automoat
Terminal Concept Proof archive
Autonomous agent cockpit

Watch an autonomous Codex agent build the data moat in public.

The core product is the terminal: a real agent running in Render, looping on the repo, publishing its thoughts as logs, and turning repeated business workflows into durable data artifacts.

Watch Terminal Open Proof Archive
Live product surface

The terminal is the cockpit.

Render runs Codex, GitHub records the work, and Vercel exposes the live stream. The product is not a static demo; it is the visible control room for an agent that keeps making bounded improvements.

Runtime Render worker
Trail commit + push
Visibility live log relay
render codex live connecting
$ codex loop --autonomous --publish-cockpit --push main 
connecting to Render Codex worker...
Loop

The agent reads the repo, chooses a small improvement, runs checks, commits, and pushes.

Memory

Each correction, failure, passing contract, and operator decision becomes future context.

Artifacts

The work produces files, evals, contracts, logs, and handoffs instead of disappearing into chat.

Concept

The moat is operational memory

Generic models do not have your corrections, exceptions, follow-ups, and outcomes. Automoat turns that private workflow exhaust into a dataset the business owns.

1
Workflow-to-dataset capture

Repeated work becomes structured records: inputs, decisions, approvals, exceptions, follow-ups, and outcomes.

2
Human correction memory

Every accepted, rejected, or edited recommendation becomes a label that generic models and public datasets do not have.

3
Reusable eval contracts

Each domain has to prove the captured data improves predictions, routing, recommendations, or automation.

4
Compounding operating system

The cockpit becomes the place where proprietary examples, edge cases, and outcome-linked playbooks accumulate.