This is the question everyone asks first, and it should be. You're handing an AI agent access to your support inbox, your billing system, and your customers' account data. Here's exactly what it touches, what it never touches, and how we keep it safe.
What the agent learns your product from. Stays in the agent's own prompt and index — never shared with anyone else. Full export if you ever cancel.
Needed for the thread context to answer coherently. Read in real time, not archived anywhere outside the tools you already use.
A restricted key: customer:read, invoice:read, subscription:read. No write access, ever. It cannot process a refund — the key physically doesn't allow it.
A read-only application role scoped to only the tables support needs. No inserts, updates, or deletes.
Only the integrations you explicitly connect. Nothing else on your computer, other tools, or personal accounts.
Responses are generated in real time via the Claude API under a Zero Data Retention agreement — nothing trains a model, and we don't archive your customer emails or messages.
Not now, not ever. We don't collect enough to sell even if we wanted to.
Full export of your knowledge base, system prompt, and configuration any time you ask, or automatically if you cancel.
We won't use your company as a case study or testimonial without your explicit written OK.
Happy to jump on a 15-minute call and walk through any of this in detail. No commitment, no pitch.