AI email assistant for managing your mailbox
An AI email assistant reads your mailbox so you don’t have to: it sorts incoming mail, summarises long threads, finds the message you half-remember, and drafts replies. Most of them do this by uploading your email to a provider’s servers. EmailOps uses a language model that runs on your own computer.
What an AI assistant does
Sorting. Every incoming message is tagged automatically by priority, intent and topic, so the inbox arrives organised instead of as one undifferentiated list.
Chatting with your inbox. Ask a question in plain language - “what did the supplier say about the delivery date?” - and get an answer drawn from across the whole mailbox, with the source messages cited under it. Follow-up questions keep the context.
Finding. Keyword search fails when you remember the substance but not the wording. Semantic search matches on meaning, so describing what you half-remember is enough to surface the message.
Drafting. A reply proposed from the thread’s context, which you edit rather than write from scratch.
The differences between products are mostly not in this list. They are in where the processing happens.
Local instead of cloud
To summarise your mail, an assistant has to read it. The question is where.
Most assistants - including those built into large mail providers - send the message to a server, frequently to a third-party model provider, and send the answer back. That is a reasonable engineering decision and part of why they work well. It also means the contents of your mailbox leave your machine, under a retention policy you cannot inspect.
EmailOps runs the model inside the application, on your own hardware, using your GPU where there is one. Mail is stored in a local database. There is no EmailOps account, no server to sign in to, and no telemetry. If you would rather use a remote model you can connect OpenRouter - but that is a setting you switch on, not the default.
The source is Apache-2.0 licensed and developed in the open, so none of this has to be taken on trust.
How EmailOps does it
- Chat with your mailbox - ask in plain language, get an answer with the source emails cited
- Classification - every incoming email tagged by priority, intent and topic; rules for the obvious cases, the model for the rest
- Semantic search - finds mail by meaning rather than exact wording
- AI drafts - replies grounded in the thread, in a persona and style you define
- Translation - of mail you receive in another language, and of mail you send
Every feature can be switched off individually, and the whole AI layer has one master switch.
Supported accounts
Gmail, Outlook and Microsoft 365, and any provider reachable over IMAP/SMTP. macOS, Windows and Linux.
Getting started
- Download EmailOps and add your account.
- It syncs your mail into a local database and downloads a model - the recommended one needs about 3 GB of disk and 8 GB of memory.
- Ask it something.
To run the AI model you need either a Mac with Apple Silicon or a discrete GPU on Windows and Linux. On a Mac the model size you can run is set by the system’s RAM, which the CPU and GPU share; on Windows and Linux it is set by the graphics card’s VRAM.
Free, no tiers, no locked features.