路墨（Lumoswitch） inference entry
An independent inference data plane with one model control plane
Clients connect directly to 路墨（Lumoswitch） inference services while teams centrally manage multi-channel models, access keys, fixed or failover routing, limits, and usage.
Why separate the control and inference planes?
When a team uses several model services or custom compatible channels, clients should not be bound to every upstream. Lumoswitch normalizes inference differences and manages service state and business billing.
Lumoswitch does not proxy requests
The Lumoswitch control plane maintains model resources, health, routing, access keys, platform-model billing, and call observability. A separate inference runtime handles allowlisted protocol endpoints, streaming responses, and upstream calls.
Where it fits
It fits teams switching among integration channels, centrally managing API keys, or connecting OpenAI-compatible agents to a shared model-resource pool. It also fits platforms that offer platform models and meter cost.
Where it does not fit
For a personal project making a temporary call to one model, using that provider's API directly is simpler. Lumoswitch is most useful in long-running operations with many models, users, keys, resilience, and billing.