Your data. Your choice.
New accounts start with prompt and response sharing off. You can choose to enable sharing for a lower usage fee in your profile and turn it off there at any time.
When sharing is off
The public chat API and Playground process your messages in memory to route your request and return an answer. With sharing off, CyberRouter does not save prompt or response bodies to its database or application logs and does not use that content for product improvement or training.
We retain account and operational metadata: model and provider names, request IDs, timestamps, token counts, latency, status, routing decisions, credit transactions, and consent history. These support your Usage page, billing, account security, and troubleshooting. Turning off content sharing does not delete this metadata; it has a separate lifecycle and no automatic 30-day expiry.
Messages still go to the upstream services needed to fulfill your request. Client applications, third-party tools, and separately enabled storage features can retain their own copies. This guide covers the public chat gateway; it does not describe legacy engagement storage in custom deployments.
Security checks on account forms
We use Cloudflare Turnstile on registration, invitation, waitlist and account-email forms to reduce automated abuse. Cloudflare processes technical browser signals and your IP address for this security check. We validate the resulting short-lived token before processing the form. See Cloudflare’s privacy policy.
Provider keys and acceptance records
CyberRouter receives your provider key, stores it encrypted and decrypts it on the server to authenticate upstream requests. Management pages return masked metadata. Disconnecting stops new use through CyberRouter; retained encrypted records and backups are not immediately erased. Revoke the original key at its provider to invalidate it everywhere.
When you explicitly accept service terms, we keep their version, fingerprint and text, the acceptance time, and the associated account or connection reference. These acceptance records contain no provider secret. See the BYOK terms for responsibilities and incident reporting.
Website analytics
Basic analytics are on by default to help improve the site. They count views of predefined public pages, storing only daily page totals. They use no visitor or session identifiers, tracking cookies or browser storage for tracking, and do not retain individual events or connect visits across pages. Turn them off at any time using Analytics preferences. Previously declined choices and browser privacy signals keep them off.
Detailed analytics start only after you select Allow detailed. They measure page visits, approximate country, device category, referral category, selected link clicks, signup steps, scroll milestones and active reading time. We use aggregate reports to improve the site. Neither mode records sessions, keystrokes, form contents, prompts, provider keys, names or email addresses.
After you allow detailed analytics, Cloudflare Web Analytics measures page paths, referral information and loading performance. See Cloudflare’s privacy policy. Our collectors accept only predefined page and event categories; they exclude query strings and fragments. The edge uses your network address transiently for rate limiting and country detection, but does not store it in the analytics database. Basic counts do not retain country or device information.
Our engagement reports retain daily aggregates for up to 31 days, with cleanup running daily. Random tab-session and page identifiers are used for deduplication, are not linked to your account, and expire from the collector within 24 hours plus the daily cleanup interval. A session rolls over after 30 minutes of inactivity. Reading time pauses when the page is hidden or unfocused, and after 60 seconds without interaction. Cloudflare maintains its separate analytics retention.
Open Accessibility → Analytics preferences to choose basic counts, allow detailed analytics, or turn all website analytics off. We use Cloudflare’s approximate country detection to show an initial choice in the UK, EEA and Switzerland, or when location is unknown. Elsewhere, preferences remain available in Accessibility without an automatic popup. Detailed analytics require your permission everywhere. Your choice is remembered in this browser; permission for detailed analytics expires after 180 days, while a refusal stays in effect until you change it or clear browser storage. Do Not Track and Global Privacy Control disable both modes. This choice is separate from optional prompt sharing, and never changes your price or access to the service. Counts exclude visitors who decline, block tracking or exceed collection limits, so they are estimates. Basic views are not unique visitors, and must not be added to the overlapping detailed view count.
Optional Google sign-in
When available, Google sign-in uses your verified Google email to find your existing CyberRouter account. After you complete email-code verification, we save your Google account identifier and email to recognize the same account on later sign-ins. We do not request access to Gmail, Google Drive, or your contacts, and we do not retain Google access or refresh tokens. You can continue to sign in with your CyberRouter password.
Upstream providers have their own policies
No training, no prompt logging, and zero data retention are different commitments. Policies may depend on the model, hosting endpoint, account agreement, region, and API feature. Turning off CyberRouter sharing does not change your provider’s settings or guarantee end-to-end zero retention.
OpenRouter provides useful starting points: its data collection guide, provider logging reference, provider terms directory, and ZDR guide and endpoint reference. OpenRouter’s provider agreements can differ from the terms on your own direct BYOK account.
References reviewed September 9, 2026; OrcaRouter updated October 6, 2026. Provider terms change; follow the source links for current requirements.
OpenAI
API data is not used for model training unless you opt in. Default abuse-monitoring logs may contain prompts and responses for up to 30 days, with legal exceptions. Zero Data Retention requires eligibility and has endpoint and feature limits.
Anthropic
Retention depends on your agreement, model, and feature. The commercial policy describes deletion within 30 days with exceptions; ZDR is available for eligible use. Some models require 30-day retention even where other models qualify for ZDR. Review both the policy and model-specific requirements.
Google Gemini
Paid and unpaid API services have different data-use terms. Paid-service prompts and responses are not used to improve Google products, but may be logged for a limited period for abuse detection and legal requirements. Unpaid usage may be used for improvement and reviewed by humans; regional exceptions apply.
xAI
The API security guide describes 30-day storage for abuse auditing, without training unless you explicitly permit it. Team-level ZDR is available where supported, with limitations for features that require stored data.
DeepSeek
The Open Platform terms reference DeepSeek’s privacy policy for personal information. That policy covers collection of prompts and use for service and technology improvement. It does not provide a blanket zero-retention guarantee for every API request; confirm your specific API agreement.
Moonshot / Kimi
The platform privacy policy describes retaining information as needed for service, improvement, security, and legal purposes; its examples include input information while an account is active. Do not assume a fixed short retention window or zero retention without a specific agreement.
OpenRouter
OpenRouter’s own prompt logging is opt-in. The inference providers behind it have separate logging and training policies, and eligibility can differ between endpoints for the same model. Review your OpenRouter account settings and the provider selected for your request.
OrcaRouter
Its published policy says it does not retain prompts or outputs or use them for training, with a 13-month rolling retention period for usage metadata and 90 days for server/security logs. It allows longer retention for legal obligations and disputes. This describes its policy, not an independent verification. Check downstream provider terms and your agreement.
Change your mind at any time
- Open Profile → Privacy & data sharing.
- Review the consent, check or clear the sharing option, and save your preference.
- Check your effective fee on the Credits page and your spending on the Usage page.
Opting out stops new collection and deletes previously collected content from active storage. In-flight requests cannot save content after you revoke consent. Provider-held data and backup copies are separate: opting out does not erase upstream copies, and backup erasure is not immediate. Collected content in backups must not be restored for evaluation after revocation or expiry.