AI.Guard

Runtime control for every LLM request.

A drop-in gateway that sits between your applications and any upstream model. Inspects prompts, masks sensitive data with reversible tokenization, blocks injection attempts, filters outputs, and produces a signed audit trail — without the application changing how it talks to OpenAI, Anthropic, or your own models.

Explore Capabilities How It Works

Live Gateway Traffic

policy v2.34

request #1 — allowed

[REQ]POST /v1/chat — summarize patient discharge note

[DLP]PHI detected — patient record fragment

[MASK]name → [REDACTED_NAME] · ssn → [REDACTED_SSN]

[FWD]→ openai/gpt-4o · 412 tokens · 11.4 ms

[RSP]200 OK — response unmasked for the caller

[AUD]audit log signed → splunk-prod-1

request #2 — blocked

[REQ]POST /v1/chat — “ignore previous instructions…”

[SCAN]prompt-injection score 0.94

[BLK]request denied — never forwarded upstream

[AUD]audit log signed → splunk-prod-1

What you get with AI.Guard

A single control point for the runtime behavior of every model your organization uses — across providers, regions, and deployment topologies.

End-to-end architecture

See exactly where the gateway sits, what inspection layers run per request, and what leaves your perimeter under each deployment topology.

View How it Works

Deep policy & DLP

Injection scoring, reversible tokenization for PII / PHI / PCI / secrets / IP, output filtering, policy-as-code, and per-tenant quotas.

View Capabilities

Fits your stack

Drop-in OpenAI / Anthropic compatibility, SDKs in five languages, export to your SIEM / SOAR / identity / secrets backends.

View Integrations

Latency

p50 ≈ 11 ms added

p95 < 17 ms — measured on non-streaming chat completions during pilots; your numbers are validated on your own traffic.

Coverage

Every major provider

OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, Google Vertex AI, Mistral, and self-hosted vLLM / Ollama / TGI.

Deployment

SaaS, single-tenant, on-prem

Including fully air-gapped on-prem, where no prompt or response ever leaves your perimeter.

Ready to put a perimeter around your models?

Start a guided two-week pilot for one LLM. We map traffic, configure a focused policy set, and deliver a signed risk report at the end.

View Pricing & Pilot