# Privacy Policy: Pure Light API

**Effective date:** 9 September 2026
**Last updated:** 14 September 2026
**Operator:** Pure Light Ltd (Company No. 17433578), 66 Paul St, London EC2A 4NA, United Kingdom
**Contact:** info@purelight.uk

This policy applies to the Pure Light API model gateway (`pure-light-api` Cloudflare Worker and its associated model endpoints).

## 1. What we process

To serve a chat completion request we necessarily process the content you send us:

- **Prompt content**: the messages in your request, used solely to generate a completion.
- **Request metadata**: timestamp, model requested, token counts, HTTP status, and aggregate operational event counts (never message content). Used for billing accuracy, abuse prevention, capacity planning, and uptime monitoring.
- **API key identity**: the bearer key you present, matched against our key store. Keys are stored as comparison data only; we do not store raw keys in logs.

## 2. What we do NOT do

- **No prompt logging.** Request and response bodies are not written to persistent storage by us.
- **No training on your data.** Your prompts and completions are never used to train, fine-tune, or evaluate any model, ours or anyone else's.
- **No selling or sharing of content.** We do not sell, rent, or share prompt/completion content with third parties for advertising or any other purpose.
- **No profiling.** No behavioural profiles, no ad tech, no cross-service tracking.

## 3. Sub-processors and infrastructure

Inference is normally executed by **Cloudflare Workers AI** (Cloudflare, Inc.). Your request transits and is processed on Cloudflare's global edge network under Cloudflare's own data policies:

- Workers AI inference results are **not used by Cloudflare to train models** (see Cloudflare's Workers AI documentation and Terms of Service for current wording).
- Cloudflare acts as a processor for the inference call; Pure Light Ltd remains the controller of the service.

If the primary inference path is unavailable, our gateway can replay the request on a serverless GPU failover operated by **Runpod, Inc.** in European Union data centres. The failover path applies the same rules: request and response content is not logged by us, not used for training, and not retained beyond the request; failover workers are ephemeral and discarded after serving. Runpod acts as a processor for that call.

We will update this section, and this policy, if our infrastructure stack changes.

## 4. Data retention

- Prompt/completion content: **not retained** beyond the duration of the inference request.
- Request metadata (timestamps, token counts, status codes): retained in aggregate form for up to 90 days for billing and abuse-prevention purposes.
- If you contact us for support, we retain correspondence as long as needed to resolve the issue.

## 5. Your rights

Under UK GDPR you have the right to access, rectify, erase, restrict, or object to processing of your personal data, and the right to complain to the ICO. Because we do not retain prompt/completion content, most requests will concern account metadata (API keys, billing records), which we can access and correct or delete on request.

Contact **info@purelight.uk** to exercise your rights. We respond within one month.

## 6. Security

- All traffic is TLS-encrypted in transit.
- API keys are stored as secrets in Cloudflare's secret store, not in application code.
- Access to production configuration is restricted to Pure Light Ltd directors.

## 7. Changes

Material changes to this policy will be posted at this URL with a new effective date. If a change materially reduces your protections, we will provide 30 days' notice to registered account contacts.

## 8. For OpenRouter users

If you access our models through OpenRouter, OpenRouter's privacy policy additionally governs their handling of your requests. OpenRouter may forward requests to us under terms compatible with this policy; we apply the same no-logging/no-training rules to all traffic regardless of entry point.