Steward is a personal assistant that lives in your Mac menu bar and quietly runs your WhatsApp and email for you. It reads everything that comes in, deals with the noise on its own, drafts replies in your voice, and only taps you on the shoulder for the things that actually need you.
It runs on your own computer — your messages and keys never leave your Mac.
The golden rule: Steward never sends a message without your one tap of approval, and it never quietly handles something important. You're always in control.
Steward catches what matters, drafts a reply in your voice, and waits for one tap. ▶ Watch the full 60-second demo
- 📥 Reads your WhatsApp + email, 24/7 — every message, so it always has the full picture.
- 🧹 Kills the noise — newsletters, spam, promotions get filed away quietly. You're not bothered.
- ✍️ Drafts replies in your voice — you just read it and tap Send. You never type.
- ✋ Never sends on its own — every reply needs your approval tap. Always.
- 🛑 Won't auto-handle important stuff — money, investors, big decisions → it flags them and waits for you. It will never answer something that matters on your behalf.
- 📱 Knows what you do elsewhere — reply to someone on your phone yourself and it notices, and clears it from your list. No double replies.
- 👀 Tracks delivery + read — it can tell you "they read your message 2 hours ago, no reply yet."
- 🧠 Learns who matters and remembers what you promised people (commitments).
- 📊 Shows you the picture — what it handled, time it saved you, today.
See FEATURES.md for the full list, still in plain language.
You'll need: a Mac, and one API key from any AI provider (OpenRouter is easiest — one key, all models). Email and WhatsApp are optional and can be added later.
The friendliest path: open this folder in Claude Code (or Cursor) and say "set this up for me following AGENTS.md." Your AI will do all the technical steps and then hand you a short checklist of the few things only you can do (get a key, make a Telegram bot, scan a QR). See SETUP.md for the exact, beginner-friendly walkthrough.
Prefer to do it yourself? One script gets the basics running:
git clone https://github.com/jeeevan11/steward-app.git steward && cd steward
bash bootstrap.sh # installs everything, creates your .env, runs the testsThen open SETUP.md and do the handful of "your turn" steps it lists.
Steward works with any OpenAI-compatible provider — OpenRouter, OpenAI, Together, Groq, or
a local model. You pick one and paste one key in .env. Defaults to OpenRouter (one key →
every model). Full details in .env.example under "LLM — BRING ANY KEY."
- It starts in dry-run mode — it drafts and shows you everything but changes/sends nothing
until you decide to switch
MODE=live. - Your secrets (
.env), your data (data/), and your sign-ins (secrets/) are never committed to git and never leave your machine.
assistant/— the Python "brain" (reads messages, classifies, drafts, decides).relay/— a small Node app that connects to WhatsApp.mac/— the SwiftUI menu-bar app (the console you see).assistant/web/— an optional web dashboard.
Architecture notes live in CONTEXT.md; the setup playbook for an AI assistant is AGENTS.md; the house rules are in CLAUDE.md.




