Delegation trust for AI agents · Coming soon

Your team's process, run by AI agents you supervise.

Write your process once, in plain English. FanIn turns it into a workflow agents can't skip or rewrite: every step checked, agents learn only what you approve, and a person always merges.

See how it works
How it works

Describe the process. FanIn compiles it into real code, not a prompt.

A fixed set of steps, checked before anything runs, then dry-run on your real tickets or repos so you see exactly what will happen.

Illustrative example · Ledgerly, a fictional 60-engineer payments company
FanIn · Ledgerly · Workflows · Stripe API upgrade
Describe your process
workflow.spec
source: jira("PAY · label = stripe-upgrade")
intake: repo_found → investigate · else → blocked
investigate: read-only · map Stripe calls
  touches card data → blocked
code: → test
test: unit + contract · pass → review
  fail → code
review: [review, security, pci] in parallel
  approved → publish · changes → code
  out of rounds → blocked
publish: → remember # draft PR, never merge
remember: → done # notes need approval
limits: max_rounds 5 · concurrent 3
IntakeIntake agent Investigateread-only CodeCoding agent TestTesting agent ReviewReview agent SecurityAppSec advisor CompliancePCI DSS advisor Publishdraft PR · owners Rememberpropose notes Donea person merges Blockedasks you parallel · advisors only tests failed changes requested · PCI / security no repo · touches card data · out of rounds
round 1 / 5
✓ every step can finish✓ every loop has a limit ✓ reviewers only advise22 PAY tickets matched · 14 repos · 0 merges
Watch it work

Many agents, each with one job, across every repo.

Cards move as agents work. A contract test fails? Back to coding. The PCI advisor finds card data in a log? Back again, up to five rounds, then it stops and asks you.

Illustrative example · Ledgerly, a fictional 60-engineer payments company
FanIn · Ledgerly · Stripe API upgrade · BoardClick any card to open it● Running

Intake

Coding

Testing

Review

Done

Blocked

Built on trust

Zero trust, applied to AI agents.

You don't make agents safe by watching them harder. You decide upfront what each one may do, and FanIn enforces it on every step.

⛓

The process is code

A fixed state machine. Agents work inside a step and can't skip, reorder or rewrite it.

✓

Checked before it runs

Every step can finish, every loop has a limit, and a dry run shows exactly what will happen.

◎

Agents with one job

Intake, coding, testing and review, plus your own advisor agents with your team's rules.

⌁

Stops when unsure

Blocked work says why. Give it what it needs, retry from where it stopped, or skip.

☰

Every ticket, explained

A full timeline of every step, every agent and every decision, in order.

⤓

A person always merges

FanIn opens draft pull requests. The final call stays with your team.

Approved memory

Agents learn, but only what you approve.

Proposed by Review agent · from PR #1176
"ledger-service rounds amounts half-even (banker's rounding). Keep that when mapping the new Stripe amount fields." Proposed · waiting for you
What the next coding agent reads
Implement the task with a minimal diff…
reference notes · approved data
ledger-service rounds amounts half-even (banker's rounding)…
Treated as data. Never as instructions.
Plugs into your tools

No new tracker. FanIn works where your team already works.

Jira
GitHub
Confluence
AWS
Kubernetes
Splunk
PagerDuty
Splunk Observability
Webex
Datadog
MCPYour internal tools
Coming soon

Awesomeness coming soon.

We're putting the finishing touches on FanIn. Want to be one of the first teams to try it?

Get early access

Be one of the first teams to delegate work to agents you can trust. We'll email you when FanIn opens.