🤖 termux-llm

Run uncensored AI models locally on your Android phone. One-script setup via Termux + llama.cpp. No root, no cloud, no internet needed after setup.

📺 Watch the Tutorial

⚡ How It Works

📱 Browser (127.0.0.1:8080)  ← You chat here
       ↕
🌐 llama-server (HTTP API)   ← Serves chat UI + OpenAI API
       ↕
🧠 llama.cpp (C++ Engine)    ← Runs the AI on your CPU
       ↕
📦 GGUF Model File           ← The AI's brain
       ↕
🐧 Termux (Linux on Android) ← Real Linux environment

Everything runs 100% on-device. Your conversations never leave your phone.

🚀 Quick Start

1. Install Termux

Download Termux from F-Droid (recommended) or the Play Store.

2. Download a Model

Download any .gguf model file and place it in your phone's storage:

/sdcard/AI_Models/

Create the AI_Models folder if it doesn't exist. The script will auto-detect all .gguf files in this folder.

3. Run the Installer

Open Termux and paste:

curl -fsSL https://raw.githubusercontent.com/orailnoor/termux-llm/main/install.sh -o ~/install.sh && bash ~/install.sh

Or if you downloaded the script manually:

cp /sdcard/install.sh ~ && bash ~/install.sh

4. Start Your AI

After setup completes, start the server anytime with:

bash ~/start-ai.sh

Then open your browser and go to: http://127.0.0.1:8080

📦 Recommended Models

If no models are found in /sdcard/AI_Models/, the script offers these ungated downloads:

Model	Size	Type	Best For
Gemma-2-2B-Abliterated	1.6 GB	🔓 Abliterated	Permanently uncensored, very smart
SmolLM2-1.7B-Uncensored	1.0 GB	🔓 Uncensored	Ultra lightweight
Qwen2.5-1.5B-Instruct	1.1 GB	🔒 Standard	Great multilingual support
Phi-3.5-Mini-3.8B	2.2 GB	🔒 Standard	Strong reasoning

💡 Tip: Search bartowski on HuggingFace for more models. Always choose Q4_K_M quantization for the best balance of quality and speed on phones.

🔓 Uncensored vs Abliterated

Uncensored (fine-tuned): Trained to not refuse, but may revert after long conversations.
Abliterated: Refusal neurons surgically removed from the model weights. Cannot censor, ever. This is the recommended type.

📱 Supported Devices

RAM	Recommended Model Size
3-4 GB	Up to 1B models
4-6 GB	Up to 2B models
6-8 GB	Up to 3B models
8+ GB	Up to 8B models

Performance: Expect ~2-10 tokens/second depending on your processor and model size.

🔁 Switching Models

Download a new .gguf model
Place it in /sdcard/AI_Models/
Re-run the installer:
```
bash ~/install.sh
```
Select the new model from the list

The engine won't recompile — it only downloads/copies the new model.

🔧 Connecting External Apps

The server exposes an OpenAI-compatible API at:

http://127.0.0.1:8080/v1

You can connect apps like SillyTavern, Open WebUI, or any OpenAI-compatible client by pointing them to this URL.

🐞 Troubleshooting

Issue	Fix
`pkg update` fails	Run `termux-change-repo` and pick a mirror
Model download 401 error	Model is "gated" — download it in your browser instead
Phone gets very hot	Use a smaller model or reduce threads: edit `start-ai.sh` and change `-t 4` to `-t 2`
Out of memory crash	Close all other apps, use a smaller model
Server won't start	Make sure no other app is using port 8080

📜 License

MIT License — See LICENSE for details.

If you found this useful, consider subscribing to orailnoor on YouTube!

termux-llm

Specs

README