Get started

Pick what you want to do. We'll show you how.

I want to run models on my hardware

You have a GPU and you want to run AI models locally. No cloud, no accounts, no fees.

anvil — 5 min setup

I want to understand the protocol

You're building an agent framework or memory backend and want to implement A2M.

A2M Protocol — interactive explorer

anvil

Run models on your hardware

If you've used Ollama before, this will feel familiar — but without the Docker container, the account requirement, or the opaque blob store. If you haven't, this is the easiest way to run AI models on your own machine.

What you need

A computer with a GPU. NVIDIA (any modern card with 6GB+ VRAM), AMD, or Apple Silicon. That's it. No Docker, no Python, no special accounts.

Install anvil

# Install anvil$ curl -fsSL https://raw.githubusercontent.com/sovereignty-labs/anvil/main/install.sh | sh# Install the runtime$ anvil runtime install

Pick a model

Models come from HuggingFace. No registry accounts needed — just a direct download. Here are some good starting points:

Model VRAM needed Good for

Qwen3-8B-Q4_K_M 6 GB General chat, coding assistance

Qwen3-14B-Q4_K_M 10 GB Better reasoning, longer context

Qwen3-32B-Q4_K_M 20 GB Complex tasks, code review

Start chatting

$ anvil pull unsloth/Qwen3-8B-GGUF:Q4_K_M$ anvil serve && anvil load Qwen3-8B-Q4_K_M.gguf✓ Model loaded on GPU 0 — port 11434✓ OpenAI-compatible API ready at http://localhost:11434

That's an OpenAI-compatible API. Any tool that works with OpenAI's API works with anvil. Point Open WebUI, Continue, Cursor, or any other client at http://localhost:11434.