Fleet 1.17.0 is out.See what's new →
FleetFleet
Use case

AI Workflows for Customer Onboarding Documents

Every new enterprise customer gets an onboarding guide that is 80% boilerplate and 20% specifics — their integrations, their SSO setup, their environment names. Producing that 20% by hand means each kickoff eats a CS engineer's afternoon, and the boilerplate 80% drifts out of date separately in every copy.

Templates don't fix it: the copies still fork, the product still changes under them, and nobody owns reconciling either.

How it works with an agent fleet

A Fleet workflow fans out over per-customer config files and renders each guide from the single current source of truth — then a CS lead approves before each guide is shared.

genflows:
  - name: onboarding-guides
    steps:
      - {name: guide, prompt: "Write the onboarding guide for this customer file: their integrations, SSO provider, and environments, against current product docs.", corpus: ["docs/product/**/*.md", "customers/*.md"], for_each: "customers/*.md", kind: report, out: guide.md}
      - {name: accuracy, prompt: "Check the guide against current product docs. Flag stale steps or features that changed.", depends_on: [guide], kind: review, out: review.md}
      - {name: cs-approval, depends_on: [guide, accuracy], kind: approval, out: decision.md}
      - {name: share, depends_on: [cs-approval, guide], kind: publish, out: link.md}

When the product changes, re-run: incremental rebuild regenerates only what the change touches, and every customer's guide is consistent with the docs again — no per-copy reconciliation.

The fleet pattern

Fan-out over customer files → per-customer guide from shared product corpus → accuracy review → CS approval → published share link. One source of truth, N rendered guides, zero forks.

Guardrails that matter here

  • Each guide passes its own approval gate — a person confirms each customer-facing document before its link exists
  • The accuracy review compares against current product docs, catching the 'guide says the old settings path' failure mode
  • Published links serve the approved revision; an updated guide is a new approval, not a silent edit

Who this is for

CS and solutions teams onboarding customers with per-customer configuration, where guides are valuable enough to personalize and numerous enough that hand-maintenance fails.

Frequently asked questions

Where does per-customer data live?

As files in your repo — one markdown file per customer with their integrations, providers, and environment details. The for_each glob maps the draft step over them. Sensitive values stay out; the guide references names, not secrets.

What happens when the product changes?

Re-run the workflow. Step fingerprinting skips unchanged work, so only guides affected by the docs change regenerate — each through its own review and approval again.

Run your first agent fleet

One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.