← Back to agents
🛡️

AI Firewall Agent

Controls every request to your AI agents at runtime. Access policies are enforced in production and every decision is traced.

What the agent does

🛡️ Runtime control

Every request to your agents is checked in production, with no change to application code.

🚦 Access policies

Allow or block by user, context and request type.

🔁 Multi-turn tracking

Detection of progressive attacks built over several exchanges (crescendo).

🔍 Content inspection

Runtime analysis of requests to spot injections and malicious payloads.

🧾 Enforceable proof

SHA-256 chained audit trail of every decision, aligned with articles 9 to 15.

⚡ Controlled latency

Inline control designed to preserve the production experience.

How it works

1

Interception

All requests to your agents pass through the firewall

2

Analysis

Verification of identity, context and content

3

Decision

Allow or block the request at runtime

4

Proof

Decision sealed in the chained audit trail

Technical approach

Inline proxy

On the flows to your agents, with no change to application code.

Declarative policies

Readable rules, evaluated on every request.

Conversation tracking

Correlation of a client's exchanges to detect escalating attacks.

Runtime detectors

Indirect injection, dangerous actuation, alarm masking, exfiltration.

SHA-256 chained trail

Enforceable proof of every block.

Sovereignty

On-premise deployment, no cloud dependency.

Use cases

PRE-DEPLOYMENT

Exposure measurement

Quantifying the attack rate of requests before production.

PRODUCTION

Access control

Enforcing policies at runtime on an already-deployed system.

COMPLIANCE

Proof of decisions

Chained audit trail of blocks, aligned with the EU AI Act.

Control access to your agents

Submit your system, we return an exposure report with the measured attack rate.