Skip to content

Cluster Integration

Backend.AI GO is fully integrated with the Backend.AI Enterprise ecosystem. This allows individual researchers or developers to seamlessly transition from local prototyping to large-scale cluster inference without changing their workflow.

Architecture

When you connect to a Backend.AI Cluster, your local GO instance acts as a smart client.

  1. Authentication: Unlike simple API tokens (like OpenAI), Cluster connections use HMAC-SHA256 signatures. This ensures that every request is cryptographically signed with your Secret Key, preventing replay attacks and ensuring authenticity without sending the secret over the wire.
  2. Session Management: The cluster dynamically provisions "Compute Sessions" or "vFolders" based on your request.
  3. Routing: The Continuum Router inside GO handles the decision of whether to run a prompt locally or forward it to the cluster.

Key Differences from Cloud Providers

While you can add "OpenAI" or "Anthropic" as providers, "Backend.AI Cluster" is treated as a first-class mesh citizen:

Feature Cloud Provider (e.g., OpenAI) Backend.AI Cluster
Protocol Standard HTTP/REST Backend.AI RPC / Continuum Protocol
Auth Bearer Token (API Key) HMAC-SHA256 (Access/Secret Keys)
Discovery Static Model List Dynamic Model Discovery
Privacy Data sent to public cloud Data stays within enterprise private cloud

Security

  • Zero-Trust: No persistent connections are assumed secure. Every request is signed.
  • Key Storage: Your Secret Key is stored in your operating system's secure keychain (macOS Keychain, Windows Credential Locker, Linux Secret Service), not in plain text files.

Usage Scenarios

  • Fine-tuning: Use local resources for small-scale testing, then switch to the cluster to fine-tune a 70B parameter model.
  • Batch Processing: Offload heavy batch inference jobs to the cluster while keeping your local machine responsive.
  • Sensitive Data: Process strictly confidential data on an on-premise Backend.AI Cluster instead of sending it to public APIs.