Frequently Asked Questions

Straight answers to what people ask before and shortly after they build their first agent. If your question is not here, the community and support channels are one page away.

Getting started

How long does it take to get an agent answering calls?

Most workspaces have a working inbound agent in under an hour: connect a number, write a prompt, attach a calendar, test, publish. The time-consuming part is not the build, it is deciding exactly what the agent should refuse to do. Budget an afternoon for that conversation with your team.

Do I need engineers?

No, for Creator, Flows, and the standard integrations. Yes, if you want custom tools that hit your own API, webhook handlers, or CI-gated promotion. The dashboard and the API are the same product; you can start in one and grow into the other.

Can I bring my own phone number?

Yes. Import an existing Twilio number, port a number in, or connect a SIP trunk. Imported numbers keep their history and caller ID reputation, which matters more than most people expect for answer rates.

Behaviour and quality

Will callers know it is an AI?

Many will not on a short call, and you should tell them anyway. Katexs ships a disclosure line in the default greeting and several jurisdictions require one. Disclosure costs you almost nothing in outcomes and removes an entire class of complaints.

What happens when the agent does not know something?

It does whatever you told it to do, which should be one of three things: answer from the knowledge base, take a message, or transfer. Agents without an explicit unknown-path improvise, and improvisation is where hallucinations live.

How fast is it?

Typical end-to-end response latency is well under a second on the default voice stack. Tool calls add whatever your integration takes — a slow CRM is the usual cause of an agent that feels sluggish. Watch tool latency in Observability before blaming the model.

Can it handle accents, background noise, and interruptions?

Yes to all three, with the usual caveat that a bad line is a bad line. Barge-in is on by default so callers can interrupt naturally. If your callers are frequently in noisy environments, test with the mumbler and interrupter personas in Testing Lab before launch.

Cost and scale

How is usage billed?

Per minute of connected call time and per chat session, plus telephony pass-through. Silence on hold during IVR navigation counts as connected time. Data → Dashboard breaks spend down by agent and outcome so you can see cost per booked appointment rather than cost per minute.

What are the concurrency limits?

Limits are per workspace and listed on your plan in Settings → Billing. Bursty inbound traffic queues rather than failing. If you are running an outbound campaign against a spiky inbound line, cap campaign concurrency below your total so inbound always has headroom.

Operations

Can I change an agent while calls are in progress?

Yes. Publishing snapshots a version and in-flight calls finish on the version they started. Rollback is republishing an earlier version and takes effect in seconds.

Do you support multiple environments?

Yes — DEV, UAT, and PROD as linked workspaces with a promotion path, diffs, and a required-passing-suite gate. See Enterprise Environments.

What if Katexs goes down mid-call?

Configure a failover destination on every production number. If the platform cannot answer, the carrier forwards to your fallback — a human line, a voicemail box, or a second number. Numbers without a failover simply ring out, so set one.