Manual Registration¶

While Backend.AI GO handles local resources automatically, connecting to external resources, like a corporate Backend.AI Cluster or a colleague's high-performance workstation, requires manual registration. This ensures security and explicit control over where your data is sent.

Connecting to a Peer (Backend.AI GO)¶

Peer-to-peer connections between GO instances are managed in Settings > Nodes. The computer that shares its models creates a one-time connection key, and the computer that wants to use those models registers with that key.

1. On the computer that will share its models (the sub node), open Settings > Nodes > Share this computer and turn on Share with other computers. Under Connection methods > Connect with a connection key, click Create key. The key is shown as an aigo:// string and as a QR code, and it expires after the time chosen in Valid for. It carries the address this computer had when the key was made, shown with the key as Address in this key. If the computer joins another network while the key is on screen, the key is marked Made on the previous network; click Create new key there to replace it with one that carries the new address.
2. On the computer that will use the models (the head node), open Settings > Nodes > Connect other computers. Under Add Node, paste the key into Connection key from the other computer, or click Scan QR Code to read it with the camera, then click Connect. Opening the aigo:// link on that computer opens this page with the key already filled in. A computer on the same local network that allows connections without approval is also listed under Computers on your network and can be added there without a key; see Auto Discovery.
3. Once registered, the node is health-checked automatically, its running models appear in the chat model selector grouped under the node name, and you can run or stop models on it remotely. See Network Node Model Control and Distributed Routing.

The peer authenticates with a key issued during pairing that is limited to managing running models, health checks, and inference; it cannot touch model files, settings, or logs on your machine.

Checking a connected computer¶

Each check shows its result on its own button: the icon turns while it runs, then becomes a check mark when it succeeds or an X when it fails, and the button returns to normal after a moment.

• The refresh icon beside the counts under Connected computers checks every connected computer at once. Each card's status shows what its computer answered.
• Check on a card checks that computer. A computer that stops answering reads Not connected and shows Reconnect, which checks it again right away.
• Test Connection on a computer's Overview tab sends one request to it and shows the response time, or Connection failed with the reason and what to check under the actions.
• Run Diagnostics on the Connection tab runs the checks in order and lists each one as it finishes: the connection with its response time, then the model list with how many models the computer is running. A failed connection ends the run there. The last line says whether the checks passed or which one failed.

While a computer does not answer, its models are taken out of the model list chat sends requests through, so the other models, local ones included, are listed without waiting on it. In the chat model selector they stay under the computer's name but cannot be chosen, and they come back when it answers again. After a computer stops answering, Backend.AI GO checks it a few more times, waiting longer each time, and then waits for Reconnect.

When a check fails, the reason says what happened and the line under it what usually causes it:

Adding a computer reports a failure to reach it with the same reasons: under the key field when you connect with a key, or on the computer's card under Computers on your network.

A computer that answers but runs no models passes, with a note that none of its models can be used here until one is running there.

Entries of type Backend.AI GO in the legacy Mesh section (under Settings > API Server) are kept for reference only and do not affect routing. Re-register those machines from Settings > Nodes > Connect other computers.

Connecting to a Backend.AI Cluster¶

If your organization uses Backend.AI Enterprise, you can connect your GO client directly to the cluster.

Prerequisites¶

You need the following information from your cluster administrator:

• Endpoint URL: The API address of the cluster (e.g., https://api.my-cluster.ai).
• Access Key: Your public identifier.
• Secret Key: Your private signing key (never share this).

Steps¶

1. Open Settings by clicking the gear icon in the sidebar, then select API Server.
2. Open the Mesh section.
3. Scroll to the Backend.AI clusters section.
4. Click Add cluster.
5. Enter the details:
   • Connection name: A friendly name for this connection (e.g., "Company GPU Cluster").
   • Base URL: The full endpoint URL provided by your admin.
   • Access key: Paste your access key.
   • Secret key: Paste your secret key.
6. Click Add.

The system will immediately attempt to verify the connection. If successful, the status indicator will turn green, and models available on the cluster will start appearing in your model selection list.