Skip to main content

What Are Agent Chains?

Chains link several specialist agents into one workflow — each step hands its output to the next, so a multi-agent job becomes a single button press.

Written by Schae Lilley

The idea

Plenty of real work is a relay, not a single task: plan the campaign, then build the ads, then review the result. A chain captures that relay once. You line up the agents in order, give each step its own instructions, and every run passes each step's output to the step after it — the builder works from the plan, the reviewer works from the build.

You'll find them in the Agents product: choose Chains in the Automation group of the left rail, next to Schedules.

Building a chain

  1. Click New chain — or start from one of the templates on the Chains page, which pre-fill the builder with a proven sequence (like "Meta Ads — plan → build → optimize").

  2. Name the chain.

  3. For each step, choose an agent and write its instructions. Step one gets the brief; later steps describe what to do with the previous step's output.

  4. Use Add step to extend the pipeline (up to 8 steps), and the arrows to reorder or remove steps.

  5. Add clients — each run fires the whole chain once per client. The client pinned in your top-bar picker is added automatically. If a client is missing data an agent needs, the step shows a warning so you know before you run.

  6. Save chain when it's ready, or Save as draft while you're still shaping it.

From the builder you can also flip on Run on a schedule — the chain then fires automatically on the cadence you set, just like a regular schedule. Drafts never fire.

Tip: Keep each step's instructions focused on one job. A chain shines when every agent does its specialty and passes clean output forward — not when one step tries to do everything.

Chains are team assets

Every chain in the workspace is visible to the whole team. The Chains page splits them into Yours and Team:

  • Anyone can open and run a teammate's chain — the run belongs to the person who fired it, landing in their chats and notifications.

  • Only the owner can edit, delete, or revise a chain. Opening someone else's shows the full setup read-only, with Run as the only action.

That makes chains a natural way to package best practice: one person perfects the sequence, and the whole team runs it.

Note: Deleting a chain deletes its run history with it — you'll be asked to confirm.

What happens when you run one

Each fire produces a reviewable run: a bundled card on the Workflow board, per-step outputs and files in the run rail, and a notification when it settles. See "Running and reviewing a chain" for the full walkthrough.

Did this answer your question?