Skip to main content

TrustGuard is runtime security for AI agents. It inspects prompts, completions, documents, URLs, and tool calls while traffic is in production, then returns a structured verdict. It is not a gateway: it does not route or call model providers. The collector that called it decides whether to allow, mask, ask, or block.

​
How it works

1

A collector sends the event

TrustGate, a coding-agent hook, another gateway, or your SDK posts the payload to TrustGuard.
2

Detectors produce a verdict

Gates match request attributes. Input and output rules run the detectors you attached. The most restrictive status wins: block, ask, transform, report, or allow.
3

The collector enforces it

What the host can actually apply depends on the integration. See How verdicts are enforced.
A collector with no matching policy is unguarded. TrustGuard returns status: "allow" and inspects nothing. Attach a default policy before you rely on it.

Connect a collector

TrustGate, coding agents, other gateways, SDKs, or middleware.

How verdicts are enforced

Block, Ask, Transform, and Monitor — and what each collector can actually apply.
Flow diagram showing a collector sending input or output to TrustGuard, which selects a policy, evaluates gates and detector rules, and returns a verdict for the collector to apply.

​
What it stops

Data loss

PII and secrets. Set Transform to mask them in flight.

Jailbreaks and abuse

Prompt injection, toxicity, topic controls, URLs, and documents.

Tool-borne attacks

Indirect injection in MCP and other role=tool content.
Detectors find risk. The policy decides what happens next.

​
Building blocks

Detectors

A catalog detector plus its settings. Detection only — they do not enforce.

Policies

Gates, Input / Output rules, and Observe vs Enforce.

Collectors

The integration point that sends traffic to a policy.

Evaluate API

POST /v1/evaluate. Most collectors call this; TrustGate does not need a tgk_ key.

​
Get started

TrustGuard is entitlement-gated. If your team does not have it, Agent Runtime does not appear in the console. The fastest path is Agent Runtime → Getting started.
  1. Build a policy from the detector catalog. Leave it in Observe.
  2. Test it on the policy Test tab. The tab evaluates the saved policy.
  3. Connect a collector under Collectors → Catalog, then attach the policy.
Watch Activity, then switch the policy to Enforce.

​
FAQ

No. It evaluates traffic and returns a verdict. The collector applies it. When that collector is TrustGate, the gateway also routes the call.
No. Use another supported gateway, an SDK, middleware, or a coding-agent plugin. See Integrations.
Yes. Observe records findings and downgrades block, ask, and transform to report. Switch to Enforce after Activity looks right.
Yes, when the collector sends tool events. Which actions the host can apply is in How it works.