Getting started
The first-run wizard: choose an AI backend, download a model and connect your first mailbox.
The first time you open EmailOps a four-step wizard runs. It takes a couple of minutes, most of which is a model download in the background.
1. AI on or off
EmailOps inspects your hardware and recommends whether to enable local AI. Pick:
- AI enabled — chat, drafts, classification and semantic search all run on this machine.
- Plain email client — no model is downloaded and no AI call is ever made. You can turn AI on later in Settings → AI Backend & Models, and turn it off again just as easily.
2. AI backend and model
If you enabled AI, choose where inference happens:
| Backend | What it means |
|---|---|
| In-app (local) | The default. A llama.cpp runtime embedded in EmailOps. No daemon, no setup, no network. |
| Ollama (local) | Uses your existing Ollama server at http://localhost:11434. |
| OpenRouter (remote) | Sends prompts to a paid cloud API. Opt-in, per feature, and off by default. |
With the in-app backend, pick a chat model from the built-in catalog. Qwen 3.5 4B is the recommended default: about 3 GB to download, needs roughly 8 GB of memory to run, and supports the tool-calling that chat relies on. Models too large for your system memory are greyed out. The download runs in the background — you can carry on with the wizard.
The memory that counts depends on the machine: unified memory on an Apple Silicon Mac, your GPU’s VRAM on a Windows or Linux box with a discrete card, and system RAM if there is no GPU. The model catalog lists the figure for each model.
The embedding model that powers semantic search (Nomic Embed Text v1.5, ~80 MB) ships inside the app on macOS, so there is nothing to download for search.
3. Inbox layout
Choose how the mailbox is laid out — split (list on the left, message on the right) or full width (one pane at a time). Change it whenever you like in Settings → Appearance, along with the interface language (English, Spanish, French, German).
4. Connect an account
The last step adds your first mailbox. EmailOps supports:
- Gmail — sign in through your browser and grant access. Tokens go straight into your OS keychain.
- Outlook / Microsoft 365 — same browser flow, via the Microsoft Graph API.
- IMAP / SMTP — iCloud, Yahoo, Fastmail, ProtonMail Bridge or any custom server. Enter the server details and credentials directly.
Add more accounts any time with Add account in the sidebar. With several connected you get a unified “All accounts” inbox on top of the per-account views.
After the wizard
The first sync takes a while
EmailOps downloads your mail into a local database, and the first pass has to fetch everything from scratch. How long that takes depends on the size of the mailbox — a few minutes for a small account, considerably longer for one with years of history and heavy attachments. It runs in the background, and the first messages appear within seconds — mail is downloaded in slices as the mailbox is listed, rather than after the whole history has been scanned — so you can read and search what has already arrived while the rest catches up.
This is a one-time cost. Every later sync is incremental: it asks your provider only for what changed since last time, so it finishes in seconds and runs quietly on a schedule. If AI is enabled, classification and embedding also work through the backlog on first run and then only touch new mail.
Once the first sync finishes:
- Classification starts tagging new mail by priority, intent and topic — see AI features.
- Embeddings are generated in the background so semantic search has something to search. You can watch progress and rebuild the index in Settings → AI Search.
- Consider setting a main password in Settings → Privacy & Security if you want the app locked on startup — see Privacy & security.
Both classification and embedding respect Limit AI processing to recent emails (Settings → AI Backend & Models), so a decade-old archive does not get processed unless you ask for it.