AI Agents
Build AI teammates that draft replies, take actions through tools, reason over your CRM, and run on their own from triggers.
An agent is an AI teammate you configure once and reuse across your workspace. An agent has a personality, a set of tools it can use, a slice of your knowledge it can read, and rules for when it runs on its own. Once it's set up, you can @mention it in a thread, assign it to a ticket, message it directly, or let it run automatically.
What an agent can do
- Draft and send replies using your knowledge base, conversation history, and connected data.
- Take actions through tools — create a ticket, look up an order, update a record, or call any installed app or MCP server.
- Follow procedures — step-by-step playbooks for specific situations like refunds or cancellations.
- Run on its own from triggers — a mention, an assignment, a schedule, a record change, or an app event.
Two kinds of agent
| Kind | Who it serves | Where it runs |
|---|---|---|
| Internal agent | Your team | Chat, mentions, assignments, and autonomous runs |
| Chat agent | Your customers | The chat widget, with a safety-filtered set of tools |
The build loop
Building a reliable agent follows a simple loop. Each step has its own article:
Write the persona
Define the agent's character, tone, and rules in the persona prompt.
Give it tools and knowledge
Choose which tools it can use and which records it can read.
Add procedures
For repeatable situations, attach a procedure — a precise playbook the agent follows.
Prove it with simulations
Run simulations to test the agent's behavior before it goes live.
Publish
Publish a version to make your changes live.