memora

Why we included this project

Anyone who runs MCP-capable agents like Claude Code or Codex has probably hit the same wall: every session starts with a blank slate, so context, decisions, and resolved issues get re-derived over and over. Memora gives those agents a persistent memory layer they call through the Model Context Protocol. When an agent finishes work, you absorb the results into a structured store, and an LLM classifies each fact as a duplicate, update, contradiction, related item, or new entry, so old knowledge is superseded rather than deleted and retrieval follows the lineage chain to the current version. Semantic search, a live graph UI, and a chat panel make it easy to see what the agent actually remembers, and the SQLite backend with optional cloud sync (S3, R2, D1) keeps everything self-contained. If you've ever watched an agent re-derive something it already knew, that's the problem this solves.