Trust

Trust & Compliance

GWMM LLC Pre-launch United States · US-West (Seattle)

Who runs Inferway, what touches your data, and how to reach us. One LLC, one node, one model — and full transparency about all three.

01 Operating entity

Inferway is an independent, single-operator inference service. One LLC, one node, one model — and full transparency about all three. The architect you email is also the person who chose the GPU, wired the tunnel, and ships the weights.

  • Legal name: GWMM LLC
  • Registered in: To be published
  • Registered address: To be published

Jurisdiction and address are populated when legal review publishes the Terms of Service. Until then, the placeholder above is the honest state of the world.

02 Data handling

The compressed version. The full architecture, retention model, and per-field handling is on the Privacy Policy page.

We never store your prompts or completions. Metadata only — token counts and latency, kept up to 90 days for billing and reliability. Never used for training, never sold.

Read the full Privacy Policy →

03 Subprocessors

These are the only third parties that touch a request, and only metadata in transit — never stored content. Every entry below comes from the company.subprocessors field in the single-source-of-truth config, so the table updates the moment a subprocessor is added or removed.

NameRoleData touchedLocation
OpenRouterAPI marketplace & request routingRequest metadata in transit (no stored content)United States
CloudflareTunnel & edge transportEncrypted traffic in transitGlobal edge
StripePayment processing for Inferway direct walletPayment method and transaction metadata (no prompt/completion content)United States
VercelWebsite hosting, analytics, and performance telemetryPage views, Core Web Vitals, and aggregated site performance metrics (no prompt/completion content)United States

Additional subprocessors — typically the GPU host (if not self-hosted) and a payment processor (if billing moves off OpenRouter-only) — will be added before launch.

04 Security

Report vulnerabilities to legal@inferway.ai. We aim to acknowledge within 48 hours.

Report abuse, including suspected CSAM or other illegal content, to legal@inferway.ai. We review reports as soon as operational capacity allows and cooperate with law enforcement as required by law.

Single-operator service — there is no 24/7 SOC. Reports are triaged by the operator directly. For high-severity issues, encrypt sensitive details with a key we publish on request.

Architectural guarantees that materially reduce blast radius: prompts and completions are processed in GPU memory and never written to disk; traffic is encrypted in transit through the Cloudflare tunnel; the inference node is not directly exposed to the public internet. See the Privacy Policy — Security section for the full list.

06 Service reality

Honest pre-launch disclosure. Single-node service, no multi-region failover, no hot standby, no uptime SLA during pre-launch. We think this is a feature — a one-person shop can't honestly promise what a 100-engineer org can — but it's also a constraint you should know.

  • Single inference node. Node-01 (RTX 5090 · US-West) is the only place your requests run.
  • Concurrency ceiling. 64 concurrent requests per API key. Excess load returns HTTP 429.
  • No SLA during pre-launch. The public incident ledger on /status opens the day service goes live — empty, by design, and ready to be appended to.
  • Planned downtime is announced on /status in advance.

When the service flips from pre-launch to live, the /status page renders the same components with real data: live TTFT, real request counts, and the public incident ledger. No second page is built for pre-launch — same component, two states, driven by a single config field.

07 Transparency roadmap

Inferway is built on the premise that you should be able to verify our claims. The roadmap below is our public commitment for how transparency deepens over time.

  • Live metrics on model cards — status and latency numbers are published on /model and /status.✅ Live
  • Public status page — latency curves, availability, and active backend visibility.🚧 In progress
  • Transparent operations — public postmortems, failover history, and incident retrospectives.Planned
Our stance. We publish everything about our own operations. We store nothing about your data. Prompts and completions are processed in GPU memory and discarded; only metadata needed for billing and reliability is retained.

See the raw measurements behind the numbers in the public stress-test reports.