Works with OpenClaw + NemoClaw

AI Safety for NemoClaw Agents

NemoClaw guards the container. EvalGuard guards the AI inside it.

Get Started Free View Integration Guide

NemoClaw secures the runtime. But who secures the AI?

Container Isolation

Covered by NemoClaw

LLM Output Safety

Not covered

Network Policy

Covered by NemoClaw

Prompt Injection Detection

Not covered

Credential Vault

Covered by NemoClaw

PII Redaction

Not covered

EvalGuard adds the AI safety layer

A complete guard pipeline that wraps every LLM call, inspects every response, and reports everything to your dashboard.

Agent Code→EvalGuard Pre-Guard→NemoClaw Privacy Router→LLM

Response flows back through Post-Guard

Agent Logic←EvalGuard Post-Guard←Response

↓

EvalGuard Trace Collector→Dashboard

Everything you need to secure AI agents

Eight layers of protection, from prompt to response, from development to production.

Prompt Injection Detection

Real-time, 5-layer detection engine catches direct injections, indirect injections, jailbreaks, payload smuggling, and encoding attacks.

PII Redaction

Regex + NER dual-engine pipeline redacts personally identifiable information before data ever reaches the LLM.

Content Safety

Toxicity, bias, and harmful content scoring with configurable thresholds and automatic blocking.

Hallucination Detection

Factual accuracy scoring against grounding documents with citation verification and confidence calibration.

Compliance Reporting

Auto-generated compliance reports for EU AI Act, DPDP Act, and NIST AI RMF with audit trails.

Cost Attribution

Per-agent, per-model, per-team cost tracking with budget alerts and spend forecasting.

Red-Teaming

246 attack plugins across 43 strategies with scheduled scans and regression tracking.

Continuous Monitoring

Drift detection, anomaly alerts, and real-time dashboards for production agent observability.

Five lines to protect every agent

Install the SDK, wrap your agent, and every LLM call is guarded automatically.

agent.ts

import { init } from "@evalguard/nemoclaw";

const agent = init({
  apiKey: "eg_...",
  agentName: "support-bot",
  guards: ["prompt_injection", "pii_redact", "hallucination"],
});

// Every LLM call through this agent is now guarded
const response = await agent.chat("How can I help you?");

Complete the security stack

Feature	NemoClaw Only	NemoClaw + EvalGuard
Container sandboxing
Network policy enforcement
Credential vault
Resource isolation
Prompt injection detection
PII redaction
Content safety scoring
Hallucination detection
Compliance reporting
Cost attribution
Red-team scanning
Drift & anomaly monitoring