Building a Cognitive Overlay Instead of Another AI Agent

One design decision has been guiding Contorium from the beginning:

Don’t replace the developer.

Most AI tooling today is moving toward autonomous execution:

Observe
→ Plan
→ Execute

Contorium takes a different approach:

Observe
→ Understand
→ Suggest

What is changing?

Traditionally Contorium has focused on runtime awareness:

• File changes
• Workspace activity
• Context persistence
• MCP state tracking

We’re now experimenting with a new concept:

Cognitive Overlay

A lightweight layer that sits on top of runtime data and provides optional insights.

Examples:

• Detect likely development intent
• Suggest relevant skills or tools
• Surface useful resources from external ecosystems
• Recommend workflow presets

What it will NOT do

Contorium will not:

• Execute commands
• Install tools
• Modify projects
• Trigger AI actions automatically

The system remains read-only and suggestion-based.

Why?

Because we think there is a missing layer in modern AI tooling.

Today the ecosystem has:

1. Passive dashboards
2. Fully autonomous agents

But there is very little in between.

We’re exploring whether a runtime cognitive layer can bridge that gap.

A system that understands what’s happening without taking control away from the developer.

Still early.

Still experimenting.

But it’s becoming one of the most interesting directions we’ve explored so far.

https://www.contorium.dev/

https://github.com/ContoriumLabs/contorium