What Admina is
Admina is an Open Source framework to build AI applications that are governed by design. It bundles a Python SDK (in-process), a Rust network proxy, a CLI, a dashboard and a plugin system. It intercepts model calls and applies bidirectional policies with four actions: ALLOW, BLOCK, REDACT, CIRCUIT BREAK.
Four governance domains
- Data Sovereignty: PII redaction, data residency enforcement, data classification (Rust, 0.62µs per scan)
- AI Infrastructure: LLM engine (Ollama, vLLM), RAG pipeline (ChromaDB), web UI, hot model switching
- Agent Security: anti-injection firewall (15+ patterns, 2.08µs), TF-IDF loop breaker (2.38µs)
- Compliance: EU AI Act classification (Art. 6-15), forensic black box with SHA-256 hash chain on MinIO, native OpenTelemetry
Performance
Median governance overhead: 6.25µs (Rust) across the full 4-domain pipeline. P95 7.04µs, P99 7.29µs. Automated test suite with 700+ cases.
OISG adequate by default
Admina is built around the OISG paradigm (Open, Intelligent, Secure, Governed: oisg.ai). The adequacy score is computed at runtime from the live installation state; a default install reaches the adequate band.
Why Apache 2.0
The project starts from the idea that AI governance must be inspectable. That is why the core is released under Apache 2.0, with no closed version replacing it.
Official project site
Admina is a project by Stefano Noferi, sponsored by noze. Documentation, code and community live at admina.org and github.com/admina-org/admina. Admina Enterprise is noze’s distribution built on Admina, with additional modules and integration, audit and support services.