Recall

Inspiration

We built Recall for people like us — people who struggle with decision paralysis and bad memory. You know the feeling: a friend recommended a great restaurant in a group chat two weeks ago, but now you can't find it. Or someone shared three options for a weekend trip and you saved none of them. You scroll endlessly, can't remember who said what, and end up just asking again (or giving up). Our chats are full of decisions, plans, and ideas — but they disappear into the scroll. Recall exists to catch all of that before it's lost, so you never have to wonder "wait, what did we decide?" again.

What it does

Recall is an AI-powered memory platform that connects to your messaging apps and turns conversations into structured, searchable knowledge. Send "recall" in any iMessage conversation, and our agent summarizes what you missed, extracts action items, and resurfaces things you mentioned before. Through the web dashboard, you can also import from iMessage, WeChat, WhatsApp, 小红书, TikTok Saved, and more. MiniMax automatically classifies every memory into categories like Food, Events, Travel, Sports, and Ideas — so you can always find what matters.

How we built it

We used TRAE AI as our development environment throughout the entire build. The iMessage agent runs locally on macOS using Photon's imessage-kit SDK, which watches for trigger messages and reads chat history. The AI backbone is MiniMax M2.5 — we use its text generation API for summarization, action item extraction, memory recall, and automatic category classification of ingested content. The web app is built with React and deployed on Vercel, with Google OAuth for authentication.

Challenges we ran into

Getting Full Disk Access and the Photon iMessage SDK to work reliably with real-time message watching was tricky — debugging AppleScript-level issues on macOS required patience. Designing prompts for MiniMax that produce clean, scannable summaries (not walls of text) took multiple iterations. Building the multi-source ingestion flow (iMessage, WeChat, WhatsApp, etc.) in 24 hours meant making hard prioritization calls.

What we learned

Vibe coding with TRAE AI genuinely accelerated our workflow — we shipped features we wouldn't have attempted manually in 24 hours. We also learned that the real challenge with a "memory" product isn't the AI — it's the ingestion layer. Getting data out of messaging platforms cleanly is the hard part; once it's structured, MiniMax handles the intelligence beautifully.

What's next for Recall

Live iMessage agent for group chats (not just DMs), Notion export, a mobile app, and deeper cross-platform memory linking — imagine Recall connecting a restaurant your friend mentioned in iMessage with the TikTok reel you saved about it.