Skip to main content

Get the agent

The agent is one container. It registers your GPU, waits for work, runs the model, and reports back. You install it once and leave it running.

There is nothing to download — no installer, no binary to place, no Python to set up. Everything ships inside a Docker image, so the only things your machine needs are Docker and an NVIDIA driver.

docker pull ghcr.io/neurogrid-ai-exchange/neurogrid-agent:latest

Before you start

You need two things, and you get both from the console.

A machine token. Add your GPU in the console and it hands you a token that looks like ng_live_…. It is shown once. The agent uses it to claim the card, and never needs it again after that first run.

A password you choose. It encrypts the token where it is stored on your machine. You pick it, you keep it, and you need the same one every time the container is recreated — an upgrade included. We never see it, and we cannot reset it.

Losing the password means re-registering

The stored token is encrypted with a key derived from your password. A different password cannot read it, so the machine has to be registered again with a fresh token.

What happens on that first run

  1. The agent starts and finds no stored token, so it uses the one you passed.
  2. It reads your GPU, CPU, RAM, disk and rough location, and registers them.
  3. The backend checks the card matches what you listed in the console.
  4. The console flips your GPU to confirmed, and the agent goes idle, waiting for work.

From then on it heartbeats every 30 seconds. When a job arrives it pulls the model, serves it, and opens a tunnel so the client can reach it.

Requirements at a glance

MinimumRecommended
OS64-bit Linux (kernel 5.x+), or Windows 11 + WSL2Ubuntu 22.04 / 24.04 LTS
GPUNVIDIA, 6 GB VRAM16–24 GB+ VRAM
NVIDIA driver575+ (CUDA 12.9)Latest stable
CPU4 cores8+ cores
RAM8 GB32 GB+
Free disk50 GB SSD100 GB+ NVMe SSD
Network25 Mbps down, stable100 Mbps+ symmetric

Full requirements, including which cards are supported →

Then what