Introduction
Vertical AI's voice and text agents answer real phone calls. Each agent pairs one language model with a flat list of tools, and there is no call flow to draw.
We handle the voice engineering, so you can focus on what your agent says and does. Start a workspace to build your first one.
What a voice agent can do
- Hold a real conversation, including being interrupted, corrected and talked over
- Answer a phone number you provision on the platform, or one you already own
- Read and write to your systems while the caller is still on the line
- Book meetings, send texts, search a knowledge base, hand the call to a person
- Take calls and messages from a widget on your website
What happens on a call
Three things run in a loop for as long as the caller is talking.
- Speech recognition turns the caller's audio into words while they speak, rather than waiting for a pause.
- A language model reads your prompt, the conversation so far and the tools available right now, then decides whether to say or do something.
- Speech synthesis reads the reply back down the line, and stops the instant the caller speaks again.
The parts of the loop you control are the prompt, the tools, the voice and the model. The parts you do not have to think about are turn-taking, interruption, echo, retries and failover.
By default every call stays in Australia: call audio, speech recognition, the voice, the AI model, recordings and transcripts. Where each part of a call runs is in Data residency. If you are still choosing a platform, voice agents hosted in Australia covers the same ground in one page.
The shape of an agent
An agent is one language model, one system prompt and a flat list of tools.
There is no call-flow diagram to draw. Why that shape holds up on real calls is argued in Core concepts.
Ordering still matters. A tool can name prerequisites, and until those prerequisites succeed the tool is hidden from the model entirely. You get sequencing without a flow.
Where to start
If this is your first agent, read the getting started guide. It walks the whole path from creating an account to taking a live call on a real number. No coding required.
If you want to know what runs when a call arrives and in what order, read Architecture.
What lives here
- Platform: what Vertical AI is, the getting started journey, core concepts, and the models behind a call.
- Build: authoring an agent, its prompt, tools, knowledge and voice.
- Operate: going live on a number, measuring quality, and running the workspace day to day.
- API: every route of the REST API, generated from its OpenAPI specification. To drive Vertical AI from your own code, start with the REST API guide, the TypeScript SDK, the Python SDK or the vai CLI.
- Trust: security, access control, privacy, data residency, compliance and reliability. What the platform protects, what it does not offer, and where call data goes.
How calls are billed
Calls are counted per minute of audio, from connect to hang-up. Your offer sets the charge: Vertical AI plans include monthly minutes, texts and phone numbers, while the usage-based offer charges for calls and number rental. Existing or negotiated workspaces can have their own terms.
The full breakdown lives in your workspace under Plan and usage, and in Billing.