Voice Agents Are Native in Timbal

Most voice agents are a second product. A different tool, a different script, integrations wired up again by hand, and a transcript that lives somewhere else. The agent your team spent months getting right for chat is not the one that picks up the phone. A cousin is.

In Timbal, voice is not a product. It is a way of talking to the agent you already have. Turn it on and the same agent can hear and speak: same knowledge, same tools, same rules, same memory, same history. That is the whole announcement. The rest of this post is what that buys you on a live call.

One agent, now with a voice

A voice agent that is simply your agent means nothing gets rebuilt. What it knows on chat, it knows on a call. The systems it can reach, the actions it can take, the boundaries it respects, all of it carries over untouched. When you improve the agent, the phone line improves with it. There is no second version to keep in sync and no second place to look when something goes wrong.

Every call shows up where your other conversations do, with the same level of detail. You can see how quickly the agent answered, what it heard, what it said, and why.

What makes a call feel like a call

Speech recognition and a good voice are table stakes. The difference between a demo and something a customer will stay on the line with is a handful of small human behaviours, and they come built in.

• It knows when you are done. Not a fixed pause, but a real sense of when a sentence has ended, so it neither talks over you nor leaves you hanging.

• You can interrupt it. Talk over the agent and it stops, listens, and remembers only what you actually heard.

• It fills silence naturally. If it needs a moment to look something up, it says so, in your language, instead of going quiet.

• It can speak first. A greeting, an introduction, a gentle nudge if the line goes quiet.

• It can be recorded. Every call can be kept, exactly as it was heard.

In the browser, on the phone, or inside your product

The same agent answers wherever you put it. On a web page, a customer clicks and talks. On a phone number, they dial. Inside your own app, it plugs in with a few lines of code. Your credentials stay on your side; nothing sensitive ever reaches the caller’s device.

The same rules apply

A voice agent is still an agent. If a certain action needs sign-off, it still does when spoken. If a policy limits what the agent can say or do, it limits it on the phone too. You define the rules once, and every channel follows them.

The concierge on timbal.ai/contact is this in production. Choose your language and talk to it. It asks what you are trying to solve, listens, and when it has enough, prepares the contact form for you to review. You read it and press send. The agent never submits anything on your behalf.

Try it

Go to the contact page, switch to Voice, and talk. That is a regular Timbal agent built with the open-source framework, not a separate voice product. Turn voice on for an agent you already trust, and give it a phone number.

Voice was never a separate product. It is the same agent, heard.