API Hub / Gateway

Bring chat, image, video, voice, and RunPod under one entry point

This is not a single-model page. It is the first productized AI gateway layer inside your site: unified entry points, model catalog, provider management, and RunPod integration before billing, logs, debugging, and quotas.

Open Playground Manage API Keys View usage

Current gateway status

This panel reads the project-visible provider configuration state to show which capabilities can already move through the gateway layer.

API Base URL

Public paths use /v1/* (Kie-compatible jobs at /v1/jobs/*). Auth: Authorization: Bearer sk-xxx. Async jobs: balance checked at createTask; credits charged only on recordInfo success.

...

cURL example

No response yet. Pick a capability, choose a configured provider, then send a request.

Inline gateway tester

Use the page itself to test the first live gateway routes. Chat and image already go through real gateway forwarding when the matching provider key is configured.

Capability

Provider

Prompt

No enabled provider currently supports this capability. Configure one under Admin > Settings > AI first.

Gateway response

/api/gateway/chat/completions

No response yet. Pick a capability, choose a configured provider, then send a request.

Scenario chat

Unify `/chat/completions` and `/responses` first, with OpenRouter, Codex-compatible, OpenAI, and RunPod vLLM as the first-class upstreams.

Image generation

Put OpenAI, Replicate, Fal, and future RunPod image workers into one capability catalog while still routing internally through adapters.

Video generation

Keep video async and provider-aware instead of forcing it into synchronous chat semantics, which makes Fal, Replicate, and RunPod custom workers easier to support.

Speech synthesis

Create a unified TTS entry point and provider inventory, then layer on voice libraries, cost controls, and duration limits.

Speech-to-text

Standardize STT and transcription entry points so OpenAI, Replicate, backend_media, and RunPod audio workers can be managed consistently.

Unified responses

Reserve a more modern OpenAI-compatible surface for future agents, structured output, and multimodal conversations.

Provider matrix

The first version centralizes OpenAI-compatible providers, RunPod, Fal, and Replicate into one operational view.

How RunPod fits

In this architecture, RunPod should act as an upstream execution layer, not the full product layer.

RunPod vLLM

Best for chat, responses, and multimodal LLM traffic through OpenAI-compatible routing.

RunPod Custom Worker

Best for image, video, TTS, STT, or provider-specific async workloads with dedicated adapters.

Gateway API surface

The first pass standardizes naming and capability segmentation so product APIs stop being scattered across generate/* passthrough routes.

Bring chat, image, video, voice, and RunPod under one entry point

Search models

Bring chat, image, video, voice, and RunPod under one entry point