Open Source

Sentinel Observability

A transparent, fail-open MCP observability sidecar. See every agent-tool interaction with cryptographic audit trails you can verify offline.

Transparent Sidecar Architecture

LLM Client
Sentinel
Observe & Record
MCP Tools

Fail-open by design

<1ms overhead

100% visibility

Canonical Ordered Events

Single, consistent, replayable history with stable event IDs and monotonic ordering.

Session Identity

Stable session_id, trace_id, and span_id for complete request/response correlation.

Cryptographic Integrity

Hash-chained, append-only audit records with Ed25519 signatures.

The Sentinel Philosophy

Observe, never decide.Record, never enforce.Fail open, not closed.
View on GitHub

Sentinel’s Core Guarantees

Identity answers who. Ordering answers in what sequence. Integrity answers can we trust the record. Together, they form the minimum foundation for serious observability - without enforcing policy or constraining agents.

Canonical, Ordered Event Stream

A single, consistent, replayable history of agent activity - with stable event IDs and monotonic ordering derived from observation at the Sentinel boundary (not wall-clock timestamps).

Concurrent tool calls observed at the Sentinel boundary

Key property: ordering is derived from the sequence of bytes observed crossing the Sentinel boundary - not timestamps - allowing a single, replayable history even when async responses race or clocks drift.

Session Identity & Trace Continuity

Stable session_id and trace_id for a run, plus span_id per request with request ↔ response correlation - consistent across tools and errors.

session_id / trace_id / span_id
Without it: logs fragment, attribution breaks, debugging turns into guesswork.

Cryptographic Integrity of Telemetry

Hash-chained, append-only audit records with Ed25519 signatures and optional encryption at rest - verifiable offline. Cryptography applies only to telemetry, never to execution.

hash-chain + Ed25519 (+ optional encryption)
Without it: the record can be edited after the fact and you’ll never know.
Book a Demo

See EngramAI in Action

Get a personalized walkthrough of the Defense Layer Platform. See how we protect AI agents, RAG systems, and MCP servers with 100% detection accuracy.

What’s included:

Live platform walkthrough
Custom threat assessment
Integration planning
Pricing discussion
Q&A with engineering team

Perfect for:

Enterprise AI Teams

Secure your multi-agent workflows and RAG systems at scale.

Financial Services

Meet compliance requirements with cryptographic audit trails.

AI-Native Companies

Ship faster with confidence. Protect your AI infrastructure from day one.

Let’s Talk

Reach out to schedule your personalized demo

We typically respond within 24 hours

or
100%
Detection
<5ms
Latency
Enterprise
Ready