3LC Hub - Getting Started

1

What are these services?

Object Service is a lightweight data server. It makes your datasets, tables and project metadata available to the 3LC Hub in your browser. It reads from your local disk (or cloud storage) and serves it over a local port. Runs anywhere — even a laptop. No GPU needed.

Compute Service powers all AI workloads: model training (YOLO, timm), auto-labeling (SAM3), data analysis, import/export, and more. It runs on a machine with an NVIDIA GPU. You can spin it up on-demand and shut it down when not in use.

Both services stay on your infrastructure. Nothing is uploaded to the cloud. The 3LC Hub in your browser connects to them over local ports (or an SSH tunnel).

2

Where will you run the services?

Pick where your GPU is — that's where the Compute Service needs to run. Object Service can run on the same machine or separately.

3

Platform & package manager

Select the OS of the machine where services will run, and your preferred Python installer. All commands below update automatically.

Operating system

Package manager

Don't have uv? Install: curl -LsSf https://astral.sh/uv/install.sh | sh

4

Who is this deployment for?

This decides which sources the Compute Service accepts plugin installs from. Your choice is folded into the start command in a later step — changing it afterwards means editing an environment variable and restarting the service.

catalog-only — plugins can only be installed from catalogs this deployment trusts. The safe default for any machine other people can reach.

Known catalog URLs (optional, comma-separated)

Catalogs listed here are trusted from the first start, so installing their plugins from the Hub just works — no env edit or restart later. You can also add catalogs from the settings page afterwards.

5

Connect to your server

Open a terminal on your laptop and SSH into your remote machine. You'll typically need the .pem key file you downloaded when creating the instance.

Where's my key? On AWS, you download the .pem file when creating the instance. On Azure, when creating the VM. Move it to ~/.ssh/ and set permissions: chmod 400 ~/.ssh/your-key.pem

5

Install 3LC

Create a Python environment and install 3LC. Run these on the machine where the services will run. The Compute Service package brings the core library with it — including the 3lc command that runs the Object Service — so this is a single install.

On a machine that runs only the Object Service, install the core library by itself — no Compute Service, no GPU.

No PyTorch install here. The Compute Service itself does not use PyTorch — each GPU plugin builds its own isolated environment with its own CUDA build when you install it. All this machine needs is an NVIDIA driver.

6

Set 3LC API key

Get an API key from https://account.3lc.ai/api-key. Set the API key on the machine where the services will run:

Replace <api_key> with the key you copied from your 3LC account.

This will create a file in the 3LC settings directory that contains your API key.
Alternatively, you can set the API key as an environment variable TLC_API_KEY.

7

Start the services

Open two terminal windows on the machine and start each service. Leave them running.

The Object Service shows a TUI interface. The Compute Service will output the URL where it may be reached.

8

Forward ports to your browser

The services run on your remote machine, but your browser runs on your laptop. An SSH tunnel bridges the two — no public ports needed.

This makes localhost:5016 and localhost:5020 on your laptop reach the remote services.

Keep this running while using the Hub. Add -o ServerAliveInterval=60 to prevent idle disconnects.

✓

Verify connection

Let's check if your browser can reach both services now.

All services connected