← All work

10 / Agent-run studio operations · solo · stopped

Foreman

An operations console for a one-person studio working with a team of AI agents. A customer writes in, an agent sorts and answers it, anything risky stops and goes to Slack for a person to approve, and every step lands in a ledger. It stopped at a demo running on staged data and never ran for real.

No public demo
The Ledger tab: a timeline of actions by each agent and by Tom, with cost, latency and outcome columns and a chain verified mark
foreman — ledger, staged data
Role
Solo — with a team of Claude Code agents
Status
Stopped · staged demo data
Stack
Next.js 16 · AI SDK · Claude Haiku · Postgres (Supabase) · Slack

The rule

Unknown risk is treated as maximum risk, never as safe.

Build notes

Rules decide when to stop

A risk score over the limit, a high-risk tag, an effect outside the studio or an unsure model all halt the action until someone approves it in Slack. That decision sits in code, not in the model.

A hash-chained ledger

Each step is chained to the one before it and written in order under an advisory lock, so concurrent writes cannot fork it. Change one row and the chain breaks on screen.

Why it stopped

To keep the demo smooth, the data was seeded, the agent runs were replayed and the ledger was backfilled. A demo like that proves very little, so the project ended there.

More screens

The Ledger after a row is tampered with: the header reads chain broken at sequence 13 and one row turns red

Want to work with NYX?

Build a product together, distribute one that exists, commission software, or join in — write a few lines and NYX will get back to you.

Send an invitation →