Privacy & security

EmailOps is built around one rule: your mail stays on your machine. This page describes what that means concretely — where data is written, what network calls exist, and which safety features you can turn on.

Where your data is stored

Everything resides in your OS application data directory:

Inside it:

• A SQLite database — messages, threads, contacts, calendar events, classification tags, search embeddings and AI memory. This is the only copy EmailOps keeps.
• A models/ folder — the AI models you downloaded.

Point EMAILOPS_DATA_DIR somewhere else before launching to use a different location — a second profile, or an encrypted volume.

Credentials are not in there. OAuth tokens and IMAP passwords go to the system credential store: macOS Keychain, Windows Credential Manager, or a Secret Service keyring on Linux. They are never written to a config file, and they survive uninstalling the app.

There is no EmailOps server

There is no account to create, no sign-up, and no backend operated by us — so there is nowhere for your mail to be uploaded to, and nothing to breach. The app talks to exactly these hosts, all of which you can name:

The last row is the only path by which your mail can reach a third party, it is off by default, and it takes a deliberate change in Settings → AI Backend & Models plus your own API key to enable.

No telemetry

The app collects no usage analytics, sends no crash reports, and has no phone-home of any kind in released builds. There is no opt-out because there is nothing to opt out of. (The source tree contains an optional OpenTelemetry tracing feature for local development; it is compiled out of every release build.)

Local AI by default

The default backend runs models in-process via an embedded llama.cpp runtime. No daemon, no localhost server, no network socket — the model reads your email from the same process that already has it. Classification, drafts, embeddings, chat, task and memory extraction all run there.

Switching to Ollama keeps inference local too, just in a separate process on your machine. Only OpenRouter sends content off the device. See choosing a backend.

Protection from the mail itself

Email is an attack surface. The client-side defences:

• Remote content blocking — external images, tracking pixels and other remote resources are blocked until you allow them. A per-email banner lets you load them once, or you can trust a specific sender permanently. This is what stops senders learning when and how often you opened a message.
• Junk and bulk scoring — every message is scored locally for spam and unwanted bulk mail. Your “junk” / “not junk” corrections train it. Flagged mail is faded or hidden, never deleted or moved on the server unless you explicitly confirm.
• Impersonation warnings — an optional check that flags messages appearing to come from someone they do not. Off by default, because it is the one check that accuses a sender of fraud and it has the least evidence to go on.
• Sanitised rendering — message HTML is stripped of scripts, event handlers and embedded objects before it is displayed, on both sides of the app. Attachments are never opened on your behalf.

Locking the app

Set a main password in Settings → Privacy & Security and EmailOps stays locked on startup until you enter it. There is no recovery path — if you forget it, you reinstall against a fresh data directory and re-sync from your provider.

Be clear about what this does: it locks the application, it does not encrypt the database. Anyone with access to your unlocked user account and the data directory can read the SQLite file directly. If that is part of your threat model, use full-disk encryption — FileVault on macOS, BitLocker on Windows, LUKS on Linux — which is the right tool for it.

Auditing any of this

EmailOps is Apache-2.0 and developed in the open. The claims on this page are checkable against the source at github.com/emailops/emailops, and so is the network behaviour — run it behind a proxy or tcpdump and compare against the table above. If something does not match, please open an issue.