admina
An open framework for governed AI development that redacts personal data, blocks prompt injection, audits agent tool calls, and tracks EU AI Act compliance, installed as a proxy or a small Python SDK.
admina redacts personal information such as email addresses, national IDs, phone numbers, IP addresses and names from prompts and responses by default, blocks prompt injection at the proxy using regular-expression checks plus a Rust heuristic scorer, and validates agent tool calls before they run, logging them afterward into a hash chain for a tamper-evident audit trail.
It breaks runaway or looping agents using similarity scoring over the action stream, and includes a built-in gap analysis and risk classification against the EU AI Act. Existing code can adopt it with zero changes through the proxy, or with a few lines through the SDK.
A scaffolding command boots a governed proxy and dashboard with no Docker required; the default installation is pure Python, with an optional Rust engine for lower overhead once broader detection coverage is no longer the priority. It is Apache 2.0 licensed, written in Python and Rust.