HiveLLM® is a free, open-source ecosystem for autonomous AI collaboration—
with persistent memory, cognitive context, high-performance data infrastructure, and agent-native tooling.
Get Started ✨ · Report a bug 🐞 · Contribute 🫶 · View Ecosystem 🏗️
Reduce AI costs by 50-90% while keeping 100% of data on-premise. HiveLLM is a suite of independent, production-grade components—mostly written in Rust—that give AI agents what they lack out of the box: durable memory across sessions, semantic and full-text retrieval, graph and realtime data, a single binary wire protocol across six languages, and rules that survive a /clear.
Memory & Retrieval: Vectorizer (vector DB + semantic search) • VecLite (embedded single-file vector DB) • Cortex (cognitive substrate — every session, decision and lesson, queryable) • Lexum (distributed full-text search)
Data Infrastructure: Synap (in-memory KV store + message broker) • Nexus (property graph DB with native vector search) • Fluxum (realtime database-as-a-server)
Communication: Thunder (binary RPC — one wire, one codec, six languages) • UMICP (model interoperability protocol, 10 SDKs)
Agent Tooling: Rulebook (rules, specs and task orchestration for coding agents) • Transmutation / Lite (documents → LLM-ready text) • CompressionPrompt (50% fewer tokens, 91% quality) • HiveGPU (GPU-accelerated similarity search)
Platform & Apps: HiveHub.Cloud • Vectorizer Sync (desktop sync app) • TML (language built for deterministic LLM code generation) • Expert (local fine-tuned inference)
HiveLLM is an open-source project, and it's always looking for new contributions. From documentation, implementing new features, contributing to infrastructure or reporting a bug; any contribution is valued and welcome. Are you interested in contributing? Give a read to our Contributing Guide and the numerous ways you can Get Involved with HiveLLM!
The HiveLLM project follows our Code of Conduct. Please abide by this Code of Conduct when interacting with all repositories under the HiveLLM organization and when interacting with people.
Please be mindful that security-related issues should be reported through our Security Policy as security-related issues and vulnerabilities can be exploited and we request confidentiality whenever possible.
Built with 🐝 by the HiveLLM community · Apache 2.0