0 requests screened this hourv1.0 GA
The Governance OS for Regulated AI

Deploy, monitor, and prove the safety of every AI in production.

Sentry stands in front of every model call — screening prompts, enforcing policy, and writing an immutable audit trail regulators can verify.

SOC 2 Type II HIPAA BAA EU AI Act Art. 14 NIST AI RMF FedRAMP Moderate ISO 42001
Live Playground
Gateway will screen the request against all enabled policies.
sentry · live decision feed streaming
00:09:21ALLOW Draft marketing email · gpt-4o · 3.2ms
00:09:17REDACTPII: Email Redaction · redacted 1 span · 2.8ms
00:09:12DENY Prompt Injection Guard · pattern match
00:09:08ALLOW Summarize Q4 earnings · claude-sonnet-4.5 · 4.1ms
00:09:04FLAG Toxicity Flagging · escalated to review
00:08:59DENY PII: US SSN · pattern match · alerted
00:08:55ALLOW Recommend recipe · gemini-3-flash · 1.9ms
00:08:51REDACTPII: Email Redaction · redacted 2 spans · 3.1ms
00:08:46ALLOW Convert 32C→F · gpt-4o-mini · 1.4ms
00:08:42DENY Jailbreak: DAN · pattern match · alerted
00:08:38ALLOW Explain photosynthesis · gemini-3-flash · 2.2ms
00:08:34FLAG Article 14 · high-risk domain · logged
$ sentry tail --tenant demo ▍
Live from production

Every decision. In real time. On the record.

Aggregate metrics streamed from the Sentry Evidence Plane across all customer tenants. Refreshed every second.

Requests screened / 24h
0
across all tenants
+ 12.4% vs 24h prior
Denied / redacted
0
policy-blocked at gateway
0.22% overall block rate
Gateway p95 latency
0.00 ms
policy evaluation
under 4ms SLO
Signed audit bundles
0
last 30 days
7-yr retention
SentryDSL
Regex-first policy language. Version-controlled, compilable, testable in the browser.
Live Decisions
Every prompt, redaction, deny — streamed to Observatory with per-request causal trace.
Reproducible
Any decision replayable against any policy version. Regulators verify the artifact.
Hot-path fast
Sub-4ms p95 policy evaluation. Never the bottleneck for a real-time LLM app.
The Sentry Architecture

Three planes. One guarantee.

Data plane isolated, control plane shared. If Evidence goes down, Gateway keeps serving traffic. If Policy goes down, Gateway serves on the last-known-good bundle. There is no single point at which observability breaks enforcement.

Availability
99.99%
Regions
4 · US/EU
SDKs
Node · Py · Go
Data Plane

Gateway

Horizontally scaled inline proxy in Go. +6ms p95 over direct provider calls. The only component on the hot path — memory-safe, stateless, region-local.

p95 latency
3.8 ms
Deployment
K8s / Nitro
Control Plane

Policy Plane

Versioned, content-addressable policy store. Author in SentryDSL, compile to bytecode, distribute to Gateway pods in under 15 seconds. Every version SHA-256 signed.

Rollout
<15 s
Integrity
SHA-256
Control Plane

Evidence Plane

Append-only audit ledger on ClickHouse + S3. Every decision, prompt, response, and policy bundle — queryable by auditors, exportable to Snowflake and Databricks.

Retention
7 yrs
Backends
CH + S3
clientGateway · evaluatePolicy Bundle vNProvider (OpenAI / Anthropic / Google)Evidence PlaneSigned Bundle · SHA-256
The Night Watch

Observatory. Every AI decision, made queryable.

Observatory ingests every decision from Sentry's Evidence Plane and builds a Causal Trace Graph — prompt, context, retrieval, tools, response, user, and business action, all connected.

  • Decision search + 5 prebuilt dashboards
  • Causal Trace Graph (CTG) with root-cause inference
  • AI Seer — LLM-powered root cause & remediation
  • Drift detection: embedding + behavioral
  • Slack, PagerDuty, Datadog, Splunk export
REQUESTS
18.4M
last 24h
DENIED
412
policy-blocked
P95 LATENCY
3.8ms
gateway
SPEND
$12,847
provider cost
# POST /api/v1/gateway/evaluate
{ "prompt": "Email me at jane@example.com", "model": "gpt-4" }
# response
action: redact
policy: PII: Email Redaction
latency_ms: 2.14
Compliance is the moat

Built for the frameworks auditors already trust.

Every deploy ships with mapped controls, signed evidence, and reproducible policy versions. No spreadsheet gymnastics.

Ready
EU AI Act
Article 14 human-oversight logging, high-risk categorization, incident reporting.
Ready
NIST AI RMF
Govern, Map, Measure, Manage — control mappings shipped in evidence bundles.
Audited
SOC 2 Type II
Security, Availability, Confidentiality. Continuous evidence collection.
Signed
HIPAA BAA
PHI redaction pre-provider, at-rest encryption, signed access ledger.
Roadmap
FedRAMP
Moderate baseline. High baseline via Citadel + Enclave with hardware attestation.
Aligned
ISO/IEC 42001
AI management system controls mapped to Sentry policy families 1:1.

Turn AI compliance into a shipped artifact.

Signed audit bundles. Reproducible policies. Answers regulators can verify.

Live gateway · never a mock