Documentation

Welcome to the Admina documentation. Admina is an open source governance proxy (Apache 2.0) for AI agents and LLM systems. It intercepts every interaction between agents and tools, applying 6 governance pillars bidirectionally.

What is Admina?

Admina is a transparent proxy that sits between AI agents and the tools they control (MCP servers, LLM APIs, databases, filesystems). Every request and response passes through all 6 governance pillars.

Key properties:

  • Bidirectional โ€” both requests (agent โ†’ tool) and responses (tool โ†’ agent) are governed
  • MCP Native โ€” speaks JSON-RPC 2.0, designed for Model Context Protocol
  • Hybrid runtime โ€” Rust core (~14ยตs overhead) with automatic Python fallback
  • OpenClaw ready โ€” skill for ClawHub, stdio bridge, adapter included
  • EU AI Act compliant โ€” risk classification and gap analysis built in
  • Apache 2.0 โ€” free to use, modify, and distribute forever

Architecture at a glance

Admina runs as a Docker Compose stack of 9 containers: proxy, dashboard, ClickHouse, MinIO, Redis, OpenTelemetry collector, Grafana, mock MCP server, and mock agent.

# The full stack in one command
docker compose up --build

# Services started:
#  proxy      โ†’ :8080  (governance proxy + REST API)
#  dashboard  โ†’ :3000  (real-time governance UI)
#  otel       โ†’ :4317  (OpenTelemetry collector gRPC)
#  grafana    โ†’ :3001  (metrics and traces)
#  minio      โ†’ :9000  (forensic black box storage)

Enterprise features

Need more? DebugABot is built on Admina and adds ML-based injection detection, eIDAS-qualified forensics, SSO/RBAC, sycophancy detection, behavioral control, and managed cloud. See Enterprise for details.