Building an AI Memory Layer: A Problem I Didn’t Expect

While working on Contorium, I discovered something interesting.

The hardest technical problem wasn’t connecting models.

It wasn’t MCP.

It wasn’t tool calling.

It was context management.

A Common Workflow

A developer might:

1. Discuss architecture with ChatGPT
2. Debug code with Claude
3. Research with Gemini
4. Document findings in GitHub

A week later, they need that information again.

Now they have to remember:

• Which tool was used?
• Which conversation contained the answer?
• Whether the information still applies?

The knowledge exists.

Finding it becomes the problem.

What I Started Building

My goal with Contorium⁠ is simple:

Create a persistent memory layer for development workflows.

Instead of treating conversations as disposable, treat them as project assets.

An Unexpected Engineering Challenge

One challenge I encountered was balancing:

• Automatic context collection
• User control
• Searchability
• Performance

Too much automation creates noise.

Too little automation creates friction.

Finding the middle ground has become one of the most interesting parts of the project.

A Question for Developers

As AI tools become part of everyday development:

Do you think the future belongs to better models?

Or better memory systems that connect everything you’ve already learned?

I’m curious how other developers are solving this problem today.

https://www.contorium.dev/

https://github.com/ContoriumLabs/contorium