Agents

Agents that hold the brief.

An agent is a specialist you define once: a role, instructions it always follows, and the tools it's allowed to use. Brief it in plain language, and it works the job — asking approval before judgment calls.

From one prompt

“Set up a weekly scan: new competitors, pricing changes, summarized every Monday.”
  1. 1Define

    Give the agent a role and instructions it always follows.

  2. 2Brief

    Describe the job in plain language.

  3. 3Approve

    It pauses at judgment calls — you stay in charge.

  4. 4Reuse

    Reopen it next week; it still knows the brief.

What you get

  • Custom roles and persistent instructions
  • Approval cards for judgment calls
  • 4,000+ connected apps
  • Real artifacts saved to your workspace

01Define once, reuse forever

The role and instructions belong to the agent, not the conversation. Reopen it next week and it still knows exactly how you like the work done.

02Approval before action

Agents pause at judgment calls with approval cards. You stay in charge; the agent does the legwork.

03Tools, not just words

Connect your apps and agents work with real data — drafting from your inbox, filing to your drive, checking your calendar.

Frequently asked

How are agents different from Tros?

Tros are ten ready-made specialists. Agents are custom ones you define yourself — same engine, your job description.

Can agents use my apps?

Yes — connect your apps and agents work with real data, approval-gated.

Do I pay per agent?

No — agents are part of the toolkit on every plan. You pay for credits used, not seats for AI.

Related capabilities

Try it on the free plan

200 credits a month, core tools included, no card. Enough to build something real and decide if it fits your work.