Skip to content
Baguals

BagualsControl

Every agent tested, approved and on the record.

Agents tested before they go live, with human approval on sensitive actions and everything on record.

What Baguals Control does.

Tests · Customer agent, version 8
Status question with a wrong container numberPassed
Quote request without a shipping datePassed
Customer asks for a discountPassed
Must hand off a damage claim to a personFailed · blocks publishing

A future-ready operation is one you can audit.

You hand us the process; we do the rest. That means rules your managers can see, approvals in their hands and a record your auditors can read.

Explore the approach

Questions, answered.

How do you know an agent works before it reaches customers?

Every new version runs through simulated conversations with a test customer. A separate model acts as a judge and grades the answers, and code checks verify the hard rules, like calling the right tool or handing off a claim. If the new version does worse than the one in production, it can’t be published. Failures you find later become new tests.

Which actions need human approval?

By default, anything sensitive: moving money, changing customer data, committing a price or a date outside the rules. Your team defines the list with us during the pilot and can relax it once the agent has earned trust. While an approval is pending, the agent tells the customer someone is reviewing the request and doesn’t act on its own.

Can anyone change or delete the audit log?

No. The log only accepts new entries. Every message, tool call, approval and cost is added with the agent, the version and the person involved. Nobody can edit or delete entries, including the Baguals team. That way, if a customer or an auditor asks what happened in a conversation, you can show the full sequence exactly as it occurred.

What happens if we hit the spending cap?

The platform checks the budget before every model call. When the cap is reached, or if the budget can’t be checked for any reason, no model is called and no money is spent. The agent tells the customer it can’t continue right now, the conversation stays open, and your team is notified to decide whether to raise the cap.

How do we roll back a change that went wrong?

Each agent keeps its published versions, and only one is live at a time. Published versions are never edited, so going back means putting the previous version live again. Conversations already in progress finish on the version they started with, so a customer never sees the agent change behavior halfway through a case.

Let’s talk about your operation.

Get in touch