Hardware Requirements¶
Minimum Requirements¶
| Component | Minimum | Recommended |
|---|---|---|
| CPU | 4 cores | 8+ cores |
| RAM | 16 GB | 32 GB |
| Storage | 50 GB SSD | 100 GB+ NVMe |
| GPU | NVIDIA with 8 GB VRAM | NVIDIA with 16+ GB VRAM |
| OS | Ubuntu 22.04+ / Debian 12+ | Ubuntu 24.04 LTS |
GPU Support¶
ai.doo uses Ollama for local AI inference. GPU acceleration is strongly recommended.
Supported GPUs:
- NVIDIA (recommended): Any GPU with CUDA support and 8+ GB VRAM. Install the NVIDIA Container Toolkit.
- AMD ROCm: Supported via Ollama's ROCm build. Start with
make up ROCM=1. - CPU-only: Works but significantly slower. Start with
make up CPU=1.
Software Prerequisites¶
| Software | Version | Required For |
|---|---|---|
| Docker | 24.0+ | All services |
| Docker Compose | v2.20+ | Orchestration |
| curl | Any | Installer script |
| NVIDIA Container Toolkit | Latest | GPU inference |
Model Storage¶
AI models are stored in a shared Docker volume (ollama_models). Budget storage based on the models you plan to use:
| Model | Size |
|---|---|
| llama3.2:3b | ~2 GB |
| llama3.1:8b | ~4.7 GB |
| qwen2.5-coder:14b | ~9 GB |
Network¶
- All services communicate over an internal Docker network (
ollama_network). - No outbound internet required after initial setup (model downloads).
- Default ports: Ollama
:11434, Hub:2000, PIKA:8000, VERA backend:4000, VERA frontend:3000.