Nudge

My texts, calendar, and email all know different things about my life. None of them talk to each other. I kept dropping balls, forgetting I told someone I'd send something, showing up to meetings cold, missing follow-ups buried in group chats. I wanted something that already knew what I owed people before I had to remember.

What it does

Nudge is a two-part iMessage AI. The Listener silently reads all your conversations and extracts every task, commitment, and ask, cross-referenced with your calendar and Gmail. The Agent lives in a separate iMessage thread and sends you proactive voice briefings: what's urgent, what can wait, what you promised and to whom.

Imagine you're a busy student: you've got a job, 3 clubs, and friends to meet. Nudge lives in your messages and helps you keep track of everything. It'll start your day with a daily digest of your to-dos and schedule, then help you throughout the day. Whenever you have free time, it'll send you a message bumping an important to-do. When scheduling meetings, it'll help you find the most optimal time. Unlike notion or todoist, you don't need to make your own list, or tell it to remind you. Nudge does it all on its own, and lives inside of what you already use: iMessages.

How we built it

Built entirely in TRAE AI. MiniMax M2.7 for reasoning and extraction, Vision for photo understanding. Photon Basic iMessage Kit to read and send iMessages across a two-Mac listener bridge. Composio for Google Calendar and Gmail. SQLite for the task and entity graph.

Challenges we ran into

Getting the Listener to extract the right signal from casual texts was harder than expected. People don't say "I commit to sending you the deck by Friday," they say "yeah I'll get that to you." Prompt engineering for colloquial language took real iteration.

Accomplishments that we're proud of

Making something cute and something we don't have to build a complicated template for like notion. Love iMessages!

What we learned

Proactive is harder than reactive. Any chatbot can answer a question. The hard problem is knowing when to say something before you're asked, and making sure it doesn't feel like a nag when it does.

What's next for Nudge

Expanding the Listener to other messaging platforms. Building a memory layer so it gets better at predicting what matters to you specifically over time. Making the voice output feel more native, less notification, more assistant who actually knows you.

Built With

• minimax
• photon
• trae