Frequently asked questions
Short answers to what buyers ask about how Bantuva builds software — and every answer is backed by something the product actually does, named in the page's source.
Who writes the code, and who reviews it?
AI agents write the code, in a sandboxed build job with recorded steps and deliverables. Then every change gets an AI review in a fresh session, and after it a human Tech Lead reviews the change and approves it or requests changes. No change merges on the agent's word alone.
Does a person approve a production deploy?
Yes. Production requires a passed validation against your org first, and then the explicit approval of a human administrator. The approval is recorded with that person's name in the audit log — agents can never pass a human gate.
What happens if a release goes wrong?
The release can be rolled back: the previous release tag is redeployed to production. The rollback asks you to acknowledge its limit first — it does not undo destructive changes or data changes — and a first release has no earlier tag to return to, so it is fixed forward with a hotfix.
Where does our code live?
In its own Git repository, one per project — either on Bantuva's Gitea, where the repository is provisioned for you with branch protection, or in your own repository on GitHub, GitLab, Bitbucket or Azure Repos, connected with credentials you provide.
Who can see our project?
Only the people invited to it, each with a project role. Access is filtered on the server by membership: someone who is not a member of your project is answered as if the project did not exist. Client users never see credentials, other projects or agent tokens.
What is recorded about what the agents did?
Every state change, gate decision, deploy, lock, token use and takeover goes into an append-only audit log with its actor and a Human/AI badge — entries cannot be edited or deleted. Each build job also keeps its own record: agent, model, times, deliverables, branch, commits, tests, coverage, outcome, log and transcript.
How is a change tested before we accept it?
The agent tests the change on your development org, and every changed class must carry a test class meeting your project's coverage threshold — 90% by default — measured on the change itself. The change then walks your delivery route, org by org, and on your UAT org you record Pass or Fail per test case: acceptance requires all passed, or a recorded reason for each failed one.
How does an engagement start?
With a demonstration. Before anything is signed, a demonstration project is created and a story from your own problem is built, tested and reviewed through the same path a paying client gets — on Bantuva's own demonstration org or a sandbox you supply, never a production org. You try the working software and say whether it is right.
The page's source names the document or file behind every answer above — because an answer about how we work should be checkable the same way our work is.