Build an agent, hear it in the browser, put it on a number, then inspect its first real call. The same loop applies from then on — build, test, deploy, listen, improve.
Before starting
Two prerequisites: an account — polite.ai is in private beta, so request an invite. Nothing else: where a website crawl was consented to at signup, it has been running since the account was created, so the agent has something to talk about before anything is built.
The product tour walks these exact stops inside the dashboard, with a checklist that ticks off along the way.
Build and test
Open Knowledge. Any website crawl consented to at signup is listed with its status. Wait for the crawl to complete if you signed up very recently. Augment by adding PDFs or other docs for private information and processes not exposed on the public website — see Website knowledge.
Open Agents and press New agent. The Builder opens: describe in plain English what your objectives are. The builder will ask clarifying questions to help it scope out what you need and then it will start building the team. The agent structure is sketched out live as it is built, and you can ask questions and guide the structure. More in Describe what you want.
Test it in the browser — no number needed. Say hello, ask something a customer may ask, try to catch it out. On hang-up the transcript goes back to the Builder, so "it got the opening hours wrong" is a fix that can be asked for in the next message.
Put it on a number
Buy a number from Numbering → Buy number, then point it at the team — the Builder's Connect inbound calls panel does the wiring. Details in Connect a phone number.
Everything so far has been building up a draft version. Press Go live on the amber draft bar to publish — the number then routes to this version. See Drafts, versions & going live.
Ring it, then listen back
Call the number from an ordinary phone. The agent answers.
Open Calls and click the row the call just made: the recording, the transcript, and a feed of everything the agent actually did. If there are any issues, Troubleshoot with AI gathers the evidence and Fix with AI hands it straight to the Builder. See Finding calls, Inspecting a call and Troubleshoot a call with AI.
Where next
- Give the team more to draw on — Knowledge bases
- Let callers book real appointments — Calendar booking
- Add specialists and human transfer — Teams, transfers & subagents
- What an agent does on every call, without being asked — What every agent promises your callers