Shodh Memory

This platform provides persistent memory for AI agents without calling a language model to manage it. Where other memory systems use an LLM to store, summarize and organize memories, it relies on local learning rules instead: Hebbian strengthening, decay curves and a knowledge graph, all running as fast algorithms on the user's own hardware. Memories that are used together form stronger links, and rarely used knowledge fades naturally, so recall improves with use. The system combines a knowledge graph, temporal indices and hybrid ranking rather than plain vector search, and it includes detection of blind spots, shallow knowledge, orphaned clusters and stale areas in what an agent knows. Because no model is involved in the memory loop, the same input always produces the same memory, and each memory operation costs nothing in model fees. It ships as a single binary of about 30 MB written in Rust, with no Docker and no external dependencies, and it runs offline on anything from servers to a Raspberry Pi. The vendor states that data never leaves the user's machine. Packages are published on crates.io, npm and PyPI, and a research paper and a live demo are linked from the site. It is aimed at developers building agents who want an alternative to cloud based memory services, including for robots, IoT and air gapped systems. The project is licensed under Apache 2.0 and is free. The site publishes no paid plans.