# Privacy Policy

> How AnyRouter collects, uses, and retains data when you route LLM traffic through anyrouter.dev.

**Web page**: https://anyrouter.dev/privacy
**Last updated**: 2026-11-18

AnyRouter ("AnyRouter," "we," "us") runs an API gateway for large language models at [anyrouter.dev](https://anyrouter.dev). This policy explains what data we collect, why, how long we keep it, who receives it, and the controls you have. It covers the API, dashboard, documentation site, MCP server, CLI, and plugins we publish. Your use of AnyRouter is also governed by our [Terms of Service](/terms).

## Summary

- Request and response bodies **are logged by default** so you can debug your traffic. You can turn this off with the Body logging switch in Privacy settings.
- We always log request metadata (model, provider, tokens, cost, latency, status, IP address, user agent) to run, bill, secure, and debug the service.
- The upstream provider that serves a request receives your full prompt. Some providers, including some free ones, may keep or train on prompts. Zero-data-retention (ZDR) routing, available on Pro and higher plans, keeps your requests off them.
- We never sell your data, and we never train our own models on your prompts.

## Our commitment on data use

- We keep only what we need to run, bill, secure, and debug the service, and to show you your own logs.
- We never sell personal data, prompts, or outputs.
- We never use your prompts or outputs to train our own models.
- We never use your data for advertising, for profiling you, or for any purpose unrelated to running the service.
- You can ask us to delete your data, as described in [Your rights](#your-rights).

## Data we collect

### Account data

When you sign in, our authentication provider (Clerk) gives us your email address, a stable user ID, and any profile fields you choose to share (name, avatar). We use them to tie keys, settings, and usage to your account. Clerk handles passwords; we never see them.

### API keys and provider keys

We store your AnyRouter inference keys (`sk-ar-`) and management keys (`ak_`). Provider keys you add for bring-your-own-key (BYOK) or donate to the shared key pool are encrypted at rest and decrypted only to forward a request, to check that the key works, or, for a donated key, to compute its fingerprint. A BYOK key is used only for your own requests unless you choose to donate it to the shared key pool.

### Donated keys

If you donate a provider key to the shared key pool, we also store:

- the encrypted key, as for any BYOK key, and a short masked preview of it;
- a keyed one-way fingerprint of the key, so that the same key cannot be donated from two accounts. The fingerprint cannot be turned back into the key;
- the version of the donation terms you accepted, with the time and IP address of that acceptance;
- the models you allowed, any prices you set, the key's health checks, and request and usage counts for the key.

The public pool page may show a masked form of your username or name as a donor, such as "duy•••t". We never show donors other users' prompts, outputs, or identity. When you withdraw a donation, the pool stops using the key, and the donation record and the fingerprint may stay until you delete the key. Deleting the key from your BYOK keys removes the encrypted key and its donation record, including the fingerprint. An audit log entry of the donation, with its time and IP address, may remain. The [Terms](/terms#donations) set out the rules for donors.

### Request metadata

For every request we record:

- timestamp, the API key used, your account, and the model you asked for;
- the provider and upstream model that served it, and whether a BYOK or pooled key was used;
- token counts, cost, latency, time to first token, status, and any error message;
- IP address, user agent, referrer, and app name or version headers your client sends.

### Request and response bodies

Body logging is **on by default** on every plan. When it is on, we store the full request and response (messages, tool calls, and attachments) so you can inspect them in Request Logs. You can turn it off with the **Body logging** switch on the [Privacy settings](https://anyrouter.dev/docs/features/privacy-controls) page. Turning it off stops new bodies from being stored; it does not remove bodies already stored. A PII redaction option is rolling out and may not yet apply to your traffic, so do not rely on it to remove sensitive data. A request made with ZDR on (below) is never written to our body archive, whatever the Body logging switch says. Separately, Cloudflare AI Gateway, which carries requests to providers, records requests and responses for its own short log window whether or not Body logging is on.

### What upstream providers receive

We forward your request to the provider chosen by your routing configuration. That provider receives the full prompt and any attachments, and its own terms and privacy policy apply (see [Third-party models](/terms#third-party-models)). Some providers keep prompts or train on them, especially on free offerings. We show a "Trains on prompts" badge where we know of it, but that list may be incomplete.

**ZDR routing.** With "Always enforce ZDR" on (Pro, Pro+, Max, and Enterprise plans), we route only to providers that we have recorded as offering zero data retention. If no such provider serves the model, the request fails instead of falling back. ZDR is off by default. It relies on each provider's published commitments, which we cannot independently verify. With ZDR off, your request may reach any provider that serves the model, including ones that keep or train on prompts.

**Shared key pool.** A request served by a donated key goes to that provider under the donor's account, so the donor's provider account may record it, and the donor may be able to see its content in that provider's console. We never show donors your prompts, outputs, or identity, and our Terms forbid donors from inspecting, copying, or using content that passes through their key. We cannot enforce that inside a provider's own console. Do not send secrets, personal data, or confidential material through free models or the shared pool; use your own key or a paid model with ZDR on for that.

**Local relay.** A request served by a relay device goes to that device's owner's machine and model. The owner's device sees the request. The same rules and the same advice apply as for the shared pool.

### Cookies and analytics

We use session cookies for sign-in (Clerk) and first-party storage for UI preferences. We use PostHog for product analytics, which links usage events in the site and dashboard to your account when you are signed in. We do not use advertising trackers.

## How we use data

- **Run the service.** Authenticate requests, route them, return responses, enforce rate limits, and bill usage.
- **Show you your logs.** Request Logs, usage, and cost views.
- **Debug and improve reliability.** Find errors and slow or failing providers, and tune routing.
- **Security and abuse.** Detect abuse, fraud, stolen or leaked credentials, and policy violations, and investigate [abuse reports](/report-abuse). When a key, account, or payment looks stolen, leaked, or abused, we may share the key identifier, account details, and the related request records with the affected provider, our payment processor, or law enforcement, and we may keep those records past the normal retention periods for the length of the investigation.
- **Communication.** Send account, billing, key, and incident email.

## Sub-processors and service providers

These companies process data for us to run the Service. Each is bound by its own terms and privacy policy. We update this list when it changes, and business customers can ask to be told about changes by emailing [privacy@anyrouter.dev](mailto:privacy@anyrouter.dev).

| Provider | What it does for us | Data it handles |
| --- | --- | --- |
| Cloudflare | Hosting (Workers, D1, KV, R2), AI Gateway, data lake storage | All request data, account data, and keys we store (provider keys are encrypted) |
| Clerk | Sign-in and sessions | Email, name, sign-in identifiers |
| Polar (merchant of record) | Checkout, subscriptions, credit top-ups, tax, invoices | Billing details and purchase records. We never see card numbers. |
| PostHog | Product analytics | Usage events from the site and dashboard, linked to your account when signed in |
| Resend | Transactional email | Your email address and the content of emails we send you |
| ClickHouse and Postgres that we run ourselves | Analytics and archive copies of request metadata and logs | Request metadata and logs |
| Upstream model providers (OpenAI, Anthropic, Google, xAI, Z.ai, DeepInfra, Together, Groq, Mistral, and others) | Run the model that answers your request | The full prompt and attachments of requests routed to them |

**Models differ.** Each model page lists which providers can serve it and what we know about their data practices, including whether they keep prompts or train on them. See [Model and provider data practices](/terms#data-practices) in the Terms.

## Retention

We keep data for these periods today:

- **Account, key, billing, and settings records**: for the life of your account. Billing records may be kept longer where tax or accounting law requires.
- **Request metadata in our primary database**: about 30 days, then pruned.
- **Archived request and response bodies** (in Cloudflare R2 and ClickHouse): kept so Request Logs and support stay complete. We do not delete them automatically today.
- **Request and response bodies in our analytics lake** (Cloudflare R2 Data Catalog): kept so history and analytics stay complete. We do not delete them automatically today. If that changes, we will update this page first.
- **Archive and analytics copies of request metadata** (in ClickHouse, Postgres, and Cloudflare R2): currently kept with no fixed deletion date. We will update this page when a retention period applies.
- **Cloudflare AI Gateway logs**: kept for the gateway's log window, then removed by Cloudflare.

What you can see in the dashboard and API is separate: read-access is clamped to your plan window. Older rows are not returned, even where we still hold them.

| Plan | Access window |
| --- | --- |
| Free, Go | 7 days |
| Pro, Max | 30 days |
| Enterprise | 90 days |

## Your rights

Depending on where you live, you may have the right to access, correct, export, or delete your personal data, and to object to or restrict some processing.

- **Export**: download your account data from Settings.
- **Correct**: update your profile on the Account page.
- **Close your account**: delete it from Settings. This signs you out, removes your sign-in, replaces your email with a placeholder, and deactivates your keys. We then erase the request logs and bodies we hold for your account where we can. Copies inside shared archive and analytics files are kept until we can remove them individually, and archive and analytics copies of request metadata follow the retention section above. Billing and credit records are kept where tax or accounting law requires.
- **Erase your data**: email [privacy@anyrouter.dev](mailto:privacy@anyrouter.dev) and we will delete or anonymize your request logs and other personal data, except records we must keep by law (such as payment records).

We acknowledge requests within 2 working days and answer within the period the law that applies to you sets, and no later than 30 days.

## Security

Traffic is encrypted in transit with TLS. Provider keys are encrypted at rest. Access to production systems is restricted. No system is perfectly secure. To report a vulnerability, email [security@anyrouter.dev](mailto:security@anyrouter.dev). We do not take legal action against researchers who report issues in good faith.

## How we keep your data safe

- Traffic is encrypted in transit with TLS, and provider keys are encrypted at rest.
- Access to production systems and stored request data is limited to the people who run the service.
- We never sell your data and we never train models on your prompts or outputs.
- If a breach of personal data happens, we notify the affected users and the authorities that the law names within 72 hours of detecting it, or as soon as the law requires.
- You decide what is stored: turn Body logging off, turn ZDR routing on, or use your own key.

## International transfers

AnyRouter runs on Cloudflare's global network, and data may be processed wherever Cloudflare and our other service providers operate. Many upstream providers are in the United States, and some are in China or other countries. The provider that serves a request determines where it is processed.

## Children

AnyRouter is not for 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.

## Changes to this policy

We may change this policy at any time without notice. Changes apply when we post them, and we update the date at the top. If you keep using the Service after a change, you accept the updated policy.

## Contact

Data controller: ONENI CO LTD., Ho Chi Minh, Vietnam.

- Privacy questions: [privacy@anyrouter.dev](mailto:privacy@anyrouter.dev)
- Security disclosures: [security@anyrouter.dev](mailto:security@anyrouter.dev)
- General support: [support@anyrouter.dev](mailto:support@anyrouter.dev)
