Open beta. Registered accounts may submit free browser scans of eligible public model revisions through the 7B class, within their account quota. Models above the 7B class are not scanned in this beta. Results remain research-stage. Payment collection and invoicing are disabled.

1. Architecture. Web/app plane (VPS, EU) is separated from the scan plane (GPU workers). HF-reference weights are fetched worker-side. The planned small-upload path goes directly from the customer computer to a dedicated local GPU scan node's transient intake; the planned large-upload path goes directly to one ephemeral GPU pod. Neither customer-upload path sends model bytes through or stores them on the web/VPS control plane. Both remain disabled pending independent intake, isolation and verified-purge review.
2. Ingestion modes. Public HF-reference scans of eligible revisions are the only open path in this beta. Private HF-token and customer-upload modes are hard-gated off and require a separate isolation, egress, secret-handling and deletion review before any future release.
3. safetensors-only. Pickle-based checkpoints (.bin/.pt/.pth/.ckpt) are refused — they are an arbitrary-code-execution vector (CVE-documented). Repo code files are not even downloaded; trust_remote_code is never enabled.
4. Private-job routing. Private launch remains disabled. Enabling it requires an explicitly provisioned private-capable worker identity and job-scoped capability; request-body claims cannot elevate a worker.
5. Egress. The worker application is intended to contact bootstrap/package sources, the model registry and our result API. A platform-level outbound allowlist is not yet technically enforced or independently attested; private-model launch remains gated on that hardening work.
6. Operational records. Public-reference jobs receive an opaque signed public review receipt; their deterministic provenance stays owner-only. Any future approved customer-upload job requires source-pinned, signed records for its exact scan-node or pod route, execution, result and resource cleanup before a report can succeed. The Ed25519 signature authenticates Tetracta's trusted-worker operational record; it is not third-party or hardware-backed proof of execution, origin, or physical erasure.
7. Self-serve report deletion. You can remove any report from the live service from your dashboard (two-step confirmation). The report, metrics, images and job record become inaccessible immediately. Disaster-recovery backup retention, expiry and deletion-after-restore controls remain under review; no fixed expiry is promised in this pre-release.
8. Contractual commitments. No training on customer models; no third-party tracking or behavioral telemetry; no intentional report-external data egress.
9. Access control. Founder-only production access at this stage; least-privilege tokens; signed cookies; rate-limits; strict security headers.
10. Compliance roadmap. GDPR DPA + KVKK; independent pentest before enterprise onboarding; SOC 2 Type II on the roadmap (stated honestly — not yet certified).
11. Vulnerability reports. Email hello@tetracta.ai — 24h acknowledgement.