AI-powered iMessage response generation using fine-tuned models and agentic workflows.

Demo Video Link: https://screen.studio/share/72wVyGXE?state=uploading

Startup Page

Inbox View

Tab Mode

Agent Mode

• Cmd + I - Navigate from start page to unread messages inbox view
• Enter - Select a conversation and enter tab mode
• Cmd + P - Switch between tab and agent modes
• Esc - Return to start page from tab or agent mode

pnpm=9.15.4 other versions might not work

textreme/
├── apps/
│   └── overlay/          # electron overlay app
├── ml/
│   ├── data/             # iMessage extraction scripts
│   └── training/         # LLaMA 3.1 8B fine-tuning (Modal + Axolotl)
├── packages/
│   ├── client/           # shared client library
│   └── schema/           # shared type definitions
└── notes/                # development notes

• Open System Settings
• Open Privacy & Security -> Full Disk Access
• Allow for whichever app in which you run pnpm dev (typically your IDE or terminal)

# install dependencies
pnpm install

# run the overlay app
pnpm dev

# get contacts info to the app (compile Swift binary for your macOS version)
cd apps/overlay && swiftc -o contacts_dump contacts_dump.swift

Fine-tune LLaMA 3.1 8B on your text messages to predict responses.

cd ml

# 1. Install: uv sync && uv run modal setup
# 2. Set up Modal secrets (HuggingFace + W&B)
# 3. Prepare data: cd data && uv run python prepare_data.py
# 4. Train: uv run modal run -m ml.training.train --data=data/training_data.jsonl

See ml/README.md for the complete guide.

cd ml/data
python extract_recent_messages.py

This extracts your recent conversations to ml/data/recent_conversations/.

cd ml

# prepare training data (convert conversations to JSONL)
cd data && uv run python prepare_data.py && cd ..

# start training on Modal
uv run modal run -m ml.training.train --data=data/training_data.jsonl

Monitor training at https://wandb.ai/

• ML Training: ml/README.md - Training setup and usage
• Data Extraction: ml/data/README.md - iMessage extraction guide
• Notes: notes/ - Development notes and explorations