If you use AI coding assistants across multiple platforms (like jumping between Claude and ChatGPT), you’ve probably hit the "Empty Room" problem.
Every time you start a fresh chat session, the model has zero context about your database schema, your environment setup, or the architectural decisions you locked down yesterday. I got completely burnt out spending the first 10 minutes of every coding session pasting in massive context blocks.
So, I decided to build a lightweight experiment to solve it: NICO (Neural Information Cache Object).
The Architecture
The goal was to keep latency near zero without managing heavy persistent infrastructure.
The Logic: An asynchronous background extraction layer that parses finalized chat decisions into structured JSON objects.
The Stack: Built using Next.js and deployed on serverless edge infrastructure with a fast key-value backing store to instantly retrieve and inject state back into active sessions.
Why I'm Sharing This
Right now, I'm running a simple validation test to see if other engineers actually struggle with this cross-model memory gap before building out the full browser extension client.
If you deal with this same LLM amnesia, you can check out the rough MVP landing page and early developer list here: nico.aimrrs.in
How do you currently handle context management when switching between different AI models? Would love to hear your thoughts or critique on the edge-caching approach in the comments!
Top comments (2)
Caching the context is the cheap half. What costs is what gets loaded into each call: if the whole project goes back in on every request, those tokens are billed again each time. I would check how often it loads before anything else.
The website NICO