← Work

Repository intelligence

LORE

An enterprise platform for repo search, agentic SDLC, and a persona that is allowed to comment — never to act.

Without this, every router change is a vibes deploy.

LORE is a repository-intelligence platform I built: index the archive, search it, chat against it, and run agentic work through a gated Factory. The public description stays at the architecture. The internals stay inside the shop that paid for them.

Agents do not get the laptop. They get a sandbox — a locked-down Linux image with Docker-in-Docker, browsers, and the language toolchains they need — orchestrated as cloud work orders with phases, human gates, and cost tracking. TypeScript automations are compiled into the .NET host and executed in those sandboxes.

Lortimus is the observer. Identity manifests, visual canon, drift tests that fail CI if a string goes stale. Flavour in the HUD. No authority. The line in the identity file is the product: a passive observer does not dispatch, approve, merge, or change anything.

What holds

  • Agentic sandboxes with Docker-in-Docker, not agents on the host
  • MCP server so external agents share the same repo tools
  • Embedded TypeScript automations inside a .NET host
  • Architecture-as-tests: dependency rules fail CI, not code review
  • Persona system with identity manifests and drift guards

Stack

.NET · React · Postgres · AWS ECS · MCP