Three documents, one page: what we do with your data (Privacy), the terms you accept by using the API (Terms), and the specific retention/handling of prompts and completions (Data Policy). Last updated .
Privacy Policy
This policy describes how Piko ("we", "the service") handles information when you use the
hosted API at agnosia.tech, the playground and the
account pages.
What we collect
- Account data — the email address you sign up with, and the account identifier derived from it. We do not ask for a name, an address or a phone number to use the API.
- Billing data — the credit balance and an append-only ledger of every movement (top-up, spend). Card details are never received or stored by us: payments are collected by Stripe on its hosted Checkout page, and we only learn whether a payment succeeded, its amount, and a payment identifier.
- API keys — we store a hash (SHA-256) of each key, its visible prefix, its name and its dates. The full secret is shown to you once and never retrievable.
- Request metadata — timestamps and token counts needed to meter usage. See the Data Policy for the content of requests.
How we use it
Account and billing data are used to authenticate you, enforce your credit balance, issue and revoke keys, and produce your ledger. We do not sell personal data, and we do not use your content to train models.
Who we share it with
- Stripe — payment processing (card data, billing address and, where tax applies, the country used to compute it).
- Supabase — authentication and the credit database.
- Inference engine — the self-hosted model that answers the readout. Your prompt content is relayed there to produce the answer; see the Data Policy.
Your rights
You may request access to, correction of, or deletion of your account data at any time by writing to [email protected]. Deleting your account removes your balance, keys and ledger rows.
Terms of Service
By calling the API or using the account pages you agree to these terms. If you do not agree, do not use the service.
The service
Piko is a hosted inference API: it reads the next-token distribution of a model over a closed menu of options and returns a choice, a probability and the possibility of abstention. It is not a general-purpose text generator, and it makes no guarantee of accuracy: a verdict is a probability, not a fact, and you are responsible for how you act on it.
Your account
- You must provide a valid email address and keep your API keys secret. You are responsible for activity under your keys and your balance.
- Requests are billed against your prepaid credit at the published rates; the balance must be positive for a request to be served. Credit is non-transferable and, except where required by law, non-refundable.
- We may rate-limit, throttle or refuse traffic to protect the service and its upstream providers, and may suspend an account that abuses the API or breaks the law.
Acceptable use
Do not use the service to break the law, to infringe others' rights, to attempt to extract model weights, or to resell access in a way that hides its origin. You retain responsibility for the content you submit.
Availability and liability
The service is provided "as is", without warranty. Upstream capacity is
shared and can be saturated; when it is, requests may be refused with a 429. To
the maximum extent permitted by law, we are not liable for indirect or consequential damages,
and our total liability is limited to the credit you spent in the month preceding the claim.
Changes
We may update these terms; material changes will be reflected by the date at the top of this page. Continued use after a change is acceptance of it.
Data Policy
This is the policy the provider application asks about: what happens to prompts and completions.
Logging
We do not log the content of your prompts, the media you attach, or the returned choice. The API is a stateless relay: a request is authenticated, its cost is recorded, it is forwarded to the model endpoint, and the answer is streamed back. What is persisted is metadata — a request identifier, a timestamp, the token counts and the amount charged — which is what lets a spend be traced and a retry be charged once.
Retention
- Prompt / completion content: not retained by us (kept only in memory for the duration of the request).
- Billing and usage metadata: retained for as long as your account exists, so your ledger stays auditable.
- Account data: retained until you delete your account.
Upstream processing
To produce an answer, your prompt (including any inline image, video or audio you attached) is relayed to the self-hosted inference engine that serves the model. That engine processes the request under our control; if your content is sensitive, review the provider's terms before sending it. We do not add prompts to any training set, and we do not sell them.
Security
Traffic is served over HTTPS. API keys are stored only as hashes. Payment card data never reaches our servers (Stripe Hosted Checkout handles it).
Contact
Privacy, legal and data questions: [email protected]. Support: [email protected].