Khive
Manage agent memory, operations, and structured relationships using an embedded typed knowledge graph and hybrid vector retrieval system.
Khive is an embedded system of records and operations built specifically for agent-native organizations and autonomous AI workflows. Operating entirely on local SQLite without external database dependencies, it merges typed entities and directional edges with vector search and full-text retrieval to let agents build traversable, structured world models. Agents interact through a unified Model Context Protocol (MCP) surface that exposes over 160 operational verbs—spanning salience-weighted long-term memory, GTD task lifecycle tracking, Bayesian user profiles, and sandboxed execution with Git integration.
主要功能
01Broad MCP Verb Surface: Dispatches 160 verbs across 14 packs covering task management, memory, messaging, and tool execution via a unified request interface.
02Dev and Runtime Tooling: Provides sandboxed execution over materialized trees, Git provenance tracking, and content-addressed blob storage.
03Zero-Infra Embedded Architecture: Operates directly within SQLite on disk with a warm daemon runtime to eliminate agent cold-start latencies.
04Typed Knowledge Graph: Models entities and 20 directional relations across 10 categories with support for graph traversals, shortest path, GQL, and SPARQL queries.
05Hybrid Retrieval Engine: Blends FTS5 full-text search with vector RRF fusion, embedding reranking, and sub-millisecond in-process Vamana ANN indexing.
0633 GitHub stars
使用案例
01Persistent Research Agents: Tracking academic concepts, paper lineages, dependencies, and temporal insights across multi-session research runs.
02Agentic Task Orchestration: Coordinating autonomous agent operations using structured GTD task lifecycles, scheduled triggers, and threaded agent communication.
03Codebase Ingestion and Dev-Loops: Inspecting source code dependencies, managing Git workflows, and running sandboxed commands with complete execution audit logs.