Skip to main content
An agent runs on a schedule, checks your pipeline against a filter you define, and takes an action when deals match. Where Signals extract patterns from your context, agents watch the state of your pipeline and do something about it. You can build agents in the Deal Brain web app — click Agents in the left sidebar — or over MCP from a connected AI platform. The Agents page has two tabs: Agents for the agents themselves, and Run History for past runs.

​
What you’d use one for

  • Post to Slack every Monday with deals closing this quarter that haven’t moved a stage
  • Flag opportunities over a threshold amount with no activity in two weeks
  • Alert a channel when a deal in late stage has a close date in the past
  • Remind a team about deals missing a champion or economic buyer

​
How an agent works

Three parts:
1

A schedule

When the agent runs — a cron expression or a recurrence rule, in a timezone you choose.
2

A filter

Which deals it cares about. Built from opportunity, account, and contact fields with comparison operators, and combinable with and / or / not.
3

One or more actions

What happens when deals match. Today that’s posting a Slack message — either literal text or a rendered Canvas template.

​
Scope

​
Safety model

Agents are built to be hard to fire accidentally:
  • New agents are always created disabled. Nothing runs until you explicitly enable it.
  • Editing configuration disables the agent again. A half-edited agent can’t run.
  • Enabling validates dependencies — if the Slack channel it posts to isn’t connected, or the agent lacks execution authority, enabling fails with a specific blocker rather than silently doing nothing.

​
Where to go next

Build your first agent

A stalled-deal Slack alert, end to end.

Filter reference

Every field, operator, and action available.