Skip to main content
An agent is an AI teammate that handles one kind of work: new internet leads, calls to a line, follow-up after the sale, and so on. Every agent works from the same contact record and calendar, so a lead who calls, then texts, is one conversation. Agents handle the logistics. Your salespeople still sell: when a lead is ready, the agent books the visit or hands the conversation to a person.

What an agent can do

Open Agents and choose New AI agent. Start from the work you need done. None of these fit? Choose Or describe it and write the job in a sentence. RevDesk drafts the agent for you.

Set one up

1

Start

Pick a card or describe the job. Under More options you can name the agent, choose Where they work (the whole workspace or one team), limit Knowledge access, and add a playbook.
2

Review

Read what the agent is told (Instructions), what it must never do (Never), What it sends, and Where it runs. Change anything that doesn’t sound like your store.
3

Simulations

Run a Quick simulation by Call or Text. Nothing is sent or booked. For deeper testing, see Simulations.
4

Go live

Read What happens when you turn it on, add a phone line if the agent should answer one, and choose Turn it on. Nothing starts before this step.
A phone line is optional. An agent with no line still works its playbooks, such as follow-up texts and outbound calls. Choose Get a number to add one. See Phone numbers.

Change how it sounds and what it says

Open an agent to see its tabs: Results, Voice & Personality, Instructions, Knowledge & Tools, Call Flow, Replies & Handoffs, and Playbooks. Business hours live on the line (Bookings › Hours), and quiet hours live in Shared defaults, so every agent follows the same rules. For outbound calls, the opening comes from the playbook step, not the inbound greeting. For reusable instructions across agents, see Custom prompts.
If someone asks whether they’re talking to a person, the agent says it’s an AI. When your workspace requires it, AI disclosure and consent is on for every agent.

Give it playbooks

A playbook is a piece of work the agent runs: follow up with new leads, remind owners about recalls, win back quiet customers. Open the Playbooks tab and choose Add to agent. One agent can run several. To see each person a playbook is working, open the run and pick a person. See Tasks for how a run decides the next step.

Keep it working

  • Status. Each agent shows what it’s doing now: On a call now., Ready to work., Finish setup., or No number rings this agent yet.
  • Pause. From the ⋯ menu. A paused agent doesn’t take or place calls, and its running work pauses too. Resume it from the same menu.
  • Duplicate. Copy an agent to run the same job for another team.
  • Remove. Choose Remove agent. Move its active work and phone lines to another agent first.

What agents will and won’t do

Will

Answer from your knowledge and inventory. Book on the right salesperson’s calendar. Qualify with your questions. Transfer to a person when the customer asks or the conversation calls for it. Follow up by call, text, or email.

Won't

Guess when your knowledge doesn’t have the answer. Promise terms outside your policies. Claim to be human. Contact someone who opted out or during quiet hours.
When a question falls outside what it knows, the agent takes the details and offers a callback from a person. Read those conversations and add the missing answers to Knowledge.

Check the first leads

After you turn an agent on, open its first few conversations in Messages:
  • Did it answer quickly and ask the right questions?
  • Did the visit land on the right salesperson’s calendar, or the handoff reach the right person?
  • Can a manager see why the lead was booked, handed off, or closed?
If a lead repeats information, check that your lead source maps the fields. If the wrong salesperson gets booked, check routing and calendar settings.
For developers. The /agents resource in the API reference returns runtime assistants, the voices attached to phone numbers and campaigns. They are not the same objects as the agents on this page, so counts won’t match. Scopes are agents:read and agents:write.