Documentation on EmailOps - Open Source AI Email Client for Mac, Windows & Linux

Installation

Mon, 01 Jan 0001 00:00:00 +0000

System requirements

Running the AI locally is the part that needs more powerful hardware, and it is optional. You can decline it in the first-run wizard and run EmailOps as a plain email client, or keep every AI feature and route it to a remote provider instead. The two modes have very different requirements.

With local AI

One of the most important requirements for running local AI is the memory available to load the model and its context. Depending on your machine, that means one type of memory or another:

Getting started

Mon, 01 Jan 0001 00:00:00 +0000

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:

Standard Features

Mon, 01 Jan 0001 00:00:00 +0000

Everything on this page works with AI switched off. The AI layer is covered separately in AI features.

Accounts and sync

Connect as many mailboxes as you like — Gmail, Outlook / Microsoft 365 (Graph API), and any IMAP/SMTP server (iCloud, Yahoo, Fastmail, ProtonMail Bridge, self-hosted). Mail is synced into a local SQLite database, so reading and searching stay fast and work offline.

The name on each account is the sender name recipients see on the mail you send from it. Set it in the Sender name field of the account’s settings, or leave it empty to send with the address only. Gmail accounts start with the name from Gmail’s send-as setting, and IMAP accounts with the display name you gave when connecting them. Mail sent through Outlook carries the name Microsoft has for the mailbox.

AI features

Mon, 01 Jan 0001 00:00:00 +0000

Every AI feature below runs through whichever backend you selected, and each one can be turned off individually. With the default in-app backend, no prompt or email ever leaves your machine.

Choosing a backend

Settings → AI Backend & Models controls where inference happens:

In-app (local) — an embedded llama.cpp runtime. Nothing to install, no daemon, no network traffic. This is the default. It uses your GPU automatically where there is one — Metal on Apple Silicon, Vulkan on Windows and Linux — and the CPU where there is not. On a Mac it requires Apple Silicon (M1 or newer); on an Intel Mac it stays unavailable.
Ollama (local) — an Ollama server you already run at http://localhost:11434. Useful if you keep a shared model library. Note that on an Intel Mac it gets no GPU acceleration either, so it will be slow.
OpenRouter (remote) — a paid cloud API. Requires an API key, supports a monthly budget cap, and sends email content to a third party — so it stays off unless you enable it.

The model catalog

The in-app backend downloads models from a curated catalog, each pinned to a verified checksum:

Privacy & security

Mon, 01 Jan 0001 00:00:00 +0000

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:

Platform	Location
macOS	`~/Library/Application Support/com.emailops.app`
Windows	`%APPDATA%\com.emailops.app`
Linux	`~/.local/share/com.emailops.app`

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.

Command line (emailops-cli)

Mon, 01 Jan 0001 00:00:00 +0000

emailops-cli drives the same local engine as the desktop app — your mail, your accounts, your local AI — from a terminal. It reads the database the app already synced, so there is no separate setup and no second copy of your mail.

Currently macOS only.

Install

Download EmailOps-CLI-macos.dmg from the latest release, mount it and put the binary on your PATH:

hdiutil attach ~/Downloads/EmailOps-CLI-macos.dmg
cp /Volumes/EmailOps\ CLI/emailops-cli /usr/local/bin/emailops-cli
hdiutil detach /Volumes/EmailOps\ CLI

emailops-cli doctor # confirms it sees your data and accounts

The binary is universal (Apple Silicon + Intel), signed and notarized, so Gatekeeper lets it through without a prompt.

Troubleshooting

Mon, 01 Jan 0001 00:00:00 +0000

AI features are unavailable

With the in-app backend, check that the recommended model finished downloading in Settings → AI Backend & Models. An interrupted download leaves the model unusable — remove it and download it again.

If you switched to Ollama, make sure the daemon is running and reachable at http://localhost:11434, and that you have pulled a model:

ollama pull llama3.2
ollama pull nomic-embed-text

On an Intel Mac the in-app AI cannot run at all — it needs an Apple Silicon chip (M1 or newer), so EmailOps keeps it switched off. Use OpenRouter instead. Ollama will install, but it gets no GPU acceleration on Intel either, so expect it to be too slow to be pleasant.