EN LV
Get Harness

Getting started

Install, activate, connect your own AI provider and get one routine finished — the shortest honest path from a fresh machine to a result you can check.

Overview

What this covers

This page is the reference version of the setup guide: installation, licence activation, connecting an AI provider, granting the first capability, and running one routine to a finished result.

It is written for the person who will own that routine day to day — not for an IT project. You do not need a server, a cloud account or a integration project to start.

How it works

How it works

Harness runs on one computer you control and acts as the supervision layer around an AI model:

  • The model reasons. It drafts, analyses and proposes what to do next, using the provider account you connect.
  • Harness controls. It exposes only the capabilities a routine needs, pauses defined actions for approval, and records the run from request to result.
  • You decide. The scope, the permitted sources, the approval policy and the final sign-off are yours.

Nothing is connected by default, so a fresh install can do nothing at all until you switch something on.

Step by step

Step by step

  1. Install Harness on the computer that will do the work.
  2. Sign in with the email of the person who runs the server; the machine registers itself.
  3. Add your Claude or GPT account or API key under providers.
  4. Create a routine and describe it in plain language.
  5. Grant only the capabilities that routine needs.
  6. Run it once and read the restated scope before approving anything.
  7. Approve the sensitive step when it appears in the queue.
  8. Open the audit trail and check what it actually did.
  9. Save the routine and give it a schedule.

Common problems

Common problems

  • Registration is refused. That workspace already has a machine, or the account signing in is not its digital member. See licence & activation.
  • The provider key is not accepted. Check it is a current key with credit or an active plan — see AI providers & keys.
  • The routine does nothing. Almost always a missing capability: the worker cannot see a folder or mailbox it was never granted. See capabilities & permissions.
  • It waits and nothing happens. Something is sitting in the approval queue — see approvals & audit.

Next

Build the first routine properly

Once it runs once by hand, the next step is making it repeatable.

Routines & scheduling All documentation