Memosyne

An AI assistant that lives in WhatsApp and keeps track of what you said you would do.

Built at: On my own
Role: Founder and sole engineer
Timeline: 2024 to present
Stack: Node.js, TypeScript, WhatsApp Cloud API, PostgreSQL, LLM APIs
Live: memosyne.co

Every task manager asks you to open it. That is the whole problem. The thought arrives while you are walking, or mid-conversation, or three seconds before you forget it, and opening an app is already too much friction. Memosyne takes the note where the thought happens: a WhatsApp message.

The problem

Note-taking tools compete on features inside their own interface, which is the wrong axis. The binding constraint is capture — whether the thought gets recorded at all. WhatsApp is already open and is where messages to yourself go, but a chat log is a terrible database: no structure, no dates, no way to ask what is due.

One inbox, three kinds of thing

The same plain sentence has to become a reminder, a list item or a fact to remember, and the assistant decides which without being told. “Remind me to call the dentist Thursday at 2:30” is a dated reminder; “add olive oil and parmesan to my grocery list” is two items on an existing list; “I parked on level 3, bay 118” is a fact to hand back when asked where you parked.

The Memosyne reminders list showing reminders with categories, one-time and recurring types, and due dates. — What the messages became: typed, dated, categorised, and recurring where that was implied.

The Memosyne lists screen showing a groceries list with its items alongside other lists. — Lists accumulate across messages instead of being rewritten each time.

A web view for the things chat is bad at

Chat is excellent for capture and poor for review, so everything is also browsable — what is due, what is outstanding, what has been completed.

The Memosyne dashboard showing message usage, pending reminders, active lists and completed tasks. — The same data, in the shape chat cannot show you.

Memosyne settings with the connected WhatsApp number, timezone and a daily briefing toggle. — Timezone is a setting because every relative date depends on it.

How it works

The interesting engineering is the gap between a loose human sentence and a structured record. A message has to be classified, its entities extracted, and its relative dates resolved against the sender’s timezone — and all of it correctable in conversation, because the extraction will sometimes be wrong and the only interface available for fixing it is a reply.