Alpha
Hivenet Router is an open-source project maintained by Hivenet. It puts one client-facing API in front of multiple inference servers and routes each request using current agent health, load, latency, model, and capability data.
One endpoint for self-hosted inference
Route requests across vLLM, Ollama, SGLang, llama.cpp, Infinity, and custom OpenAI-compatible backends using policies, live signals, fallback, authentication, and observability.
How it works
Connect
Agents connect inference backends to the router and report their models, capabilities, capacity, health, and live signals.
Route
The router applies model and capability constraints, policy gates, ranking, queues, retries, and fallback.
Observe
Prometheus metrics, Grafana dashboards, traces, and audit records show how requests move through the fleet.
Start building
Quickstart
Deploy a router and three agents, then send and observe your first routed inference request.
Architecture overview
Understand the control plane, data plane, storage, authentication, and request flow.
Docker quickstart
Run a router and two vLLM agents across Linux hosts with Docker.
Use the API
Send chat, embedding, reranking, model-list, and administration requests.
Explore by task
Routing and policies
Filter, rank, queue, and fall back across agents.
Security and auth
Authenticate agents and clients and enforce access controls.
Observability
Monitor router, agent, engine, and hardware behavior.
Integrations
Connect Claude Code, OpenCode, Pi, Open WebUI, and applications.
Reference
Look up configuration, architecture, errors, and performance.
Project
Contribute, report security issues, and review project policies.