Start here

Getting started

You sign up and take a real call in about thirty minutes. The sign-up wizard builds your first agent and lets you talk to it in your browser. Test calls are free.

Sign up, build an agent, and take a real call. No coding required. Set aside about thirty minutes. If you have not decided yet, what Vertical AI does is the shorter read.

Before you start

  • Have a microphone ready. The wizard lets you talk to your agent in your browser before you sign up.
  • If you want to personalise the agent with information about your business, have your website URL ready.

1. Sign up

Create an account and you land in a workspace. Everything you build lives here.

2. Build your agent

Sign-up runs a wizard that builds your first agent. It reads your website, attaches tools, and lets you talk to the agent in your browser. Then you create your account. At the end you can pick a plan and a number and add a card to go live, or come back to those later.

If the wizard can't read your website, try another address or choose Start without a website. You get a ready-made agent, and the details in square brackets, such as [Business], are the first thing to replace.

Open the agent afterwards. Three sections matter most for a first build:

  • Prompt: your agent's persona and rules. Every agent ships with a starter prompt. Swap the [square brackets] for your business and tighten from there.
  • Voice: how the agent sounds. Pick one and listen before you publish.
  • Model: we recommend the default, a Fast-tier model that runs in Australia.

3. Add a tool

A tool is how the agent does real work instead of just talking: look up an order, search a knowledge base, transfer to a human, or end the call.

Open Tools and choose Add tool. Five ways to create a tool:

  • Built in: prebuilt actions like Transfer to a human or End the call. See Builtin tools.
  • Plugin: runs on our infrastructure, like an address check.
  • App: connect an app via Connected apps.
  • Your system: a custom API call to your own HTTP endpoint. See Tools for parameters, variables, secrets and mocks.
  • From a website: point at a URL and it wires up a matching lookup tool.

Tool descriptions are the lever

If the agent calls a tool at the wrong time, or never calls it, fix the description before you touch the prompt.

4. Test it

Press Test to open the test panel and have the conversation you expect a real caller to have. Then try to break it: mumble, change your mind, ask something out of scope.

Iterate until it behaves every time, not most of the time. Test calls are free.

For automated grading, see Evals.

5. Publish

Your edits live on a draft. Publishing promotes it to the live version callers reach. The validator checks for problems first and blocks anything that would break a call.

6. Get a number

Phone numbers belong to the workspace. Buy one from Phone numbers in the workspace sidebar, then assign it to your agent.

7. Take a call

Call the number. After the call, the transcript shows what was said and which tools ran.

That is the loop: read the calls that went wrong, tighten the prompt or a tool description, publish, repeat.

Working from an AI assistant or from code

You can do everything above without the app's screens too:

  • Ask your AI assistant. Connect the AI assistant you already use and describe what you want in plain language. The steps for each one are in Connect your assistant.
  • For developers. The vai CLI keeps an agent's config in a repository and runs its tests in CI; the REST API and the TypeScript and Python SDKs reach every operation from your own code.

Where to next

  • Core concepts: the six ideas the platform is built on.
  • Hearing: how the agent listens and takes turns.
  • Knowledge base: give the agent facts to draw on.
  • Billing: how calls are metered and what numbers cost.

On this page