VelesDB
A local-first memory engine for AI agents that fuses vector, graph, and columnar storage under one query language, and can explain the evidence behind each recall.
VelesDB is a memory engine for AI agents that combines vector search, graph relationships, and columnar storage in a single binary, queried through its own language, VelesQL. Its why function returns the evidence path behind a given recall, and a deterministic context compiler is intended to reduce the token cost of retrieved context. Nothing in the memory path leaves the machine: no LLM call and no API key is required to store or retrieve memory.
It installs with pip for basic vector search, or as a separate local MCP memory server for use with Claude Code or another MCP client, and ships integration hooks that resume and save working context around a coding agent's session lifecycle and require a successful memory recall before certain tool use.
It is written in Rust.