Building a Shared Intelligence Layer for AI Coding Agents

Most AI development workflows have a hidden inefficiency:

Project understanding disappears between sessions.

The result is familiar:

• Repeated explanations


• Lost decisions


• Fragmented reasoning


• Context switching overhead

What We Changed in Contorium v3

We unified our architecture around a single concept:

Project Intelligence Layer (PIL)

Instead of focusing on chats, prompts, or agent memory, PIL focuses on preserving project intelligence itself.

The system records:

STATE

INTENT

DECISION

WHY

and tracks:

TIMELINE

IMPACT

CONFIDENCE

EVOLUTION

PROVENANCE

Runtime Contract

All runtimes now expose the same capabilities:

Capture

capture_focus

capture_note

capture_decision

Inspect

inspect_state

inspect_intent

inspect_decision

inspect_why

...

Transfer

transfer_context

transfer_intelligence

transfer_handoff

This creates a consistent interface across:

• IDE


• MCP


• CLI

while sharing a single local intelligence repository.

Why This Matters

As AI tooling becomes more capable, continuity becomes more important than raw intelligence.

The challenge is no longer generating code.

The challenge is preserving understanding.

Contorium v3 is our attempt to make project intelligence a first-class development artifact.

⸻

GitHub: https://github.com/ContoriumLabs/contorium

Website: https://www.contorium.dev