PraxisAI P900

AI assistant powered by Qwen3.5, branded as PraxisAI P900 by CozzyLabs.

Quick Start

# 1. Build llama.cpp
./build-llama.sh

# 2. Download models (optional - they'll download on first run)
./download-models.sh 9b

# 3. Launch PraxisAI
./launch.sh                    # 9B thinking server
./launch.sh -m 4b -M general    # 4B general mode
./launch.sh -m 2b --cli         # 2B CLI mode

Model Variants

Model	Parameters	VRAM	Quantized Size
Qwen3.5-0.8B	0.8B	~1GB	~500MB
Qwen3.5-2B	2B	~2.5GB	~1.2GB
Qwen3.5-4B	4B	~5GB	~2.5GB
Qwen3.5-9B	9B	~12GB	~5.5GB

Available Modes

• thinking - Reasoning mode with thinking enabled (llama-server)
• general - General tasks with thinking enabled (llama-server)
• cli - CLI mode for general tasks (llama-cli, non-thinking)
• reasoning - CLI mode for reasoning tasks (llama-cli, non-thinking)

Project Structure

PraxisAI/
├── data-assets/
│   └── system-prompt.txt     # PraxisAI identity & instructions
├── config/
│   ├── models.env            # 0.8B configuration
│   ├── 2b.env                # 2B configuration
│   ├── 4b.env                # 4B configuration
│   └── 9b.env                # 9B configuration
├── scripts/
│   ├── server/               # Server scripts (thinking mode)
│   ├── cli/                  # CLI scripts (non-thinking)
│   └── general/              # General task scripts
├── models/                   # Downloaded GGUF models
├── llama.cpp/                # Compiled llama.cpp binaries
├── launch.sh                 # Main launcher
├── download-models.sh        # Model downloader
└── build-llama.sh            # Build script

Requirements

• llama.cpp (compiled)
• CUDA (optional, for GPU acceleration)
• ~12GB RAM/VRAM for 9B model
• 256K context length support

Identity

PraxisAI is configured with a custom identity based on the system prompt in data-assets/system-prompt.txt.

• Name: PraxisAI
• Model: P900 (built on Qwen3.5 foundation)
• Creator: CozzyLabs / Praxis Solutions
• Owner: itzC9 / C9 / Yuri