Use when you have a written implementation plan to execute in a separate session with review checkpoints
npx skills add DrBaher/youos
Or install specific skill: npx add-skill https://github.com/DrBaher/youos
# Description
>
# SKILL.md
name: youos
description: >
YouOS β your personal AI email copilot. Learns your writing style from your Gmail history
and drafts replies that sound like you. Improves automatically from your feedback.
Runs entirely locally on Apple Silicon. Use when: drafting email replies, reviewing
how you've handled similar situations before, or setting up a self-improving personal
communication assistant.
homepage: https://github.com/DrBaher/youos
version: 0.1.0
metadata:
openclaw:
emoji: "\u2709\uFE0F"
requires:
anyBins: ["gog", "python3"]
env: []
YouOS β Personal Email Copilot
YouOS drafts email replies in your style, grounded in your real past replies.
Trigger phrases
- "draft a reply to this email"
- "write this email for me"
- "how would I respond to this"
- "what would I say to"
- "help me reply"
- "draft in my style"
- "youos"
- "my email copilot"
- "generate a draft"
- "reply draft" / "email draft" / "draft reply"
- "compose reply"
- "write a response"
- "email response"
- "how do I usually reply to"
- "reply to this"
- "help me write"
- "write an email"
- "compose a response"
- "email assistant"
- "my writing style"
- "train on my emails"
Requirements
- Apple Silicon Mac (M1/M2/M3/M4) with 8GB+ RAM (16GB recommended)
- Python 3.11+
- gog CLI configured with your Gmail account(s)
- ~5GB free disk space
Quick start
# Install
cd ~/Projects/youos
pip install -e .
# Run setup wizard (15 min, mostly ingestion)
youos setup
# Draft a reply
youos draft "paste inbound email here"
youos draft --sender [email protected] "email text"
# Open web UI
youos ui
# Check status
youos status
# Run nightly pipeline manually
youos improve
# Add sender note
youos note [email protected] "integration partner, prefers bullet points"
# View stats
youos stats
# Teardown (remove all data, keep code)
youos teardown
How it works
- Ingests your sent Gmail history (stays local, never uploaded)
- Builds a retrieval index of your real past replies
- When you ask for a draft: retrieves the most similar past replies and generates a new one in your style
- Every email you review trains the model further
- Nightly: ingests new emails, fine-tunes the local Qwen model, runs autoresearch to optimize retrieval
Privacy
All data stays on your machine. No email content is ever sent to a cloud service (except the LLM call for initial drafts before the local model is trained). See PRIVACY.md.
# README.md
YouOS βοΈ
Your email. Your model. Your style.
YouOS is a local-first AI email copilot that learns from your sent Gmail history and drafts replies that sound like you β not a generic AI. It runs entirely on your Mac. No cloud. No subscriptions. Your data never leaves your machine.
Gmail (sent mail) Your feedback
β β
βΌ βΌ
Ingestion pipeline Review Queue
(gog CLI + SQLite) (10 emails/batch)
β β
βΌ βΌ
Reply Pairs DB βββββββΊ LoRA Fine-tuning
(FTS5 + BM25) (Qwen, nightly)
β β
βΌ βΌ
Draft Generation βββββ Autoresearch
(local Qwen MLX) (80 iterations/night)
β
βΌ
Draft Reply β
Privacy: Everything stays local. Your corpus, model, and drafts never leave your Mac.
π Landing page Β· πΉ Demo coming soon
What it does
- Ingests your sent Gmail history
- Learns your writing style through corpus analysis
- Drafts email replies grounded in your real past replies
- Improves from your feedback via LoRA fine-tuning
- Self-optimizes nightly via autoresearch
- Runs entirely locally on Apple Silicon
Requirements
- Apple Silicon Mac (M1/M2/M3/M4)
- 8GB+ RAM (16GB recommended)
- Python 3.11+
- gog CLI configured with your Gmail account(s)
- ~5GB free disk space
Quick start
# Clone and install
cd ~/Projects/youos
pip install -e .
# Run the setup wizard
youos setup
# Or run directly
python3 scripts/setup_wizard.py
The setup wizard walks you through:
1. Dependency check
2. Gmail account configuration
3. Email corpus ingestion
4. Writing style analysis
5. Optional initial fine-tuning
6. Server setup
Usage
# Draft a reply
youos draft "Hi, can we schedule a call next week to discuss the proposal?"
# Draft with sender context
youos draft --sender [email protected] "email text here"
# Open the web UI
youos ui
# Check system status
youos status
# View corpus stats
youos stats
# Add a sender note
youos note [email protected] "prefers bullet points, decision-maker"
# Run nightly pipeline manually
youos improve
# Start the web server
youos serve
Web UI
The web UI provides:
- Draft Reply: Paste an inbound email, generate a draft, edit it, and submit feedback
- Review Queue: Review auto-generated drafts against your real sent emails
- Stats Dashboard: Corpus health, model status, benchmark trends
- Gmail Bookmarklet: One-click drafting from Gmail
Architecture
app/
main.py # FastAPI application
api/ # HTTP endpoints
core/ # Config, embeddings, sender classification
db/ # SQLite bootstrap and migrations
generation/ # Draft generation (local Qwen + Claude fallback)
ingestion/ # Gmail, Google Docs, WhatsApp importers
retrieval/ # FTS5 + semantic search
evaluation/ # Benchmark scoring
autoresearch/ # Automated config optimization
scripts/ # CLI tools and pipeline scripts
configs/ # Persona, prompts, retrieval settings
templates/ # Web UI (feedback, stats, bookmarklet)
Privacy
All data stays on your machine. No email content is ever sent to a cloud service unless you explicitly use an external LLM for draft generation (configurable). See PRIVACY.md.
Configuration
All settings are in youos_config.yaml, created by the setup wizard:
user:
name: "Your Name"
emails: ["[email protected]", "[email protected]"]
model:
base: "Qwen/Qwen2.5-1.5B-Instruct"
fallback: "claude" # or "none" for fully local
autoresearch:
enabled: true
schedule: "0 1 * * *"
License
Open source. See LICENSE for details.
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.