AI Setup

This page documents ElastOS v1 (PC2), the proof of concept. ElastOS v2 is the current system. To install it, see Install ElastOS. To see how the stages fit together, see ElastOS in stages.

Configure local AI for your PC2 personal cloud. This guide covers Ollama installation, model selection, GPU acceleration, and connecting external providers.

With Ollama (local AI), your conversations stay 100% private - they never leave your machine.

Installing Ollama

curl -fsSL https://ollama.com/install.sh | sh

Verify installation:

ollama --version
ModelSizeRAM NeededBest For
deepseek-r1:1.5b1GB4GBFast responses, basic tasks
llama3.2:3b2GB6GBGood balance
phi3:mini2GB6GBMicrosoft’s efficient model
mistral:7b4GB8GBStrong general purpose
llama3.2:8b5GB12GBComplex reasoning
codellama:7b4GB8GBCode generation

Install a Model

ollama pull deepseek-r1:1.5b

Or via PC2: Settings → AI Setup → Click “Install” on any model.

List Installed Models

ollama list

GPU Acceleration

GPU dramatically improves AI speed.

NVIDIA GPUs (CUDA)

Ollama automatically uses NVIDIA GPUs if CUDA is available.

# Check GPU is being used
nvidia-smi

Apple Silicon (M1/M2/M3)

Ollama automatically uses Metal acceleration. No configuration needed.

No GPU?

CPU-only works for smaller models (1.5b-3b). Larger models will be slow but functional.

Connecting to Remote Ollama

If running Ollama on a different machine (like a powerful server):

On the Ollama Server

OLLAMA_HOST=0.0.0.0 ollama serve

In PC2 Settings

  1. Go to Settings → AI Setup
  2. Set Ollama URL to http://server-ip:11434
  3. Save

External AI Providers

PC2 also supports cloud providers for when you need more power:

ProviderModelsGet API Key
OpenAIGPT-4, GPT-3.5platform.openai.com
AnthropicClaude 3console.anthropic.com
GoogleGeminiaistudio.google.com
xAIGrokxAI dashboard
⚠️

Cloud providers send data to their servers. Use Ollama for maximum privacy.

Troubleshooting

”Ollama not available”

# Check if running
curl http://localhost:11434/api/tags
 
# Start Ollama
ollama serve

Slow Responses

  1. Use smaller model (deepseek-r1:1.5b)
  2. Enable GPU (see above)
  3. Check system resources: htop

Out of Memory

  1. Use smaller/quantized model
  2. Close other applications
  3. Add more RAM or swap

Model Recommendations by Use Case

Use CaseRecommended Model
General chatllama3.2:3b, mistral:7b
Codingcodellama:7b, deepseek-coder:6.7b
Writingmistral:7b, llama3.2:8b
Fast responsesdeepseek-r1:1.5b, phi3:mini
© 2026 Elacity Labs. ELASTOS® is a registered trademark of Elastos DAO. ElastOS is open source under the MIT licence.