If you use Claude Code, Codex, or any MCP-compatible AI coding tool, you have probably experienced the daily grind of re-explaining your project context from scratch. I often waste a lot of time re-pasting instructions or context and on occasion, I forget which repository or project I’m working on and built something to help me save time and token costs 
My build runs entirely on my machine (local-first architecture) and I’ve set it to watch my code, decisions and failures. I have about 10-14 repositories so I built other capabilities like:
- Language-aware code indexing that understands my code at function and class level
- Cross-repo contract conflict detector that warns me when a change in one repository may break another
- Poisoning detection that quarantines anomalous or adversarial memories before they pollute search
This is a tool (not AI), that other AI-coders use. So Codex will know what Claude did, Claude will know what Kimi K2.6 did, and Kimi K2.6 knows what Codex did. The project is still a work in progress and I’m really happy about how its coming along.