Security
Last updated 2026-05-18
ATE·IQ is designed for engineering teams whose IP — silicon designs, IG-XL test programs, STDF lots, customer specs — cannot leave their network. The architecture reflects that constraint. This page describes the security posture at the level a customer's due-diligence team needs. For implementation specifics required by a vendor questionnaire, email [email protected] with subject prefix SECURITY and we'll share what's appropriate under NDA.
The product runs on your network
- Single Windows installer. Single-tenant. Runs on your engineer's laptop or a dedicated bench machine.
- All product services bind to the local loopback interface — never to a LAN-routable address. Nothing in the install is reachable from another machine on the network without an explicit reverse-proxy setup that you control.
- No telemetry. No phone-home. No usage analytics flowing out of the product.
- The product's database is bundled with the install and lives in the user profile directory on the engineer's machine.
Bring your own LLM keys
- You configure an API key for your chosen LLM provider — Anthropic, an OpenAI-compatible endpoint, or Azure OpenAI. The product does not bundle a key and does not see prompts until you provide one.
- Outbound LLM traffic flows directly from your machine to your chosen provider over your network. ATE·IQ has no proxy or relay in the middle.
- If you point the provider config at an internal LLM (e.g. Azure OpenAI in your tenant, or a self-hosted gateway), outbound stays inside your perimeter.
Encryption
- Stored secrets — git access tokens, LLM provider keys, requirements-system credentials — are encrypted at rest with industry-standard authenticated symmetric encryption before insertion into the local database.
- The encryption key is loaded from a customer-controlled environment variable. It is never bundled in the application image.
- The local HTTPS surface used by the Excel task pane uses certificates from a CA trusted in the Windows root store of the engineer's machine. No public-internet certificate authority is involved.
File access
- When the Excel task pane reads a workbook from disk, the local server enforces a path allow-list: paths must resolve under the engineer's user profile or a recognised OneDrive sync root.
- Path-traversal attempts and absolute paths outside the allow-list are rejected before any file handle is opened.
- File extensions are validated against the expected test-program formats before parsing.
Authentication (multi-user installs)
- If the install is used by more than one engineer, authentication is required. Sessions are HTTP-only cookies bound to the local loopback host.
- Failed login attempts are rate-limited per source. Password verification runs in constant time relative to whether the username exists, to remove the side-channel that would let an attacker enumerate valid accounts.
- Page-level access checks are enforced at the route boundary; API endpoints re-check authorisation independently rather than inheriting it from middleware.
The marketing site (this site)
- Served from a global edge CDN. The only dynamic surface is the contact form, which posts to a single edge function.
- No analytics, no tracking pixels, no third-party JavaScript. The only outbound from the form is to a transactional email provider, then on to a Google Workspace inbox at
[email protected]. - The Cloudflare zone has WAF and bot management enabled at the edge.
What we don't claim
ATE·IQ is a pre-launch product operated by a small Ireland-registered company. We have not yet completed a SOC 2 / ISO 27001 audit and we don't claim certifications we don't hold. What we offer pre-pilot is:
- A signed NDA, turned in 24 hours from your template.
- A custom Data Processing Agreement covering the marketing-site contact form and the pilot engagement separately.
- Answers to your vendor-security questionnaire at the level of detail appropriate under NDA.
- Direct access to the engineer who owns the code.
Reporting a security issue
Email [email protected] with subject prefix SECURITY. Please don't open a public GitHub issue for vulnerabilities. We acknowledge within 48 hours.