Subprocessors
This page explains in plain English how Flinzo works today. If anything here is unclear, write to support@flinzoai.com and we will answer.
These are the companies that process data on our behalf so Flinzo can work. Each one receives the narrowest set of data its job needs. None of them receives your footage unless the table says so, and none of them receives your whole video for AI analysis — a model sees a transcript, scene boundaries and a small set of sampled still frames.
Read the last column carefully, because it is the one we refuse to overstate. "No training" appears only where the provider's own terms make it true, or where we have exercised an opt-out and hold the confirmation. Everywhere else it says "Pending confirmation", which means exactly that: we have not yet obtained the evidence, so we are not making the claim.
1. Current subprocessors
| Provider | Purpose | Data categories received | Region | Model training |
|---|---|---|---|---|
| Vercel | Hosting for the web application: pages, API routes, authentication. Never receives or streams video bytes. | Identifiers, request metadata, IP address | United States — exact region pinned at provisioning | Not an AI provider and does not train on customer data. Data processing agreement: pending confirmation |
| Fly.io | Runs the background workers that do the actual editing: probing, transcoding, cutting, reframing, captioning, rendering. | Your footage and everything derived from it, during processing only | United States — exact region pinned at provisioning | Not an AI provider and does not train on customer data. Data processing agreement: pending confirmation |
| Neon | Managed PostgreSQL: accounts, projects, credit ledger, transcripts, edit plans, crop geometry. | Identifiers, commercial information, transcripts, edit plans, crop geometry and salience tracks | United States — exact region pinned at provisioning | Not an AI provider and does not train on customer data. Data processing agreement: pending confirmation |
| Cloudflare R2 | Private object storage for uploads, derived audio, sampled frames, overlays and rendered output. No public buckets; access only through short-lived signed links. | Your footage, derived audio, sampled frames, rendered output | United States — bucket location hint set at provisioning; pending confirmation | Not an AI provider and does not train on customer data. Data processing agreement: pending confirmation |
| ElevenLabs | Speech-to-text. Receives the extracted audio track, returns a transcript with word-level timings. | Extracted audio and the resulting transcript text | United States; pending confirmation | Pending confirmation. The account-level training opt-out must be switched on and the confirmation stored before the first real user file is sent. Default retention is enabled, deleted items can persist in backups for up to 30 days, and zero-retention mode is enterprise-only. |
| Google Gemini API | Generates the Edit Plan. Receives the transcript, scene boundaries and 20–40 sampled still frames — never the video file. | Transcript text, scene boundary timings, sampled still frames | United States; pending confirmation | No training on the paid tier: with an active billing account, Google states it does not use prompts or responses to improve its products. Prompts and output are still logged for about 55 days for abuse detection. Confirmation that our project is on the paid tier: pending confirmation. We will never send customer footage to the free tier, where Google does use submitted content. |
| Anthropic | Powers the conversational editor: turns your notes into a revision of the plan. Receives the plan and your message, not your footage. | Edit plan JSON, your chat messages | United States; pending confirmation | Anthropic documents that prompts and outputs are not retained by default and are not used for training. Content flagged by its trust-and-safety systems can be retained for up to two years. A contractual zero-data-retention arrangement: pending confirmation, because it is granted per organisation by their sales team, not self-serve. |
| Stripe | Payments, subscriptions, invoices and tax calculation. We never see or store your card number. | Identifiers, billing contact details, transaction and subscription records | United States | Does not train AI models on your footage; it never receives footage. Stripe processes payment data under its own terms and uses it for fraud prevention and its own services. |
| Resend | Transactional email: sign-in links, renewal and price-change notices, receipts, job-complete notices, DMCA notifications. | Email address, message content | United States; pending confirmation | Not an AI provider and does not train on customer data. Data processing agreement: pending confirmation |
| Sentry | Error and performance monitoring so we can find failures. Secrets, tokens and request bodies are redacted before anything leaves our servers. | Error events, stack traces, request metadata, account ID | United States — exact region pinned at provisioning | Not an AI provider and does not train on customer data. Data processing agreement: pending confirmation |
A fallback speech-to-text provider is planned for the case where the primary one is unavailable. It is not enabled, no audio has ever been sent to it, and it will appear in this table with its own training status before it is switched on.
We do not use any advertising network, analytics broker, session-replay tool or customer-data platform. That is why the privacy notice can say we do not sell or share personal information.
2. How this page changes
We update the version and date at the top whenever a provider is added, removed, or its role changes. Adding a subprocessor that receives your footage is a material change and account holders are told before it takes effect.