AI CodeOps for agencies and consultancies

Agencies and consultancies run many codebases at once, for clients who each have their own Git host, their own rules and their own budget. AI coding agents can take on a lot of that work, but only if you can keep clients separate, show what each one cost, and prove the work was checked before it shipped.

The problem

An agency's engineers switch between client repositories all day. Each client has a different stack and a different idea of what "reviewed" means. Billing needs to know how much time, and now how much AI, went into each client. And at handover, a client often asks for proof: what was changed, who approved it, and were the tests run?

How AI CodeOps helps

One place for every client's code

AI CodeOps connects to GitHub, GitLab (including self-managed instances), Bitbucket, Gitea and Forgejo. Each project links one repository, and every AI session belongs to a project, so client work stays organised by repository. See projects and the integrations list.

AI work that arrives checked

You describe a task. Agents plan it, write the change, run the client's tests and review the diff for correctness and security. The change then waits at a release gate until tests pass, AI findings are reviewed and the approvals your policies require are in. Your senior engineers review finished, tested work instead of first drafts.

Cost you can bill

Every session shows its cost, so you can see AI spend per task and per client project. Spend caps can be set for the workspace and for each workflow, so a fixed-price engagement does not quietly overrun. See cost controls.

A handover record

Each change has a signed evidence pack: the task, the models used, approvals and each vote, policy decisions, findings, tests, the pull request and the deployment. It is signed with Ed25519, and your client can check it at /evidence/verify without an AI CodeOps account. That turns "trust us, it was tested" into a file.

Roles for mixed teams

Members get roles: owner, admin, developer or viewer. A billing admin can handle billing without reading code. The audit log records who changed what.

Model choice per engagement

Run on the platform's Economy, Standard or Premium tiers, or Auto. If a client wants work done on their own model provider account, add their key to the workspace you use for them.

What to try first

1. Follow the quickstart and connect the Git host of one client you know well.
2. Create a project for one repository and run a small, well-understood task with the "Plan, Build, Review & Test" workflow.
3. Check the session's cost against what you would have billed for the same work.
4. Download the change's evidence pack and walk through it as if you were the client.
5. Set a workflow spend cap before you hand the agents larger tasks.

Good to know

AI CodeOps is browser-only: there is nothing for your engineers or your clients to install. It is free during the open beta. How many workspaces and seats you can have depends on your plan; see pricing. For SSO or custom roles, talk to us about Enterprise.