Legal
Privacy Policy
Effective 13 September 2026. What Teraflock, Inc. collects, where your prompts actually go, and what we keep. Inference needs plaintext on the serving machine; this page says so instead of pretending.
1. What we collect, by surface
The website (teraflock.ai)
- Waitlist: your email address and which side you picked (build, operate, both). Your IP address is used in memory to rate-limit the form and is not stored with your signup.
- Nothing else. The site sets no cookies and runs no analytics or third-party trackers. Our web server keeps ordinary access logs (IP, path, user agent) for a short time to operate the site.
The console (app.teraflock.com)
- Sign-in identity: your email, and the name and avatar URL your provider (Google or GitHub) shares, or just your email for a sign-in link. We ask providers for the minimum scopes (OpenID email and profile; GitHub read-user and email).
- Session: one HttpOnly, Secure, SameSite cookie holding a random session token; the server stores a hash of it. No advertising or analytics cookies.
- Account data: your API keys (we store a hash and a display prefix), usage, balance, and billing history.
The API
- Per request we store metadata, never the text: the model, privacy tier, serving node, token counts, timing, the random seed, and the credits charged and paid. Prompts and completions stream through the gateway in memory and are not written to our database or logs.
- Batch API: the JSONL file you upload and the output and error files we generate are stored in our object storage so you can download them. There is no self-service delete yet; email us and we remove them. This is the one place customer text is at rest on Teraflock systems.
Payments
Stripe processes card payments and operator payouts. We store Stripe's identifiers, amounts, receipt links, and refund or dispute status. We never see or store card numbers, bank details, or the identity documents Stripe collects for payouts.
Node operators
- Enrollment and heartbeats: your node's public-key fingerprint, operating system and architecture, GPUs and their memory, RAM, CPU model and core count, free disk, measured bandwidth, daemon and runtime versions; then, every few seconds, capacity, queue depth, temperatures, and throughput. No hostname, no user name, no file listing.
- Connection metadata: the coordinator sees your node's IP address because the node connects to it. We do not geolocate or resell it.
- Trust records: verification verdicts (canary comparisons, fingerprint challenges, timing checks), reputation changes, and slashes, kept for 180 days so a dispute can be examined on the record.
- Earnings: a double-entry ledger of every credit earned, escrowed, vested, spent, forfeited, or paid out.
- Payout identity: your Stripe Express account ID and whether verification is complete. Stripe holds your identity and bank details.
2. Where your prompts go
A request is routed to a node according to the privacy tier you set on it:
open— any qualified node in the mesh, which is a third-party operator's computer. The plaintext must exist in that machine's memory to be processed, so the operator could in principle observe it. Operators agree not to read, log, or store requests, the daemon never writes them to disk, and a node caught doing so is banned. That is a contractual and technical promise, not a cryptographic one.verified— only nodes with high reputation, staked credits, and a signed operator agreement.private— only first-party hardware operated by Teraflock.
On every tier, the node receives the request content and an ephemeral request ID. It never receives your identity, account, API key, or IP address. A sample of requests is duplicated to a first-party reference node so the two outputs can be compared; that copy is processed the same way and not retained.
We do not use your prompts or outputs to train models, and we do not sell personal data.
3. Who else touches the data
- Amazon Web Services (United States): hosting, database, object storage, and transactional email through Amazon SES.
- Stripe: payments, payouts, and identity verification for operators.
- Google and GitHub: sign-in, when you choose them.
- Node operators: as described in section 2, for the tier you choose.
Model weights are downloaded by nodes from Teraflock's own distribution bucket; nodes do not contact the model publishers.
4. Email
We send transactional email only: sign-in links, invites when your waitlist spot opens, payout and billing notices, and notices about changes to these terms. The waitlist confirmation says it plainly: we write when the flock is ready for you, and not about anything else.
5. How long we keep it
- Account and identity data: while your account is open, then deleted on request except where the next line applies.
- Ledger, payment, and payout records: as long as accounting and tax law require, because money moved.
- Request metadata: for billing and abuse investigation; it contains no text.
- Batch files: until you ask us to delete them.
- Trust records: 180 days.
- Waitlist entries: until you are invited or ask to be removed.
- Database backups: seven days, then gone.
6. Your choices and rights
Email hello@teraflock.ai from your account address to see what we hold about you, correct it, export it, delete your account, or remove a waitlist entry. We answer within 30 days. Deleting your account removes identity and node data; financial records stay for the retention period above.
7. Security
Nodes reach the control plane only over mutually authenticated TLS and never listen on a public port. Session tokens and API keys are stored hashed. Card and bank data never reach our systems. If a breach affects you, we tell you.
8. Scope
The Service is offered in the United States, to adults, and data is processed in the United States. It is not directed at children under 18 and we do not knowingly collect their data.
9. Changes and contact
When this policy changes we update the date at the top and email account holders about changes that matter. Questions or requests: hello@teraflock.ai.