Security

Last updated: 29 July 2026

Mnemor aggregates your mail, messages, calendar and code activity into one place. That concentration is exactly why its security model starts from a simple premise: keep the data on the device, and give you the controls.

Data at rest

• All app data is stored in the iOS app sandbox and written with complete file protection — the files are encrypted and remain unreadable while the device is locked.
• Credentials never touch the filesystem. OAuth tokens and any API key you supply are held in the iOS Keychain with device-only accessibility, so they are excluded from iCloud and iTunes backups and from the data export.
• Travel documents in the wallet are additionally excluded from device backup and gated behind biometric authentication.

Data in transit

• Every network call uses TLS. There are no plaintext endpoints.
• Connector traffic goes directly from your device to Google, Microsoft, Slack or GitHub. It is not proxied through our infrastructure.
• AI requests go directly from the device to the model provider.

Access control on the device

• App lock with Face ID, Touch ID or passcode, plus a configurable inactivity auto-lock.
• Jailbreak detection, which warns you — and can be set to refuse to run — on a device whose sandbox guarantees are compromised.
• Panic wipe erases every token, key, cached item and note from the device immediately.

AI controls

• Ranking is not AI. Deciding what needs you, who has waited longest and what is urgent is deterministic code executing on the device. No model, no network. AI is only invoked for optional extras you request — a draft, a summary, meeting notes — so the core of the product works with the cloud switched off.
• Data minimisation. When a model is called, it receives the specific excerpt for that task, not your mailbox and not your history.
• On-device first. Where the hardware supports it, Apple's on-device model handles requests and nothing leaves the phone.
• On-device only mode is a hard switch enforced at a single choke point in the code. With it on, no request reaches any cloud provider; features that need the cloud stand down rather than degrade silently.
• Images never reach a model. There is no image input path in the app at all — receipts, badges and documents are read with on-device text recognition. This is structural, not a setting.
• Bring your own key. Supply your own Anthropic API key and the app calls Anthropic directly with it; the key stays in your Keychain.
• Prompt-injection hardening is applied to every model call, so text arriving from an email or web page cannot redirect the assistant.
• Credential scrubbing. Before any text reaches a prompt — cloud or on-device — Mnemor strips things that should never be in one: API keys, bearer tokens, JWTs, private-key blocks, explicit password: lines, and card numbers that pass a Luhn check. This is hygiene, not anonymisation. It is deliberately tuned for precision so it cannot corrupt your draft, and it makes no attempt to remove names, companies or other personal detail. If you want a guarantee that nothing leaves the device, use On-device only — that is the control designed for it.

For the step-by-step version — every workflow in the app, each step labelled with whether it runs on rules, on-device, from cache or in the cloud, plus a complete list of every AI call the app can make — see how it works.

Auditability

The in-app Activity log records every AI request — when it ran, whether it stayed on-device or went to a named provider, and roughly how much text was sent — plus security-relevant actions such as replies sent, exports, key changes and disconnections. It never records message content. You can review the last 3, 7 or 30 days.

Your data rights, self-serve

• Export everything the app holds as a single JSON file. Credentials are deliberately excluded.
• Delete everything instantly with the panic wipe, or by removing the app. Because there is no server-side copy, that is a complete deletion.
• Revoke Mnemor's access at any time from the provider's own settings.

Sub-processors

Mnemor uses a deliberately short list of third parties, and only for the features described:

The services you connect — Google, Microsoft, Slack, GitHub — are your own accounts, accessed with your authorisation, not sub-processors acting for us.

This website

This site is static. It sets no cookies, runs no analytics, loads no third-party scripts or fonts, and contains no trackers — which is why you will not see a cookie banner here. It is served over HTTPS with HSTS and a strict Content Security Policy.

Reporting a vulnerability

Please report security issues privately to @saysereze on X rather than disclosing them publicly. We aim to acknowledge within two business days and will keep you updated through to a fix. We do not currently run a paid bounty programme, but we credit reporters who want it.

What we do not claim

Being straight with you: Mnemor is an independently built product. It does not currently hold SOC 2 or ISO 27001 certification, and there is no admin console, SSO or centrally managed deployment. If your organisation requires those, tell us what you need — but do not assume them.