WhatsApp reader
A small script that reads my own WhatsApp history straight from my Mac, so my AI can learn who I actually talk to and how I write, with no third-party app.
- Why
- WhatsApp has no connector, and the unofficial ones work by pairing as an extra device, which breaks WhatsApp's terms and risks the account. The Mac app already keeps my chats in a file on disk. Reading my own file needs nothing else.
- How long
- One session on 20 Aug 2026
How I built it
- I found where the Mac app keeps chats: a database file inside WhatsApp's folder in ~/Library/Group Containers.
- I had Claude write a read-only script that opens it without locking it, so WhatsApp keeps running.
- It writes three files to a scratch folder: contacts with how much each side replies, my own messages, and style stats.
- The AI turns those into notes about the people I talk to and how I write, then deletes the extract.
- People are ranked by how evenly both sides reply, not by volume: about half and half is a real relationship; nearly one-sided is a broadcast.
- To copy it you need
- Claude, or the AI of your choice, with Claude Code and the WhatsApp Mac app. The gotcha: timestamps are counted from 2001, so add 978307200 seconds to get real dates.
Try it yourself
Copy this prompt and paste it into ChatGPT or Claude to try it yourself.
I'm looking at Shweta Kamble's WhatsApp reader on whatarewebuilding.ai: a read-only Python script that opens the WhatsApp Mac app's own chat database (no third-party app, no pairing), writes contacts ranked by how evenly both sides reply, plus style stats, to a scratch folder, and deletes the extract after the notes are written. Help me build the same for my Mac. Find the database, write the script read-only so it never locks or changes it, and show me the top 20 two-way contacts first. The gotchas Shweta hit: dates need 978307200 added, and pasted text pollutes the writing sample. Ask me for anything you need.
Tools I used for this project
