Zero-Trust Containment for Autonomous AI

Stop your LLM agents from escaping the sandbox. API Guardrail is a semantic egress filter: it inspects every outbound request, evaluates intent against a System One decision model, and blocks SQL injection, sandbox escapes, and data exfiltration (hype measurements pending ~tens of ms) before the payload reaches the public internet.

Point your agent’s egress traffic to https://api.fastroute.fyi (also reachable via api.apiguardrail.com) to begin.

What it blocks

• SQL injection & path traversal payloads
• Command execution, port scanning & sandbox/jailbreak escape
• Credential & API-key exfiltration
• Unauthorized data exfiltration

Built for engineers, not hype

No CLM-8B. No “our model escaped” stories. Uses the real protectai/deberta-v3-base-prompt-injection classifier as a secondary signal, backed by the operator’s own Jev System One router as the primary decision layer.

Hosted on sovereign infrastructure (Contabo VDS). Self-hostable.