Privacy

Email and a password hash to sign you in. The agent configs you set up in the dashboard, including any provider API keys you paste — those are encrypted at rest. Your agent's conversation history and logs live inside your VM's /data overlay disk; we don't pull them out into a central store.

Provider keys (OpenAI, Anthropic, OpenRouter, etc.) and Telegram / Slack tokens are not stored in our database. They flow from the wizard straight into your VM's persistent disk (at /data/.env) and never enter our logs, audit trail, or backups in plaintext. The dashboard keeps a "…last4"-style preview so the Configure form can show which fields are set.

One exception, if you connected a Discord bot to your agent: our Discord relay holds an open gateway socket for that bot 24/7 — including while your VM is paused — so its Discord fields are stored encrypted (AES-256-GCM) and decrypted in memory only when the relay needs to forward an event.

Stored in your VM's persistent disk. Cagebox operators do not read them as a matter of course. We have admin access to the underlying host for support and incident response — we will not poke at your data unless you ask us to, and we'll tell you when we do.

So you can find a conversation while your agent sleeps, Cagebox stores each conversation's title (your agent writes it from the first message), channel, timestamps and message count; they are removed when you delete the agent. For the usage view we also keep per-day numbers (model, channel, tokens, estimated cost) for as long as your account exists. Message content stays on your agent's disk.

With Cagebox credits, your agent's model requests — your messages, the files it sends to the model, and the model's replies — pass through our model relay to the model provider (currently OpenAI). The relay keeps token and cost counts per key so we can measure your credits; it does not keep the content of requests or replies.

With your own key (OpenAI, Anthropic, OpenRouter, and so on), requests go straight from your agent's machine to your provider, under your key and on your terms with them. They don't pass through us.

We don't sell your data or hand it to anyone for advertising. These companies process data on our behalf, each only for the job listed:

• Hetzner — the servers in Germany that your agents, their computers and our database run on.
• Stripe — payments and invoices.
• Google — sign-in with Google; Google Analytics only with your consent (see below).
• Cloudflare — DNS and cookieless web analytics.
• Sentry — error reports from the web app.
• Zoho — our support email.
• OpenAI — the model provider behind Cagebox credits.

Billing is handled by Stripe. When you subscribe, your card details are entered on Stripe-hosted pages and never touch our servers — we only receive a Stripe customer id, the plan you picked, and the subscription status (active / trialing / past_due / canceled) or when your Day pass ends. PCI-DSS compliance is inherited from Stripe. Invoices and saved payment methods are managed through Stripe's customer portal, which you can open from the Billing page.

One strictly-necessary cookie keeps you signed in. For understanding how the site is used we run Cloudflare Web Analytics, which is cookieless and stores no personal data, and Google Analytics 4 in Google Consent Mode. Until you make a choice in the cookie banner — or if you select “Necessary only” — Google Analytics receives only anonymous, cookieless signals (no identifiers, used solely for aggregate visit counts). Only after you select “Accept all” do we set Google Analytics cookies that measure your session in full. We use no advertising pixels and share no data for ad targeting. You can change your choice any time by clearing site data.

Email support@cagebox.dev and we will wipe your account and your agent overlay disks within 7 days. You can also destroy individual agents from the dashboard at any time — that removes the overlay immediately.