
Catalyst AI Fleet
A public-facing account of Catalyst Labs' internal agent-company experiment: named roles, review chains, evidence gates, and human authority stated plainly.
Overview
Catalyst AI Fleet is the public explanation of a private operating experiment: what happens when a group of agents receives named roles, reporting lines, budgets, review chains, and a durable company ledger instead of being treated as one interchangeable chat window.
The central claim is intentionally modest. Catalyst Group remains human-led. The agent company has no day-to-day business authority, and the site does not market the experiment as a product.
What the site explains
The project documents four things that are easy to blur together:
- the squad’s hierarchy and role boundaries;
- how goals move through orchestration, implementation, review, and acceptance;
- which outputs have actually shipped;
- lessons and known failure modes from operating the fleet.
The public copy makes the human gates visible because an org chart full of agents can otherwise imply authority the system does not have. Drafting, implementation, and review may be delegated. Business direction and high-impact acceptance are not.
What exists now
The repository contains a handcrafted multi-page static site with sections for the squad, architecture, outputs, lessons, blog, and FAQ. A deterministic Python builder produces the public tree, and CI rejects remnants of the discarded documentation-theme prototype before GitLab Pages deployment.
That is evidence of a built publication surface, not evidence that every organizational claim is permanent. The fleet changes; the site therefore has to be maintained as an operating document rather than launched once and forgotten.
Status
Active internal experiment with a public-facing site build. The site source and Pages pipeline exist on main. This page does not claim a separate public custom domain until one is independently verified.
Related
- Source: ai/catalyst-fleet-site
- Catalyst Group: catalystgroup.tech