Browse docsMenu

Concept

Agent

An AI worker with a role, instructions, and scoped access to your tables, files, and apps.

What a agent is

An agent is hired like a teammate: you write its instructions (the job description), pick its runtime, and grant it access to exactly the data and apps it needs. Agents can be asked things in chat, put on a schedule, given steps in a flow, or wired to a surface so they handle inbound work on their own.

Reach for it when

  • The work needs judgment: drafting, classifying, extracting, summarizing, deciding.
  • Inbound items vary too much for fixed rules.
  • A person currently does the step by reading and thinking, not by following a checklist.
  • You want chat-style help that knows this pod’s data.

Hire your first agent

  1. 1
    Open Agents in the sidebar and create one with a single, narrow job.
  2. 2
    Write instructions like a job description: what comes in, what should go out, what to do when unsure.
  3. 3
    Grant it an access scope — only the tables, folders, and apps the job needs.
  4. 4
    Run it once from chat with a real example before wiring it anywhere.

Where it lives

Agents in the left sidebar. Each agent has Runs, Edit, and History tabs.