Private beta / autonomous action firewall

Secure intent before decision passes.

Aurels is the intent firewall for autonomous actions: policy, behavior, semantic intent, and a signed audit record before an agent can execute high-consequence work.

Run console Request pilot access Request pilot access →

Ecosystem & visibility

Built within a trusted ecosystem

Public ecosystem relationships and coverage — not customer endorsements.

target p99

< 50ms

Rules and policy checks stay out of the critical path.

control layers

Policy, behavior, and intent analysis before execution.

hard-rule misses

Deterministic denials remain deterministic.

Before execution

Every action earns its verdict.

Aurels turns a proposed agent action into a decision your runtime can enforce — with the evidence to explain it afterwards.

Try a live action

agent actionPay €4,800 · Acme Corp

REQ_8F2A

Policyscope · limit · recipientPASS

Behaviorvelocity · anomaly · driftPASS

Intentpurpose · context · mandatePASS

verdict ALLOWsigned audit · sha256: 4afd…1990

Live console / no signup required

Action checkpoint

Submit a financial action as the live example and watch the gate decide: policy pass, behavior pressure, semantic intent, final verdict, signed audit.

intent analyzer · online

Choose your view

01 / Intent

Describe the action

Agent

Amount

Currency

Recipient

02 / Analysis

Run the gate

Policy checks●IDLE

Behavior pressure●IDLE

Semantic intent●IDLE

Aurels checks policy, behavior, and intent before execution.

03 / Decision

Enforce the verdict

Current result

Ready

Defense layers

Three stops before execution.

Aurels checks the action, the operating lane, and the evidence before a tool runs.

Policy

Checks targets, limits, routes, and permissions before execution.

Behavior

Flags unusual bursts, retries, drift, and repeated attempts.

Evidence

Signs each verdict with policy, payload, node, and timestamp.

POST /api/verify

One call. One verdict.

Send the action details and execution trace. Aurels evaluates deterministic rules first, checks behavior, analyzes intent, and returns a verdict the runtime can obey.

allowAction is safe to execute

flagRequires human review before proceeding

blockAction must not execute

Request

POST /api/verify
x-api-key: ig_your_api_key

{
  "intent_id": "act_abc123",
  "agent_id": "agent_gpt4",
  "amount": 4800.00,
  "currency": "USD",
  "recipient": "vendor@acme.com",
  "agent_context": "User approved a $4,800 invoice
    payment to Acme Corp for Q2 services.",
  "metadata": { "action_type": "payment" }
}

Response · 200 OK

{
  "decision": "allow",
  "reason": "Intent verified before execution",
  "risk_score": 8,
  "evaluated_at": "2026-06-22T14:30:00Z",
  "intent_id": "act_abc123"
}

Access tiers