Sentinel

About Sentinel

Sentinel is an open-source observability and trust layer for AI agents. With one line of code, every prompt, response, tool call, and retrieval becomes a traceable, scored, auditable event — with hallucination detection, cost tracking, guardrails, and alerts built in. Runs entirely on your infrastructure.

Features

• Universal interceptor (wrap any client, trace any function)
• Trace waterfall (Jaeger-style view of agent runs)
• Trust scoring (confidence, groundedness, repetition, refusal, factuality)
• Guardrails and alerts (secret leaks, prompt injection, runaway loops)
• PII redaction before egress
• Prompt registry with git-like versioning
• Cost and token tracking per-step and per-model
• OpenTelemetry and Prometheus support

Installation

pip install sentinel-llm