AI Governance & Security v1.0

Ship AI experiences with defense-in-depth.

The comprehensive control plane for AI safety. Enforce policies across prompts, retrieval, and responses without slowing down your builders.

See How It Works Start Securing

Inline PII Redaction & Secret Detection

Retrieval Firewall & Trust Registry

Real-time Capture & Policy Replay

sentix-proxy-logs — tail -f

Architecture

Full Round-Trip Protection.

Sentix inspects the inbound prompt for attacks and the outbound response for data leaks.

Traffic Sources

Apps • Chat UI

Prompt

Response

Bidirectional Proxy

Tilius Sentix

Inspect Input ⇄ Scan Output

Sanitized

Completion

LLM Provider

OpenAI / Anthropic

The Platform

Complete control over AI behavior.

Guard every stage—ingress prompts, retrieval sources, and outbound responses—with tenant-aware policies.

Egress Output Scanner

Strip secrets, PII, and high-entropy tokens in both streaming and batch responses.

DLP

Regex

Entropy

Retrieval Firewall

Inspect RAG queries and sources. Automatically enforce trust tiers on retrieved chunks.

Trust Registry

Citations

Sensitivity

Policy-as-Code

YAML-based rules with hot-reload, linting, and test preview via our CLI or UI.

GitOps

Diffs

Access & Quotas

Tenant-scoped API keys with granular RPS/RPM limits and model allowlists.

Multi-tenant

RBAC

Capture & Replay

Replay past traffic with current policies to measure drift, validate fixes, and compare decisions.

Replay

Drift

Playground & Regex Tester

Batch suites with diffs plus a regex tester to tune patterns and policies before production rollout.

Batch

Regex

Suppressions Workflow

Mark findings as false positives with scoped suppressions, expiry, and suggested regex adjustments.

False Positive

Suppress

Integration

Two ways to consume.
One security layer.

Developers can drop our proxy into their code, while analysts can use our secure, tenant-scoped chat interface immediately.

Compatible with OpenAI/Anthropic SDKs
Zero latency overhead (< 20ms)
Drop-in replacement (base_url change)

main.py

from openai import OpenAI

# Option 1: The Proxy Approach
client = OpenAI(
    base_url="https://proxy.tilius.com/v1",
    api_key="sentix-tenant-key"
)

# Your existing code stays exactly the same
resp = client.chat.completions.create(
    model="gpt-4",
    messages=[...],
    stream=False
)

Enterprise Capabilities

Prompt Injection ML

Detects jailbreaks and hostile inputs before they reach the model.

Privacy Controls

Log scrubbing, redact-at-ingest, and configurable retention windows.

Observability

Full visibility via Prometheus metrics, audit logs, and SIEM connectors.

Org Store Indexing

Auto-attach trusted demo corpus data (MD/PDF) for instant RAG testing.

Tuning & Replay

Replay past traffic against new policies to visualize drift and catch regressions.

Cost Governance

Hard token caps per request to prevent wallet-draining loops or attacks.

Suppressions

Mark-and-expire false positives with scoped suppressions and regex suggestions.

Trust & Sensitivity

Domain trust tiers, citations, and sensitivity labels to tighten thresholds automatically.

Embedded Chat UI

Tenant-scoped chat experience that routes through Sentix policies—ready for analysts and read-only users.