Run AI Models on Your Local Mac / CPU / GPU Rig. Share Them With The World.
Turn your Apple Silicon Mac, NVIDIA GPU, or any computer into a free, public AI API.
Run Llama, Qwen, Mistral, DeepSeek — any open-source LLM. OpenAI-compatible.
Loading available models...
./start-app.sh --tunnel
Search & download via the built-in Web UI
Sign in with Google at llamanet.app
OpenAI-compatible, auto-routed
import openai
client = openai.OpenAI(
base_url="https://llamanet.app/v1",
api_key="your-api-key"
)
response = client.chat.completions.create(
model="Ministral-3-3B-Instruct-2512-Q4_K_M.gguf",
messages=[{"role": "user", "content": "Hello!"}],
stream=True
)
for chunk in response:
if chunk.choices[0].delta.content:
print(chunk.choices[0].delta.content, end="")curl https://llamanet.app/v1/chat/completions \
-H "Authorization: Bearer your-api-key" \
-H "Content-Type: application/json" \
-d '{"model": "Ministral-3-3B-Instruct-2512-Q4_K_M.gguf", "messages": [{"role": "user", "content": "Hello!"}]}'Start instantly, download models via Web UI
Switch models without restarting your node
Drop-in replacement for any OpenAI client
Automatic Cloudflare tunnels, zero config
Search Hugging Face, download with progress tracking
DeepSeek-R1, Qwen reasoning models
Auto-route to the best available node
NVIDIA, Apple Silicon, CPU fallback
Run any GGUF model. These tiers map to your hardware.
| Tier | Model | Size (Q4) | Min RAM | Best For |
|---|---|---|---|---|
| Tiny | Phi-4 Mini | ~1.5 GB | 8 GB | Quick tasks, edge devices, fast iteration |
| Fast | Qwen 4B | ~2.5 GB | 8 GB | Daily driver — fast responses, solid quality |
| Standard | Ministral 3 8B | ~4.5 GB | 16 GB | Best all-round balance of speed and quality |
| Coding | Qwen 14B | ~8 GB | 32 GB | Code generation, technical tasks |
| Premium | GPT-OSS 20B | ~12 GB | 32 GB | High-quality general-purpose inference |
| High Quality | Qwen 35B-A3B | ~20 GB | 48 GB | Expert-level reasoning, long-form generation |
| Reasoning | DeepSeek-R1 Distill 32B | ~18 GB | 48 GB | Chain-of-thought reasoning, math, analysis |
Sizes are approximate for Q4_K_M quantization. Apple Silicon uses unified memory. Use Model Search above to find and download any model.
Pick your platform. Same command. Automatic GPU detection.
Metal GPU acceleration is automatic. No drivers, no CUDA, no setup. Works on MacBook Air, MacBook Pro, Mac Mini, Mac Studio, and Mac Pro with any M-series chip.
git clone https://github.com/machaao/llama-net.git
cd llama-net
pip install -r requirements-inference.txt
sh start-app.sh --tunnel --bootstrap-peers https://llamanet.app
8GB RAM (M1/M2/M3/M4 base): Tiny & Fast tiers at Q4_K_M
16GB RAM (Pro): Up to Standard tier comfortably
36GB+ RAM (Max/Ultra): Up to Coding tier and beyond
Works on any Linux or Windows system with NVIDIA drivers and CUDA installed. LlamaNet auto-detects your GPU and offloads model layers automatically.
git clone https://github.com/machaao/llama-net.git
cd llama-net
pip install -r requirements-inference.txt
sh start-app.sh run hf.co/mistralai/Ministral-3-8B-Instruct-GGUF:Q4_K_M \
--tunnel --bootstrap-peers https://llamanet.app
8GB VRAM (RTX 3070/4060): Tiny & Fast tiers
12GB VRAM (RTX 3060 12GB/4070): Up to Standard tier
24GB VRAM (RTX 3090/4090): Up to Premium tier
No dedicated GPU? Smaller models (Tiny & Fast tiers) run well on modern CPUs. Significantly slower than GPU inference, but works on any machine.
N_GPU_LAYERS=0 sh start-app.sh \
--tunnel --bootstrap-peers https://llamanet.appBest with Tiny tier models: Phi-4 Mini, Qwen 1.5B