AI risk monitoring
for production LLMs
Catch hallucinations, prompt injections, and jailbreaks before they reach your users.
pip install sentinelai-riskPyPICatches what filters miss
Naive filters only match keywords, so attackers paraphrase around them. Three detectors run in parallel — prompt anomalies, jailbreak attempts, and risky output across 8 categories like violence, hate speech, and self-harm.
Learn moreKnow why it was flagged
A bare score tells you nothing. Every verdict ships with the reason, the flags that triggered it, and the thresholds applied — so you can replay any score against the policy that produced it.
Learn moreAct, don't fail
A risky response isn't a dead end. The policy engine allows, warns, blocks, or escalates — and correct() returns a cleaned response instead of an error.
Learn moreWatch the whole conversation
One prompt can look harmless while the conversation drifts. ConversationTracker scores every turn and rolls them up into conversation-level risk.
Learn moreLive in three lines
pip install sentinelai-risk, call verify(), and get a 0–100 score with a clear verdict. Retries, parallel batches, and API-key auth are handled for you.
Learn moreKeep data on your network
The core is open source and self-hostable. Same SDK, same API, your own server — with SQLite or PostgreSQL behind it.
Learn moreDeploy AI with confidence
Every organization deploying LLMs needs to know when the model is wrong. SentinelAI makes risk visible, explainable, and controllable.
Get started in 60 seconds