1. Governed execution
AQEVON is designed around explicit identity, authorization, and policy checks before any consequential enterprise action is taken, and independent outcome verification after it. In our architecture, an AI model's reasoning is intended to inform an execution decision, never to create authority on its own — every action that touches enterprise systems passes through a separate, deterministic identity, authorization, and policy layer first, and its outcome is verified and recorded afterward.
2. Data boundaries
The product architecture separates experience (chat/voice interface), intelligence (model orchestration), trusted execution (policy-gated connectors), and enterprise infrastructure (customer systems of record) into distinct layers, so that enterprise data and the actions taken on it can remain behind explicit, auditable control boundaries rather than inside a single undifferentiated model context.
3. Encryption & infrastructure
- The website uses HTTPS, and the live voice preview uses secure WebSocket connections (WSS).
- Public-facing endpoints — including the voice preview's registration and session infrastructure — sit behind Cloudflare, which provides DDoS mitigation, bot detection, and TLS termination in front of our own infrastructure.
- The live voice preview routes to a single, purpose-dedicated network endpoint, isolated from AQEVON's internal admin and customer-facing product infrastructure, specifically so that a public, anonymous-visitor-facing demo never shares a trust boundary with authenticated customer systems.
4. Secrets and access
Production deployments are designed to keep credentials and API secrets outside source-controlled client code, injected at runtime through environment configuration instead. We apply least-privilege access to enterprise connectors and infrastructure, and administrative access to production systems is limited to the personnel who need it to operate the service.
5. Auditability
Meaningful execution steps in the product are designed to produce a trace covering the originating request, identity and authorization checks, the policy decision made, the action taken, and outcome verification. The exact controls and retention available for a given deployment depend on that customer's agreement and configuration.
6. Live voice preview
Because this is the one place the public internet directly reaches a live AI pipeline on this Site, it gets its own deliberate constraints, independent of the rest of the product:
- Isolated, synthetic demo tenant. The preview runs against a dedicated demonstration tenant containing synthetic data only — it has no path to any real customer's data or any other AQEVON tenant.
- Gated access. A verified work email and a Cloudflare Turnstile bot-check are required before a session can start, to keep the preview from being abused as an open, anonymous compute resource.
- Hard server-side limits. Every session has a maximum duration and a maximum number of conversational turns enforced on our servers (not merely displayed in the browser), and a per-visitor concurrent-session cap, to bound cost and abuse risk.
- No training on your voice. Audio captured during a preview session is used only to run that session; we do not use it to train or fine-tune AI models.
7. Responsible AI principles
We build AQEVON on the premise that an AI system should never be the sole authority for a consequential action. Policy, authorization, and verification are deterministic layers the model cannot talk its way around. Where a system's output could materially affect a person or a business outcome, we favor designs that keep a clear, inspectable decision trail over designs that are merely convincing. We also try to be plain about what AQEVON does not yet do, rather than imply a maturity the product hasn't reached — see the next section.
8. Current stage & what we don't claim
AQEVON is in MVP and market-validation stage. This page describes our design principles and current practices in plain language. It is not a claim of certification, independent third-party audit, or compliance with a particular security framework (such as SOC 2, ISO 27001, or HIPAA) unless and until we say so explicitly, with evidence, on this page. If a specific compliance posture matters for your evaluation of AQEVON, ask us directly — we would rather tell you exactly where we stand than let a generic security page imply more than is true.
9. Reporting a vulnerability
If you believe you've found a security vulnerability in this website, the live voice preview, or any AQEVON system, please report it to support@aqevon.com. Please include enough detail to reproduce the issue, and avoid accessing or modifying data that isn't yours in the course of testing. We ask that you give us a reasonable opportunity to investigate and address a report before any public disclosure, and we will acknowledge genuine, good-faith reports promptly.
