khive is a knowledge-graph runtime for AI agents, written in Rust and shipped as a single binary that speaks MCP (Model Context Protocol). It is open source under Apache-2.0, at github.com/ohdearquant/khive.
The runtime gives an agent a place to keep what it learns. It stores typed entities and connects them through a closed edge ontology, rather than a free-form graph where any relation can point anywhere. Retrieval runs over hybrid search, combining lexical and vector matching so an agent can find what it needs whether it remembers the exact words or just the idea.
All of it runs from one process. There is no separate database to stand up and no Python runtime in the loop. A client that speaks MCP, such as Claude Code, Codex, Cursor, or Windsurf, connects to the binary directly.
The codebase is a Rust workspace of 34 crates. The shipped binary, kkernel, installs from crates.io. It sits alongside the rest of the open-source stack we build on, listed on the products page.