shodh-memory

AI agents forget everything between sessions. Vector DBs give you similarity search, but not actual learning. Shodh is different – it's a cognitive memory system: - Hebbian learning – memories recalled together strengthen their connections - Activation decay – unused memories fade (like biological memory) - Knowledge graphs – entities extracted and connected automatically - Three-tier storage – working → session → long-term (based on Cowan's working memory model) Technical details: - Single ~15MB Rust binary, runs 100% offline - Sub-millisecond graph operations (entity lookup: 763ns, 3-hop traversal: 30µs) - MiniLM-L6 embeddings + TinyBERT NER, bundled - MCP server (works with Claude, Cursor), Python SDK, REST API - RocksDB for persistence Edge deployment: Designed for local-first and air-gapped environments. Runs on Jet

Hey HN, I built this because most AI memory solutions (mem0, Zep) need cloud connectivity. That doesn't work for drones flying beyond cell coverage or robots in warehouses. Shodh-Memory is a local-first AI memory system: - 100% offline, runs on edge devices - 4MB binary, sub-100ms retrieval - Multi-tier memory (working → session → long-term) - Geo-spatial queries for GPS-tagged memories - Mission tracking for autonomous systems Install: pip install shodh-memory Quick example: ```python from shodh_memory import MemorySystem memory = MemorySystem("./robot_memory") memory.record( content="Obstacle detected at X=5.2, Y=10.1", experience_type="observation" ) results = memory.retrieve("obstacles near position 5") Rust core with Python bindings via PyO3. Website: https://shodh-rag.com/memory PyPI: https://p