If you are looking for persistent memory across sessions, check out claude-brain too. Full lossless capture with search, not just checkpoints. https://github.com/mikeadolan/claude-brain
Interesting approach. I went a different direction and built the memory layer as a SQLite database instead of plain text files. Every conversation captured losslessly via hooks, with keyword,semantic, and fuzzy search across all projects. The tradeoff is you need the MCP server or hooks to surface context, but you get full search and zero orphaned memories when projects move or rename.Running on 1,300+ sessions.
Same here. That 3am energy is real. I have not felt this way about technology in awhile. Been working on something to solve the context loss problem. Should be sharing it soon. How do you handle it when all three models disagree on the same problem?