OPEN SOURCE · SELF-HOSTABLE · APACHE-2.0
Own your agents.
Eparch is an open-source, self-hostable control plane for building, deploying, and governing AI agents on open-weight models — on infrastructure you own.
$ make quickstart✓ secrets generated✓ monolith started (2 containers)→ open http://localhost:8080Agent SaaS you can't run in your environment.
DIY plumbing that takes quarters to build.
Eparch: the platform layer, on your hardware.
ARCHITECTURE
Ten services. Your infrastructure. Nothing else.
Eparch runs entirely on components you operate: PostgreSQL with pgvector for state, events, vectors, and encrypted secrets; Redis for rate-limit and budget counters; MinIO or any S3 store for documents; and OPA for authorization. The model backend is an OpenAI-compatible endpoint you point at your own inference — vLLM, Ollama, or any compatible provider. No managed cloud dependency; a single VM or an air-gapped network is enough.
GOVERNANCE
Every request governed, before it reaches a model.
Authenticated, policy-checked, rate-limited, budget-checked, cached, audited — by default.
CAPABILITIES
The plumbing every serious agent deployment needs.
LLM gateway
Hybrid RAG
Tool registry + MCP
Canary + evals
Hard multi-tenancy
Immutable audit
QUICKSTART
Running in one command.
$ git clone https://github.com/eparchai/eparch && cd eparch && make quickstart# requires only Docker + openssl — no model server, no accounts✓ secrets generated✓ monolith started (2 containers)→ open http://localhost:8080 and register the first account (it becomes the admin)To wire up a real model backend or enterprise SSO, follow theself-host walkthrough.
ENTERPRISE
Deploying somewhere the cloud can't go?
Air-gapped networks, regulated industries, sovereignty requirements — Eparch was designed for the environments other agent platforms can't enter.