Developers

Documentation.

Everything you need to ship Huxகn in production.

From signup to your first inference call in under five minutes. One GPU VM, one curl command.

How Hux and Kan fit together. The data plane, the control plane, what crosses the wire.

The OpenAI-compatible inference endpoint, the Kan management REST API, and the agent enrollment protocol.

Multi-fleet topologies. Custom TLS certs. Rate limit tuning. Budget alert wiring. Backups & restore.

Coming soon

Concepts.

Org — your tenant. Isolated by Postgres row-level security.
Fleet — a logical group of GPU nodes that share an apisix.yaml manifest and a Redis cache. Usually one per region or per environment.
Node — a single GPU machine running Hux.
Model — a logical inference target (named by slug). Backed by one or more deployments.
Deployment — a model running on a specific node, in a specific container, on a specific port.
Consumer / API key — your end users. Keys are issued to consumers, consumers belong to consumer groups.
Consumer group — bundles rate limits + ACL policies. Apply to N consumers.
Budget — token or USD cap per period. Scope: api_key, consumer_group, or org.