Last updated: 21 August 2026
llmfa.st operates the inference API at api.llmfa.st and the
website at llmfa.st.
For anything about privacy — a question, a request about your data, or a complaint — write to [email protected]. A person reads it, and we answer within 30 days.
This policy covers data we receive when you use our inference API — whether you call us directly or a platform such as OpenRouter routes your request to us — and when you visit our website.
For prompts and completions we act as a processor: you decide what to send and why, and we process it on your instructions. For account and billing data we act as a controller.
Where OpenRouter routes traffic to us, our customer is OpenRouter. We hold no account for their end users and generally cannot identify them.
We operate zero data retention for inference content.
To speed up requests that share a common beginning, the inference engine keeps a key–value cache in GPU memory. This holds derived numerical state, not readable text; it is never written to persistent storage, is continuously evicted under normal load, and is destroyed when the process restarts.
We publish compliance.zdr: true for every model on our
models endpoint.
Whichever option above applies, we keep limited metadata about each request. It contains no prompt or completion content.
| What | Why |
|---|---|
| Timestamp | Billing, capacity planning |
| Model requested | Billing, routing |
| Token counts (prompt, completion, cached) | Billing |
| Latency — time to first token, total duration | Reliability monitoring |
| HTTP status and error class | Reliability monitoring |
| API key identifier | Attribution, abuse prevention |
Per-request metadata is kept for 30 days and then deleted. Aggregated statistics that cannot be traced to an individual request are kept indefinitely for capacity planning.
For direct customers we hold a contact name, email address, company name and billing details. Card data is handled by our payment provider and never reaches our servers. We keep billing records for as long as the tax and accounting law that applies to us requires, which is typically seven years.
API keys are stored only as a SHA-256 hash. We cannot display a key again after issuing it, and a lost key must be replaced rather than recovered.
llmfa.st sets no cookies and runs no analytics. There is no tracking script on this site, no advertising pixel and no third-party embed. Nothing follows you from page to page, which is also why you are not being asked to dismiss a cookie banner.
Our web host keeps standard access logs — IP address, timestamp, requested path and user agent — for a short period for security and abuse prevention, and we do not use them for anything else.
Where GDPR or similar law applies, we rely on:
We publish the physical serving location of every model in the
datacenters field of our
models endpoint, as a country
code and region. That endpoint is generated from our live configuration, so
it is accurate at the moment you read it — which a sentence on this page
could not promise.
If you need inference confined to a particular region, or a written commitment about where your data is processed, write to [email protected] before you start sending traffic.
We rely on a small number of third parties to run the service: a provider of the GPU hardware inference runs on, a payment provider for billing, and a provider for transactional email. None of them receives prompt or completion content, because we do not retain it to pass on.
A current list, naming each provider and what it handles, is available on request from [email protected], and we will give notice before adding a sub-processor that handles customer data.
We do not sell your data. We disclose it only:
We use measures appropriate to the risk: TLS 1.2 or better in transit, encryption at rest for anything we persist, hashed credential storage, network isolation of inference hardware from the public internet, least-privilege access control, and audit logging of administrative action.
No system is perfectly secure. If a breach affects your data we will notify you without undue delay and within any statutory deadline that applies.
| Data | Kept for |
|---|---|
| Prompts and completions | Not retained |
| Per-request metadata | 30 days |
| Aggregated statistics | Indefinitely (not linkable to a request) |
| Account data | Life of the account, then 90 days |
| Billing records | 7 years (legal requirement) |
| Web server access logs | 14 days |
Depending on where you live, you may have the right to access, correct, delete, port or restrict processing of your personal data, to object to it, and to complain to a supervisory authority — in the EU, your national data protection authority.
One consequence of section 3.1 is worth stating plainly: if we do not retain your prompts, we cannot produce them in response to an access request. There is nothing to produce. This is a feature of the design, not a refusal.
Write to [email protected]. We respond within 30 days.
The Services are not directed to anyone under 18 and we do not knowingly collect their data. If you believe a child has given us personal data, contact us and we will delete it.
We will post changes here with a new date. Where a change materially affects how we handle prompt content, we will email active customers at least 30 days before it takes effect.
llmfa.st
Privacy requests: [email protected]
Security reports: [email protected]
General enquiries: [email protected]