses and connected to the version control system the business already uses. Permissions are set so it reads only the repositories it needs, and every proposed change goes through a review step before it reaches the main branch. Before handover, the owner agrees who looks after the setup, who holds the access keys, and who decides when the model or tool version changes. The owner and one named developer then receive a short written summary of those decisions.
Builds it fits
Industries it fits
OpenCode in practice
Inside a team that uses OpenCode, much of the work starts with a question put to the code rather than a request for a finished change. A developer might ask where a discount is calculated across the checkout module, or why a nightly export writes duplicate rows. The agent traces the path through the files and reports what it found, and the developer decides which parts of that answer to act on. The useful habit is to ask for the reasoning as well as the edit, since the explanation is often easier to check than the diff.
A second pattern appears when a change is ready for review. The developer hands the agent a narrow task, such as adding a validation rule to one form or moving a configuration value out of a script, and asks for a test that fails before the change and passes after it. The review then concentrates on three things: whether the test checks the right behaviour, whether the change touches anything outside the task, and whether the naming matches the rest of the project.
Frequently asked questions
- Where does OpenCode store data?
- OpenCode runs on the machine or server where it is installed, so the files it reads stay under the business's control. Data sent beyond that depends on the model provider chosen in setup. When a hosted model is called, the prompts and code sent with them are handled by that provider under its own terms. MILabs records which provider is used and confirms it matches what the owner agreed before handover.
- When is OpenCode the wrong choice?
- OpenCode is a weaker fit when the business has no software it maintains in version control, or when the work is mostly office administration, such as email, scheduling and spreadsheets. It also suits a team poorly when nobody on staff can read code changes critically, because the output still needs a reviewer. In those cases a no-code workflow tool or a general assistant is the better fit, and MILabs will say so in the first conversation rather than a setup that sits unused.
- Can my team use OpenCode without training?
- No, a team needs some onboarding first. A member must be comfortable working in a terminal and reading a code change before accepting it. Staff who mainly write documents or manage customers will find the tool hard to use without a developer beside them. MILabs covers the setup, the permissions and a written note on how the business has chosen to use the tool, so everyone relying on it knows its limits.
- Can MILabs build on top of OpenCode?
- Yes, where the work calls for it. MILabs can build custom development around a business's own codebase, using OpenCode for drafting and review tasks while the team keeps ownership of the code. Any build is scoped with the owner first, and the business keeps the repositories, the access and the decisions about what ships. Where another tool fits the job better, MILabs will recommend it rather than force OpenCode into the work.
- What does an OpenCode setup need from me?
- It needs a named owner who can make decisions about access and model use, and a developer who will review each proposed change. You also need read access to the repositories the tool should work on, and a decision on which model provider the setup will use and whether code may leave your systems. Send MILabs the list of those repositories before the setup conversation so the scope is clear from the start.
