⚠️ This project has moved. The new and improved version is Zeno — cleaner architecture, better product, actively maintained.
The most advanced personal AI OS built on Apple Silicon. Voice + Memory + RAG + 5 Agents + Local LLM. No cloud. No subscriptions. No data leaving your machine. Built by one person in Haryana, India.
🎬 Demo video coming soon — follow to get notified
Every AI tool today requires internet, charges monthly, and sends your data to servers you don't control.
SAMEER AI TWIN runs entirely on your MacBook. Every conversation, every task, every memory — stays on your machine. Forever.
This is not a chatbot wrapper. This is a personal AI operating system with memory, knowledge, agents, and voice — built from scratch.
Speak naturally → Whisper transcribes → AI thinks → Kokoro speaks back. Full offline voice conversation. No internet. No API calls.
- Qwen2.5-7B for text responses (15 seconds)
- Qwen2.5-3B for voice responses (8 seconds)
- Decision engine with 4 modes: assistant, strategist, coach, analyst
- SQLite-backed task manager with deadlines
- Voice task listing — "what are my tasks?" reads directly from database
- Pattern detection — knows your most active hour and work preferences
- Logs every interaction — queries, tasks, voice sessions
- Builds daily activity summary automatically
- Detects behavioral patterns over time
- AI knows what you did today without being told
- Answers from your own notes, PDFs, and documents
- Fully local embeddings — no OpenAI, no internet
- Similarity threshold — never hallucinates missing knowledge
- Add any file to ~/Notes → AI instantly knows it
| Agent | What it does |
|---|---|
| TaskAgent | Manages SQLite tasks via natural language |
| ResearchAgent | Queries your notes, falls back to LLM |
| CodingAgent | Writes code for your exact stack |
| AutomationAgent | Opens 30+ Mac apps via voice |
| Brain | Planning, strategy, decision making |
Open any app by voice. 30+ aliases supported. "Open VS Code", "Open Brave", "Open Spotify" — all work.
| Phase | Feature | Status |
|---|---|---|
| v1 | Voice + Brain + Tasks + Dashboard + 5 Agents | ✅ Complete |
| v2 | RAG Knowledge Engine — LlamaIndex + ChromaDB | ✅ Complete |
| v3 | LangGraph Multi-Agent System | ✅ Complete |
| v4 | Coming soon | 🔨 Building |
| Layer | Technology |
|---|---|
| LLM (text) | Qwen2.5-7B-Instruct-4bit via mlx_lm |
| LLM (voice) | Qwen2.5-3B-Instruct-4bit via mlx_lm |
| Speech to text | whisper-large-v3-turbo via mlx-audio |
| Text to speech | Kokoro-82M-bf16 via mlx-audio |
| Knowledge / RAG | LlamaIndex v0.10+ + ChromaDB |
| Embeddings | BAAI/bge-small-en-v1.5 (fully local) |
| Agents | LangGraph StateGraph |
| UI | Streamlit |
| Storage | SQLite |
| Hardware | Apple Silicon M-series (MLX native) |
Requirements: Mac with Apple Silicon (M1/M2/M3/M4) · 16 GB RAM minimum
git clone https://github.com/sameersharma06/sameer-ai-twin
cd sameer-ai-twin
pip install -r requirements.txt
streamlit run app.pyFirst run downloads ~6 GB of models once. Fully offline after that. No API keys. No accounts. No internet required after setup.
# Drop any .txt .md .pdf into ~/Notes
cp your_notes.md ~/Notes/
# Rebuild the index — takes 30 seconds
python knowledge/ingestor.py
# Your AI now knows everything in those filesUser Input (voice or text)
↓
Router (LangGraph)
↓
┌───────────────────────┐
│ TaskAgent │ → SQLite database
│ ResearchAgent │ → ~/Notes (ChromaDB)
│ CodingAgent │ → LLM (7B)
│ AutomationAgent │ → Mac (subprocess)
│ Brain │ → LLM (7B) + context
└───────────────────────┘
↓
Memory Logger → SQLite events table
↓
Response → Text + Voice (Kokoro TTS)
This project is built and maintained by one student in Haryana, India.
If SAMEER AI TWIN helped you or you want to support local AI development:
Sponsor this project:
- GitHub Sponsors — coming soon
- for sponser [sameersharmaa95@gmail.com]
For companies: If your company builds AI tools, developer tools, or Apple Silicon products and wants to be featured in this README — reach out.
Contact: [sameersharmaa95@gmail.com] · [https://www.linkedin.com/in/sameersharma0028/]
What sponsors get:
- Logo in README (seen by every developer who finds this)
- Mention in every LinkedIn/Twitter update
- Early access to Pro features
- Direct feedback channel
This is an open-source personal AI project. Contributions welcome.
Areas where help is needed:
- Windows/Linux port (currently Mac only)
- Better embedding models for Indian languages
- Voice model improvements
- UI enhancements
Open an issue or PR — all contributions acknowledged.
The public repo contains the core open-source system.
Interested in early access? [sameersharmaa95@gmail.com]
I am a solo AI builder from Haryana, India actively looking for:
- AI/ML Engineering Internship (remote or Delhi-NCR)
- Research Internship (AI, LLM, Agents)
- Part-time AI consulting
- Open source collaboration
What I bring:
- Built a full local AI OS from scratch — voice, memory, RAG, agents
- Deep hands-on experience with MLX, LangGraph, LlamaIndex, Whisper, TTS
- Apple Silicon ML stack (rare skill in India)
- Ships fast, builds in public, learns daily
Stack I work with: Python · MLX · LangGraph · LlamaIndex · ChromaDB · Streamlit · SQLite · Whisper · Kokoro TTS · Qwen2.5 · Apple Silicon
If you are building AI products and need someone who:
- Understands local LLM deployment
- Can build multi-agent systems
- Moves fast and ships working code
→ Reach out directly:
📧 Email: [sameersharmaa95@gmail.com] 💼 LinkedIn: [https://www.linkedin.com/in/sameersharma0028/] 🐦 Twitter/X: [sameersharma28_] 📍 Location: Haryana, India · Open to remote globally · Available for in-office roles
Star ⭐ this repo if you want your own personal AI OS. Follow for weekly updates as I build this live.