Docs 11

Running a node

Put a graphics card to work, in a browser tab, as a dedicated rig, or rendering images and get paid in USDC for every job it completes.

Three ways in

Node type	Hardware	Serves	Barrier to entry
Browser	~6GB free VRAM	Light tier (Nimbus 8B)	Lowest: open a tab, click start
Rig	20GB+ VRAM / 32GB+ unified memory	Heavy tier (Atlas 30B / Atlas Vision 27B)	Higher: one CLI command
Render	24GB+ VRAM	Image generation	Higher: one CLI command

Browser node

Turn a browser tab into a node. No install.

Go to usetalos.xyz/contribute.
Sign in with a linked account.
Click Start node: Nimbus 8B downloads and caches (~4.2GB, needs ~6GB free VRAM).

The tab shows live stats while it runs:

USDC earned: your total claimable balance
Uptime: how long the node has been active
Jobs completed: number of requests processed
tok/s: your current generation speed

Keep it open

Close the tab or navigate away and the node stops. Don't minimize it either: some browsers throttle background tabs, which kills throughput.

Tips:

Chrome or Edge have the best WebGPU support.
Check chrome://gpu to confirm WebGPU is enabled and pointed at your discrete GPU.
One node per browser instance: running two fights over VRAM instead of doubling throughput.

Rig node

Rig nodes run Atlas 30B or Atlas Vision 27B on your own machine, serve Heavy tier requests and earn several times more than a browser node.

Why go rig

Higher pay: Heavy jobs are worth significantly more than Light.
Bigger model: a 27–30B parameter model produces noticeably better output.
Real GPU use: full CUDA, Metal, or Vulkan acceleration, not a browser sandbox.
Headless: runs as a background process, no tab required.

Requirements

NVIDIA: RTX 3090, RTX 4090, or better (20GB+ VRAM)
Apple Silicon: M1 Max/Ultra or M2/M3/M4 Max (32GB+ unified memory)
AMD: RX 7900 XTX via Vulkan

Quick start

bash

npx @talos-network/node --token <your-token>

That's the whole install. The local runtime handles model download and GPU detection on its own.

Choosing a model

A rig node asks which model to run and shows how many nodes are live on each, nudging you toward whichever has the fewest so new supply balances out across the network. Skip the prompt with --model:

bash

npx @talos-network/node --token <your-token> --mode heavy --model atlas          # Atlas 30B
npx @talos-network/node --token <your-token> --mode heavy --model atlas-vision   # Atlas Vision 27B

Only the model you choose gets downloaded (~17GB). Get a token at usetalos.xyz/contribute.

Getting a token

Make sure you've got a stable connection.
Go to usetalos.xyz/contribute.
Sign in with a linked account.
Scroll to the Rig Node section.
Click Generate node token.
Copy and store it: it's shown once.

Render node

A rig can run as a render node instead of a text node, serving the image-generation side of the network by running the render pipeline and the Aperture HD model on your GPU, earning per image.

bash

npx @talos-network/node --mode render --token <your-token>

Left without --mode, the CLI asks on first run whether to act as a Heavy (text) node or a Render node and remembers your answer. Render mode only downloads the image model (~14GB), not the text model. A 24GB card (RTX 3090/4090) is recommended. Set RENDER_DIR if you want the node to install and launch the pipeline for you, or point RENDER_URL at an instance you already run.

Earnings

Every node type earns the same way: 72% of the USD value of the credits spent on each job it completes, paid in USDC (82% once you've staked at least 250,000 $TALOS for 48 hours). See Staking for the node boost mechanics.

Earnings are tracked on your node dashboard and tied to your account. Withdraw your balance as USDC whenever you like.

Previous ← Using TALOS