Hey everyone -- I'm Archit, the maker. I built Memori because I got tired of re-explaining context to Claude Code every session. "We debugged this yesterday." "You chose Tailwind over Bootstrap last week." "The API uses prefix-based ID resolution, remember?" It never remembers. So I built something that does. The core insight: memory for AI agents isn't a chatbot problem. It's a search problem. You need hybrid retrieval (full-text AND semantic), automatic deduplication (agents store aggressively), and decay scoring (recent + frequently-accessed memories should surface first). I wrote the engine in Rust for speed and the CLI in Python for ergonomics. The whole thing is one SQLite file you can copy, back up, or delete. No external services. The part I'm most proud of: the behavioral snippet. Running `memori setup` injects guidance directly into the agent's system prompt that teaches it *when* to store memories (after non-obvious bug fixes, after choosing between alternatives, when leaving unfinished work) and *when to search* (before investigating bugs, before designing anything). The agent becomes self-maintaining. I wrote a deep-dive on the design principles behind the CLI: https://archit15singh.github.io/... Would love feedback -- especially from other Claude Code / Cursor / AI agent users. What memory patterns do you wish your agent remembered?

Memori – Rust+SQLite persistent memory

Rust-powered persistent memory for AI coding agents

Rust-powered persistent memory for AI coding agents