What it does

Me_Bot ingests your exported WhatsApp chat logs (or WeChat SQLite databases, if you’re on iOS), cleans the raw text, and trains a neural model to generate replies in your voice. The whole pipeline is a pair of Python scripts and two Jupyter notebooks — no API, no cloud service, just your laptop and your embarrassing chat history.

The interesting bit

The project treats your personal corpus as training data, which is either delightfully intimate or slightly unnerving depending on your stance on digital doppelgängers. The WeChat support is community-contributed and involves poking around iTunes backup files with iTools — a nice reminder that Chinese chat ecosystems live in entirely different technical worlds.

Key highlights

• Pure local pipeline: export, clean, train, generate — no third-party services
• Supports both WhatsApp text exports and WeChat SQLite databases (iOS only)
• TensorFlow-based model, though the README is vague on architecture details
• Two-step cleaning process with manual name substitution required
• Actively seeking collaborators — contact is a bare MIT email in the README

Caveats

• English-only; Chinese support is “under constructing” per the README’s phrasing
• WeChat extraction is iOS-specific and requires third-party backup tools
• No model architecture, performance numbers, or sample outputs shown
• README contains typos (“itnerested”) and dead/unclear image links

Verdict

Grab this if you want a weekend project that turns your chat history into a party trick, or if you’re curious about personal-data-as-training-corpus. Skip it if you need production-grade dialogue systems or care about the privacy implications of training models on intimate conversations without any stated data handling practices.