60 detectors out of the box
Prompt injection, jailbreaks, PII / credential leakage, hallucinated citations, repetition loops — security, hallucination and operational coverage from day one, plus opt-in EU AI Act Art. 5 detectors.
Drop-in IChatClient middleware
Wraps any Microsoft.Extensions.AI client. Two pipeline passes per call (prompt + response). No provider-specific code.
Custom detector SDK
Write your own detector by implementing IDetector. Test it with DetectorTestBuilder + FakeEmbeddingGenerator — no API keys, deterministic results.
Embedded dashboard
Live audit feed, threat scores, severity timeline. HTMX + Server-Sent Events — no JS framework.
Per-pipeline configuration
Multiple isolated pipelines under string names — multi-endpoint, dev/staging/prod tiers. Per-detector tuning via opts.Configure<T>.
Audit forwarders
NDJSON file, SQLite, Azure Sentinel, OpenTelemetry — pick one or several. Hash-chained audit ensures tampering is detectable.
Redaction by default
Personal data and secrets in a prompt are replaced with placeholders before the model sees them, and put back in the reply. The in-process client redacts; a redaction is audited, never counted as a threat.
Art. 50 transparency
Opt-in: mark every reply as AI-generated in ChatResponse.AdditionalProperties and add a notice once per conversation or on every reply. Each first disclosure is audited.
EU AI Act compliance report
sentinel report turns the audit file into totals per article as Markdown, HTML or JSON. Evidence to support an assessment, not a statement of compliance.
One line wires the middleware
Configure
builder.Services.AddAISentinel(opts =>
{
opts.OnCritical = SentinelAction.Quarantine;
opts.OnHigh = SentinelAction.Alert;
opts.OnMedium = SentinelAction.Log;
});Wire into your chat client
builder.Services.AddChatClient(pipeline =>
pipeline.UseAISentinel()
.Use(new OpenAIChatClient(...)));
// optional embedded dashboard
app.MapAISentinel("/ai-sentinel");The dashboard your security team will actually open
Live threat-risk score, severity counters, and a real-time event feed — every detection is hash-chained, filterable, and one click away from the audit forwarder of your choice. No JS framework, no separate service.
15 packages, one ecosystem
AI.Sentinel
Core — pipeline, 60 detectors, intervention engine, audit store.
AI.Sentinel.Detectors.Sdk
SDK for writing + unit-testing custom detectors.
AI.Sentinel.AspNetCore
Embedded real-time dashboard. HTMX + SSE.
AI.Sentinel.Sqlite
Persistent SqliteAuditStore with hash-chain integrity.
AI.Sentinel.Mcp
Stdio MCP proxy — scans tools/call + prompts/get for any MCP host.
AI.Sentinel.ClaudeCode
Native hook adapter for Claude Code's settings.json.