AI risk monitoring for production LLMs

Catch hallucinations, prompt injections, and jailbreaks before they reach your users — with a score, a reason, and an action for every response.

Live Demo  •  Documentation  •  API Reference  •  PyPI

---

The problem

LLMs are confident — and wrong. They invent citations, flip numbers, and blend entities with zero hesitation. Most teams find out after a bad response reaches a user, or from a compliance auditor.

SentinelAI sits between your app and your LLM, scoring every prompt/response pair in real time. Six detectors run in parallel, every score ships with token-level reasons, and your policy decides the action: serve, auto-correct, or block.

No black boxes. Every verdict explains itself.

Quick start

pip install sentinelai-risk

from sentinelai import SentinelAIClient

client = SentinelAIClient(api_key="sk_...")

prompt = "What was Q3 revenue?"
llm_output = "Revenue grew 45% year over year."

result = client.verify(prompt=prompt, response=llm_output)

if result.status == "hallucinated":
    print(result.corrected)  # serve the fix, not the flaw
else:
    print(result.trust_score, result.reasons)

Live in under 60 seconds. See the quickstart guide for configuration, deployment modes, and self-hosting.

Architecture

SentinelAI is model-agnostic and minimally invasive — it observes prompt/response pairs and never sits in the generation path:

flowchart LR
    Client[Client Application]
    LLM[AI / LLM]
    Sentinel[SentinelAI API]
    PromptMon[Prompt Anomaly Detector]
    OutputMon[Output Risk Scorer]
    Aggregator[Risk Aggregator]
    DB[(Risk Logs)]

    Client --> LLM
    LLM --> Sentinel
    Sentinel --> PromptMon
    Sentinel --> OutputMon
    PromptMon --> Aggregator
    OutputMon --> Aggregator
    Aggregator --> DB
    Aggregator --> Client

Loading

1. The client application sends prompt and model response to SentinelAI
2. Prompt anomaly detection checks for distribution shifts
3. Output risk scoring flags unsafe or unstable responses
4. Risk signals are aggregated into a unified score
5. Results are returned and optionally logged for review

How a response is scored

flowchart LR
    App[Your App] --> API[SentinelAI API]
    API --> Detectors[6 Parallel Detectors]
    Detectors --> Reasoner[Risk Reasoner]
    Reasoner --> Policy[Policy Engine]
    Policy -->|0-24 Trusted| Serve[Serve as-is]
    Policy -->|25-59 Review| Fix[Auto-correct or flag]
    Policy -->|60-100 Block| Block[Block + escalate]

Loading

Trust score	Status	Default action
0–24	Trusted	Serve as-is
25–59	Needs review	Auto-correct or flag for humans
60–100	Hallucinated	Block and escalate

Deployment modes

Mode	Use case
Blocking	Verify every response before it hits your user
Monitoring	Log everything, review flagged ones later
Async	High-throughput: fire-and-forget + webhooks
Self-hosted	Keep all data on your own infrastructure

What you get back

Field	Meaning
status	trusted, needs_review, or hallucinated
trust_score	0 (safe) to 100 (critical risk)
reasons	Token-level explanations for every flag
corrected	A cleaned response when auto-correction applies

Features

• Six detectors in parallel — fabricated citations, numeric drift, entity confusion, contradictions, unsupported claims, overconfidence
• Explainable risk — every score ships with token-level reasons, no black boxes
• Auto-correction — serve a cleaned response instead of a risky one
• Conversation-aware — risk judged across multi-turn context, not in isolation
• 3-line SDK — pip install sentinelai-risk and you are live
• Self-hostable — open-source core, your data stays on your infrastructure

Documentation

Full docs — trust score, detectors, deployment modes, API reference, and self-hosting — live at https://blacksujit.github.io/Sentinel-AI/.

The docs site is a static Next.js/Fumadocs build in docs-site/, exported with a /Sentinel-AI base path and deployed to GitHub Pages by .github/workflows/pages.yml on every push to main.

cd docs-site
npm install
npm run dev

Repository layout

Path	Contents
docs-site/	Documentation website (Next.js + Fumadocs, static export)
Backend/	SentinelAI API backend (FastAPI)
Frontend/	Web dashboard
sentinelai-sdk/	Python SDK (published as sentinelai-risk)
Docs/	Source documentation & design notes

Roadmap

• Prompt drift detection, structured logging, alerting
• Feedback-driven calibration, CI/eval integration, automated red-teaming
• EU AI Act / SOC 2 / ISO 42001 compliance reporting

License

MIT