Platform overview

GridNexa provides markets, identity, a model catalog, routing, billing, task scheduling, and compute delivery. The developer platform exposes those capabilities to third-party developers.

You can consume chat, image, and video capabilities directly, or publish a complete application that users invoke from the Marketplace or an external experience.

• Model capabilities are exposed through /v1/invoke.
• Developer apps are exposed through /v1/apps/:applicationId/invocations.
• The Marketplace lists an app only when both the app and its target market are published.

GridNexa exposes two invocation endpoints: one for platform model capabilities, and one for published applications. An application worker can also call the model endpoint while handling an application request.

A. Developer application -> POST /v1/invoke -> GridNexa AI capability

B. User / experience -> POST /v1/apps/{applicationId}/invocations
   -> current release -> Load Balancer or Queue worker

An application can call /v1/invoke from inside its own worker.

End users interact with a platform experience or developer-owned UI. App developers maintain releases, manifests, and business code. GridNexa owns the unified invocation, routing, scheduling, and call record.

• Worker addresses, developer API keys, and runtime tokens remain within the platform and application runtime.
• Load Balancer applications implement an ordinary HTTP service.
• Queue containers integrate the gridnexa-runtime Worker SDK and communicate through the Runtime API.
• Platform Gradio and external experiences are UI layers, not alternate runtimes.

For /v1/invoke, capability defines the input, output, and billing contract. model narrows the candidates. The published binding and market execution policy then select either third-party services or the ModelHub compute pool.

capability (required) + model (optional) + routing preferences
  -> published model binding
  -> market execution policy
  -> third-party provider OR ModelHub + compute pool
  -> HTTP 200 OR HTTP 202 task

If you only need chat, image, or video generation, continue to the model-call quickstart. If you want to publish your own workflow, read Application Invocation and then choose a runtime.