Frequently asked

A self-hosted workspace for Claude (Anthropic's LLM). It's a quieter, more private alternative to Claude.ai or ChatGPT: your conversations live in a database you control, your costs are visible per-message, and there are no third-party trackers on the page.

No. Aurelius uses Anthropic's API but is not affiliated with, endorsed by, or operated by Anthropic. "Claude" is Anthropic's trademark.

It's invite-only during the private beta so we can keep token costs predictable and ship slowly. Approval typically takes under 24 hours. You'll get an email the moment your admin sets your budget.

They're stored in our database (Turso) under your username. Only you and an authorized admin can read them. They are NOT sent to Anthropic for training, we use the Anthropic commercial API which explicitly excludes inputs from training data.

You can delete any chat at any time, or request full account deletion by emailing the address in the footer.

No. We don't have any training infrastructure. We don't share your chats with any party other than Anthropic (which uses your prompts only to generate the response, see their commercial terms).

In a Turso (libSQL) database. Hosting region is configurable; see the privacy policy for the current configuration. Backups are encrypted and retained for up to 30 days after deletion.

Yes. Email the address in the footer with "data export request", you'll receive a JSON file of your account + all conversations within 30 days. This is a GDPR Article 15 (Right of Access) and CCPA right.

No. There are zero analytics scripts on the site. No Google Analytics, no Mixpanel, no Plausible, no Posthog. No advertising pixels. No cross-site fingerprinting. The only third-party JavaScript on the page is for Markdown rendering, syntax highlighting, and (optionally) Google Sign-In.

You buy one-off packs of tokens. No subscription, no auto-renewal. Pricing tiers are shown on the buy page. Credits never expire.

All sales are final. Token packs are digital goods that are delivered and made available for use the moment payment goes through, so we do not offer refunds. Any non-waivable consumer rights in your jurisdiction still apply. Details on the Refund Policy page.

Payment is handled by Stripe. We never see your card number, Stripe processes it on their own servers and sends us a confirmation event when the payment succeeds. Apple Pay and Google Pay are supported on devices that have them.

Email the address in the footer with your Stripe receipt. Our webhook should grant credits within seconds of payment, but if a network blip prevents that, we will manually credit your account and verify the Stripe-side state.

By default, the one your admin assigned you when approving your account. You can also pick per-turn between Haiku (cheapest, fastest), Sonnet (balanced default), and Opus (most capable, most expensive) using the model switcher in the dashboard, but only up to the tier your admin allowed.

It's an LLM. Outputs are generated probabilistically and can be incorrect, biased, incomplete, or out of date. Do not rely on AI output for legal, medical, financial, or safety-critical decisions without independent professional verification.

Yes, images (PNG, JPEG, GIF, WebP) and text (plain, markdown, JSON, CSV). 4 MB per file, 5 files per message. Images are processed by Claude's vision pipeline; text files are quoted as context.

This is a real risk we take seriously. Aurelius defends against it in multiple ways: an anti-injection contract at the top of every system prompt; filenames are sanitized before reaching Claude; web_search results are explicitly marked as untrusted; the markdown renderer strips image tags so model output can't exfiltrate data via crafted URLs. See audits/2026-05-22-ai-safety.md in the repository for the full threat model.

You can export your conversations at any time. We commit to giving 30 days notice and a final data-export window before any planned shutdown. We maintain encrypted backups separate from the live database so even a catastrophic incident doesn't mean lost data.

Yes, /status.html. It's manual right now (we update it during incidents) and will become an automatic uptime feed as we mature.

If we suffer a personal-data breach, we will notify affected users within 72 hours per GDPR Article 33. Smaller security issues are tracked in our audits/ folder and remediated in public commits. Report security issues to the address in the footer with subject "security".

Not yet. Email the address in the footer if you're interested, we want to talk to you while we design it.