Security & privacy
Your inbox is sacred.
An agent that acts on your email has to earn deep trust. Here’s exactly how we handle your data — plainly, specifically, no fear-mongering.
How the AI handles your data
Private by architecture, not by promise.
Zero-retention inference
Cloud model calls run with zero data-retention. Prompts and outputs aren’t stored by providers.
No training on your mail
Your email is never used to train models — ours or anyone’s. Full stop.
On-device option
Sensitive triage and drafting can run on a local model, so the content never leaves your device.
Bring your own key
Use your own Anthropic, OpenAI, or Google key. Keys are decrypted only in an isolated worker — never client-side, never logged.
Encryption & architecture
Crown jewels, treated like crown jewels.
Encrypted in transit & at rest
TLS everywhere; data encrypted at rest. Message bodies live in encrypted object storage, referenced by id.
Envelope-encrypted secrets
OAuth tokens and BYOK keys are envelope-encrypted via KMS — never stored inline, never written to logs.
Least-privilege access
Minimum OAuth scopes, object-level authorization on every read and write, signed webhooks, idempotency keys, strict CORS + CSP.
Untrusted-input handling
Email is treated as untrusted input to the agent: prompt-injection defenses, an action allowlist, and sanitized rendering that blocks tracking pixels.
You’re in control
Autonomy you can take back any time.
Copilot before Autopilot
In v1, a human approves before anything is sent. Autopilot is opt-in, bounded by rules you set.
Undo on everything
Every autonomous action is reversible. Changed your mind? One tap puts it back.
Full audit trail
Each action is logged with the reasoning behind it — the trust layer that makes autonomy usable daily.
Compliance
The paperwork, handled.
Found something? Responsible disclosure to support@aiemaily.com. We respond fast and credit researchers.
Email you can finally trust to act.
Start free. Add a card when you’re ready.