Clawdbot: Hosting Your Own AI Assistant

🤖 Why Self-Host?

The current AI landscape is dominated by cloud giants. Your prompts, your data, your context—all sent to a server. Clawdbot challenges this by running locally on your device or a private server you control.

Benefits

1. Privacy: No data leaves your control.
2. Cost: Run small models for free.
3. Customization: Swap models (DeepSeek, Llama 3) without waiting for API updates.

🏗️ Architecture

Clawdbot follows a client-server model, even if both run on the same phone.

Server (Node.js / Python)

• Handles LLM inference (using Ollama or LocalAI).
• Manages vector database (ChromaDB) for memory.
• Exposes a simple REST API.

Client (Android)

• UI: Jetpack Compose interface.
• Networking: Retrofit to talk to localhost:11434 (Ollama default).
• Voice: Android Speech-to-Text.

📱 Setting Up on Android

1. Install Termux

Run a Linux environment on Android.

pkg update && pkg upgrade
pkg install nodejs python

2. Run the Backend

Clone the repo and start the server.

git clone https://github.com/arceapps/clawdbot-backend.git
cd clawdbot-backend
npm install
npm start

3. Configure the App

Point the Clawdbot Android app to http://localhost:3000.

🧠 Model Selection

For mobile, you need quantized models (GGUF format).

• Llama 3 8B (Q4_K_M): Good balance of speed and intelligence. ~5GB RAM.
• Phi-3 Mini (3.8B): Extremely fast, surprisingly smart. ~2.5GB RAM.
• Gemma 2B: Lightweight, runs on almost anything.

🚀 The Future: On-Device Hardware Acceleration

With Android 15 and NPU access (via AICore), apps like Clawdbot will soon run inference natively without Termux, tapping into the dedicated AI chips on Pixel and Samsung devices.

🏁 Conclusion

Self-hosting AI is not just for hackers. It’s the only way to guarantee privacy in the age of intelligence. Clawdbot is a proof-of-concept that you can own your AI assistant today.