Hi,
We're building VAIG — an open-source runtime governance layer for LLM
inference, designed for EU AI Act Annex III compliance and
insurance-grade auditability. Apache 2.0.
How VAIG maps to ACS
| ACS principle |
VAIG |
| Runtime Interception |
Probabilistic distrust scoring at every inference |
| Proof of Control |
SHA-256 hash-chained WORM audit log, append-only |
| Mandatory Halt |
Binary HALT state, formally verified (TLA+, 4.78M |
|
states, 0 counterexamples) |
| Contextual Guardrails |
DistrustEngine L0–L4 (TRUSTED → MONITOR → WARN → |
|
DEGRADE → HALT) |
Key design principle
VAIG sits architecturally outside the model it monitors — it
cannot be influenced, overridden, or manipulated by the system it
validates. This is the same principle that makes Safety Instrumented
Systems (IEC 61511) reliable in industrial process control.
Status
- Formally verified halt mechanism (TLA+ model-checked)
- WORM audit log producing legally defensible audit trails
- Validated on MedQA and LEDGAR benchmarks
- Designed for deployment on sovereign European infrastructure
We'd like to be listed as a compatible implementation alongside
LangChain/CrewAI. Happy to contribute documentation or a reference
integration guide.
— Njål Solland, VALO Research Group
https://github.com/nsolland/VAIG
pip install vaig
Hi,
We're building VAIG — an open-source runtime governance layer for LLM
inference, designed for EU AI Act Annex III compliance and
insurance-grade auditability. Apache 2.0.
How VAIG maps to ACS
Key design principle
VAIG sits architecturally outside the model it monitors — it
cannot be influenced, overridden, or manipulated by the system it
validates. This is the same principle that makes Safety Instrumented
Systems (IEC 61511) reliable in industrial process control.
Status
We'd like to be listed as a compatible implementation alongside
LangChain/CrewAI. Happy to contribute documentation or a reference
integration guide.
— Njål Solland, VALO Research Group
https://github.com/nsolland/VAIG
pip install vaig