Code Colony
Code Colony renders a repository as a multiplayer AI-company simulation: regions, rooms, agent employees, jobs, risk, trust, and cost on a world map, with human approval gates at the spec and artifact boundaries.
- My role
- Solo developer
- Status
- Active
The problem
Agent-run software work is invisible: you cannot see which agent is doing what, at what cost, with what risk, or where a human approval is required.
What runs underneath
A Convex backend with twenty mounted components, a container-backed sandbox Durable Object whose exec and clone endpoints are HMAC-authenticated, and an infrastructure-as-code stack that owns production DNS and deploys the Worker and Pages projects.
Still mocked, and it says so
The README publishes what is still mocked: multiplayer presence and votes, runtime execution, artifact diffs, tests, the cost ledger, and marketplace installs. It labels memory and repository scanning as partially real.
Who did what
What I owned
Authorship
Lalo is the only human author in the repository's history; the other commit identities are his own agent tooling.
What the team owned
Collaborators
No other human contributor appears in the repository, its local history, or the project tracker.
Built with
TypeScript, Next.js App Router, TanStack Start, Convex, Mastra, OpenRouter, Cloudflare Workers, Pages, and Durable Objects, Alchemy IaC, PostHog
Receipts
| Source | Kind | Verified |
|---|---|---|
| Public repository | public repository | 2026-08-20 |
| Self-declared mocked-surface table | source excerpt | 2026-08-20 |
| Live deployment | live product | 2026-08-20 |
| Repository contributor history | qualitative statement | 2026-08-20 |
Withheld: Code Colony has no published adoption, usage, or benchmark figures.
Can’t trust your agent yet?