🤖 Air-Gapped AI
Offline Model Selection Guide
Enter your hardware constraints and security task requirements. Get a specific model and quantisation recommendation with full reasoning.
⚙️ Your Environment & Requirements
📊 Full Model Reference — Air-Gapped Security Use
| Model | Size | Min VRAM (Q4) | Min VRAM (Q8) | Log Analysis | Alert Triage | Tool Calling | JSON Output | Context | Best For |
|---|---|---|---|---|---|---|---|---|---|
| Phi-3 Mini 3.8B | 3.8B | 2.5 GB | 4 GB | ★★★ | ★★☆ | Poor | OK | 128K | CPU-only, resource-constrained |
| Mistral 7B v0.3 | 7B | 4.5 GB | 8 GB | ★★★★ | ★★★ | Limited | OK | 32K | Single GPU, fast inference |
| Llama 3.1 8B | 8B | 5 GB | 9 GB | ★★★★ | ★★★ | Limited | Good | 128K | Best 8B overall — long context |
| Phi-3 Medium 14B | 14B | 9 GB | 15 GB | ★★★★ | ★★★★ | Limited | Good | 128K | Best quality per VRAM on 16-24GB |
| Mistral Nemo 12B | 12B | 8 GB | 13 GB | ★★★★ | ★★★ | Limited | Good | 128K | Efficient, good instruction following |
| Qwen 2.5 7B | 7B | 4.5 GB | 8 GB | ★★★★ | ★★★ | Limited | Good | 128K | Strong instruction following at 7B |
| Llama 3.1 70B | 70B | 43 GB | 75 GB | ★★★★★ | ★★★★★ | Good | Good | 128K | Best overall air-gapped quality |
| Qwen 2.5 72B | 72B | 45 GB | 76 GB | ★★★★★ | ★★★★★ | Good | Excellent | 128K | Best JSON/structured output at 70B |
| Mistral Large 2 123B | 123B | 70 GB | 130 GB | ★★★★★ | ★★★★★ | Excellent | Excellent | 128K | Top tier — requires significant hardware |