backend-ai-guide

Backend.AI is a distributed computing platform with five core components: Manager coordinates orchestration, Agent executes workloads, Storage Proxy handles data, Webserver serves APIs, and App Proxy routes frontend requests. Together they enable scalable kernel deployment, REST/GraphQL access, and RBAC-controlled multi-tenant operations.

Backend.AI setup requires configuring container runtimes, accelerators (GPUs, TPUs), and vfolders for persistent storage. The Manager orchestrates these across Agents; deployment documentation covers runtime selection, kernel provisioning, jail isolation, and hook runtime integration for custom workload execution.

Backend.AI exposes both REST and GraphQL APIs for kernel management, session handling, and resource queries. GraphQL provides flexible, efficient queries for complex data retrieval; REST offers standard HTTP endpoints. Both support RBAC policies and WebUI integration.

Backend.AI implements role-based access control (RBAC) to manage user permissions across kernels, vfolders, and APIs. RBAC policies control who can launch kernels, access storage, and invoke endpoints, supporting multi-tenant deployments with fine-grained authorization.

The WebUI communicates with the Webserver and App Proxy components via REST/GraphQL APIs to display kernel sessions, manage vfolders, and execute code. The App Proxy routes frontend requests; the Webserver authenticates and delegates to the Manager for orchestration.

Scaling Backend.AI requires distributing Agents across compute nodes, configuring the Storage Proxy for high-throughput vfolder access, and tuning Manager resource pools. Architecture documentation covers multi-region setups, kernel affinity, and hook runtime customization for production workloads.