Modern development has a strange problem.
Our projects become smarter.
But our projects also become easier to forget.
Every AI conversation contains valuable engineering knowledge:
- design trade-offs
- rejected ideas
- architectural decisions
- implementation reasoning
- future plans
Unfortunately, most of it disappears after the chat ends.
Documentation rarely stays updated.
Commit messages are too small.
ADRs are often forgotten.
So we started asking a different question.
What if the repository itself could remember?
Contorium approaches this by separating two responsibilities:
PIL (Project Intelligence Layer)
A deterministic, local-first storage layer that continuously captures structured project knowledge.
CIL (Cognitive Interaction Layer)
An intelligence layer that allows both humans and AI agents to understand that knowledge through natural language.
Instead of searching dozens of files, developers can simply ask:
- Why?
- How?
- What changed?
- What’s next?
- What’s missing?
The interesting part isn’t another AI assistant.
The interesting part is creating a shared cognitive layer that survives across:
- Cursor
- VS Code
- Claude Code
- Codex
- Gemini
- MCP clients
- CLI
One project.
One source of project intelligence.
Every AI tool starts from the same understanding instead of starting over.
We’re not trying to replace documentation.
We’re trying to make project understanding continuous.
Because software isn’t just source code.
Software is accumulated reasoning. Contorium is designed to preserve exactly that.
https://www.contorium.dev/
https://github.com/ContoriumLabs/contorium
Top comments (0)