Keak has not authored or reviewed a subprocessor notice for Cheaper Inference. This page exists so the list of processors is factually correct while that work is outstanding. Until counsel replaces this text, treat it as engineering documentation and not as a notice, disclosure, or contractual commitment. The binding subprocessor terms are in DPA section 6.
The tables below were built by reading this application: its dependencies, its deployment configuration, and every outbound call it makes. A provider appears only where the code shows it. Each row also says how far along the integration is: in use today, committed and built but not yet running, or dependent on work that has not started. Those are three different answers to “will this provider see my data”, so the table keeps them apart.
Keak’s wider security and compliance information stays on the Keak Trust Center.
1 Platform infrastructure
These providers host, store, bill for, and support the service itself.
| Provider | Purpose | Data it receives | Status |
|---|---|---|---|
| Supabase | Managed Postgres database, authentication, and session cookies for the platform. | Account and workspace records, API-key metadata, wallet and usage ledgers, and session identifiers. No prompt or response bodies. | In use today |
| Vercel | Application hosting, serverless request execution, and the scheduled jobs that refresh prices and prune caches. | All traffic to the website and the API, including request metadata and, in transit, request bodies. | In use today |
| Stripe | Wallet top-ups, saved payment methods, auto-recharge, and payment webhooks. | Billing contact details, payment-method data entered in Stripe's own card form, amounts, and payment identifiers. Cheaper Inference never receives full card numbers. | In use today |
| Resend | Transactional email — currently team and workspace invitations. | The invited person's email address and the inviter's email address. | In use today |
| OpenAlex | Daily scan of published research that feeds the optimization-technique candidate queue. | Outbound search queries only. No customer account data and no request content. | In use today |
2 Model and inference providers
These providers serve inference requests. Everything a customer sends in a request reaches the provider selected to serve it, and the generated output comes back through the same route.
| Provider | Purpose | Data it receives | Status |
|---|---|---|---|
| Primary marketplace upstream (name pending publication approval) | The spare-capacity marketplace that serves most inference requests, and the source of the price and model feeds. | The request body, including prompts, images, and files, and the generated output. | In use today |
| Vercel AI Gateway | Alternative inference route, used when it prices below the marketplace. | The request body, including prompts, images, and files, and the generated output. | In use today |
| OpenRouter | Model routing across third-party inference endpoints. | The request body and the generated output. | Committed, not yet in use |
| DeepInfra | Hosted open-weight model inference. | The request body and the generated output. | Committed, not yet in use |
| AkashML | Hosted open-weight model inference. | The request body and the generated output. | Committed, not yet in use |
| EZModel | Hosted model inference. | The request body and the generated output. | Committed, not yet in use |
| Venice AI | Hosted open-weight model inference. | The request body and the generated output. | Committed, not yet in use |
| Weights & Biases Inference | Hosted model inference. | The request body and the generated output. | Committed, not yet in use |
| Novita | Hosted open-weight model inference. | The request body and the generated output. | Committed, not yet in use |
| Amazon Web Services (Bedrock) | Hosted model inference, reachable only through a seller's own credentials rather than an account-level integration. | Nothing today. If the seller marketplace ships, requests routed to a seller offering Bedrock capacity would reach it under that seller's credentials. | Not planned; depends on work that has not started |
3 Examined and excluded
These are recorded so an omission can be challenged rather than guessed at.
- Keak platform (platform.keak.com)
- Identity provider for “Sign in with Keak”. Operated by Keak AI, Inc., the same entity that provides Cheaper Inference, so it is an affiliate rather than a third-party subprocessor.
- Google Fonts
- Typefaces are fetched at build time and served from our own origin, so no visitor request reaches Google.
- GitHub and the npm registry
- Source hosting and continuous integration. No customer data reaches either.
- Slack
- Community links rendered in the interface. No data is sent to Slack.
4 Known gaps in this draft
- No legal entity names, registered addresses, or processing locations are given. The codebase does not evidence them, and this page does not guess. A published subprocessor notice needs all three for every entry.
- The committed providers were built in parallel with this page and have not shipped yet. Those rows must be re-checked against what actually reaches production.
- The primary marketplace upstream is described but not named, because a separate copy rule forbids naming that vendor on customer-facing pages. A subprocessor notice normally has to name it. Counsel has to resolve the conflict before this page is published.
- The change-notification mechanism promised by DPA section 6.4 — 30 days’ notice before a new subprocessor starts processing — has no subscription mechanism on this page yet.
5 Contact
Send subprocessor questions and objections to support@keak.com.
Keak AI, Inc.
651 North Broad Street
Middletown, Delaware 19709
United States