OpenRelay API Documentation

This is the fastest path onto OpenRelay: create a key, log in, deploy a GPU VM, and SSH into it. If you want an OpenAI-compatible API to call from your app instead of your own box, use the Inference API.

Create an account and an API key

Sign up at app.openrelay.inc. Add a card and make a deposit to fund your balance; your first deposit gets a $5 credit match.

Signing up creates an organization for you: your workspace that owns your keys, VMs, and balance. In the dashboard, open Account > API Keys, create a key, and copy it. The plaintext value (it starts with or_) is shown once, so paste it somewhere safe.

The key is bound to that organization. You never pass an org id to orl: it reads the org from the key.

Install and log in

Install orl (see Install), then log in and paste the key when prompted:

orl auth login

orl stores the key in your OS keyring and confirms which organization it acts for. Check it any time:

orl auth status

Check what it costs

GPU VMs are billed per hour while they run. See live rates and what has capacity right now before you deploy:

orl catalog pricing get
orl catalog gpu-availability get

Deploy a GPU VM

orl deploy takes a name and at least one --ssh-key, and creates the VM. Pass --connect to wait for it to boot and drop you into an SSH session:

orl ssh-keys list   # pick a key id
orl deploy trainer --gpu-model "h100 pcie" --gpu-count 1 --ssh-key $SSH_KEY_ID --connect

Give it only a name and a key in a terminal (orl deploy trainer --ssh-key $SSH_KEY_ID) and orl walks you through the GPU choice from live availability.

--gpu-model takes a catalog id or a model name. Names are matched loosely, so "h100 pcie", H100 PCIe and h100-pcie are the same model. Where one GPU ships in more than one board, each is its own model at its own price: H100 PCIe and H100 SXM are separate choices, not one entry the scheduler picks from. Run orl catalog gpu-model list to see what is sellable.

Boot takes a few minutes while the VM provisions and starts, longer for many-GPU shapes. Use --wait instead of --connect if you just want the id once it is ready, or drop both flags to return immediately.

Connect, watch spend, and stop

Reconnect to a running VM any time, and check your balance and current-month spend:

orl vms ssh <vm-id>
orl balance

A VM keeps billing until you stop or terminate it:

orl vms stop <vm-id>        # keep the disk, stop the meter
orl vms terminate <vm-id>   # destroy it (asks first; --force to skip)

Next steps

The signature commands and how to script orl in CI: Workflows
Serve the whole API to an agent with one key: Agents (MCP)
The same flow with curl and full API detail: API Quickstart