How-to · 17 Nov
How-to · 2 min read
How to run an AI coding agent on your own machine
If you want an AI coding agent on your own computer, with your own model and no account, you need one that keeps the code and the decisions with you. Many agents live in someone else's cloud and edit your files the moment they decide to. That is fine for a toy project. It is a problem for a codebase you are responsible for, where a wrong change or a leaked key costs real time.
- Published
- 24 Oct 2026
- Delivery
- No. 8
- Topic
- AI and agents
- Evidence
- 10 claims quoted from albertcode.dev
01
Why it's hard
Hosted agents usually need an account, send your code to their servers and keep logs you cannot see. Agents that do run locally often write straight into your working tree, so you find out what they changed after the fact and clean up by hand. Running the tests is left to you, which means an edit can look right and still break the build. And plugging in a local model is often an afterthought, so you end up tied to one provider anyway.
This part describes the usual way. It makes no claims about AlbertCode SWE, so nothing is quoted here.
02
How AlbertCode SWE does it
AlbertCode SWE Agent runs as a small service on your own machine, with a terminal, a browser workspace and a VS Code panel on top. You connect the model you want, a hosted provider with your key or a local model server, and then nothing leaves your machine. Each change is planned first, built in an isolated copy of your repository and checked with your project's own tests. You approve the plan and then the diff, and only then does anything reach your files.
AlbertCode runs as a small service on your own computer, with the terminal, the browser workspace and the VS Code panel on top.Quoted from albertcode.dev
A hosted provider with your key, or a local model server — then nothing leaves your machine. No telemetry, no account.Quoted from albertcode.dev
Nothing reaches your files until you accept it.Quoted from albertcode.dev
03
Step by step
- 1Install AlbertCode on macOS, Windows or Linux. There is no account to create.
- 2Connect a model: a hosted provider with your own key, or a local server such as Ollama or LM Studio.
- 3Describe the change in plain words and read the plan it proposes: files, steps, tests and risk.
- 4Approve the plan, then let it build the change in an isolated copy and run your own tests.
- 5Read the diff and the test results, then accept it, ask for changes or discard it.
04
Tips
Use a local model for code you cannot send out at all. With Ollama, LM Studio or a similar server, the whole loop stays on your computer.
or Ollama, LM Studio and friends, so nothing leaves your machine.Quoted from albertcode.dev
Keep the default Governed mode for changes you care about. It gives you two gates, the plan and the diff, and you can switch to Direct or Ask mode for small edits or plain questions.
Choose the pace per request: two gates for changes you care about, direct edits you watch, or questions that change nothing.Quoted from albertcode.dev
Make sure your test suite actually runs before you start. The agent judges a change by your own checks, so a project with no working tests gives it nothing to prove against, and a green result means much less.
the project's own checks decide what passed.Quoted from albertcode.dev
Treat accepted changes like any other commit. They land as ordinary edits, so review them with git diff and undo them with Git if needed.
Accepted changes are ordinary edits in your working tree: review them with git diff, commit them, undo them with Git.Quoted from albertcode.dev
05
Questions
Do I need an account?
No. AlbertCode runs on your machine with no account and no telemetry.
macOS · Windows · Linux / no account / your model, or one running locallyQuoted from albertcode.dev
What happens to a change I reject?
It is deleted along with its isolated copy, so your working tree never sees it.
Rejecting or discarding a change deletes its isolated copy: your working tree never saw it.Quoted from albertcode.dev
Where are my API keys stored?
In your system's keychain or keyring. They are never passed to the commands a task runs.
Saved keys stay in your system's keychain or keyring, and are never handed to the commands a task runs.Quoted from albertcode.dev