One thing I keep running into with Cursor is chat history degradation once work spans multiple repos or services.
Within a single repo it’s great, but once you’re bouncing between backend, infra, shared libs, and frontend, the conversation history either resets or slowly loses the thread. You end up re-explaining architecture, assumptions, naming conventions, and past decisions over and over.
In real production systems, that context is the work. How services relate, why decisions were made, what changed last week. Losing that breaks flow more than the actual coding.
I ended up building Pikai to experiment with keeping chat history persistent across repos and over time, so you don’t have to constantly rehydrate the same mental model. It integrates via MCP and is meant to complement Cursor when things get bigger than a single repo, not replace it
I’m testing it with a small group right now. If you’re working in multi-repo or microservice setups and this resonates, I’d genuinely love feedback, especially where it breaks or feels wrong