What is khive?

khive is a knowledge graph runtime designed for AI research agents that need to track structured information, such as relationships, connections, and history, rather than just searching for similar text. Where vector search finds documents that look alike, a knowledge graph tracks how things relate: which concept depends on which, which papers contradict each other, and which questions remain unanswered. It organizes everything into three types of records: entities, which are the things being tracked such as concepts, documents, datasets, projects, people, and organizations, notes, which are observations, insights, questions, decisions, or references about those entities, and events, an unchangeable audit log of what happened and when. Entities are connected by typed edges that capture specific kinds of relationships such as dependencies, derivations, or implementations, giving the graph meaning beyond simple links. Agents interact with khive through MCP, a protocol for communicating with AI tools over standard input and output, so there is nothing extra to deploy. The 11 built in tools follow simple verb patterns: create, get, list, update, delete, merge, link, traverse, neighbors, query, and search. Search combines full text matching with vector embeddings and blends the results, so both keyword and meaning based queries work. The whole system runs on a single SQLite database file on disk, with no separate graph database or query server required. khive is built in Rust and can be installed with a single cargo command, or built from source with cargo build. It is released under the Apache 2.0 license and is aimed at developers building research oriented AI agents who need persistent, structured memory that grows across sessions instead of starting fresh every time.