Inference Space Docs

Inference Space positioning, unified API capabilities (LLM / images / audio / OCR / vision segmentation), and the control-plane / data-plane split.

Inference Space is a unified model API product: one account, one gateway API key, and one billing system covering chat models, image generation and editing, speech recognition and synthesis, OCR, and image/video segmentation. Applications using OpenAI- or Anthropic-compatible clients can integrate with little or no code change.

The private deployment API is identical; only replace the Base domain → Enterprise private deployment.

Your first request

Create a gateway API key starting with gk_ in the console at https://ai.inf.space, then make a request with an Authorization: Bearer header. The default domestic route uses https://ai.inf.space, and the overseas route uses https://ai.inf.space:

curl "https://ai.inf.space/v1/chat/completions?provider=dashscope" \
  -H "Authorization: Bearer $TOS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "qwen-plus", "messages": [{"role": "user", "content": "Hello"}]}'

See Quickstart for the complete steps.

Control plane and data plane

Inference Space separates control-plane operations (console/browser) from data-plane operations (API calls) by URL path. The English public entry point uses the same host for both:

Route	Plane	Public host	Use
Public route	Control plane / console	`https://ai.inf.space`	Browser entry point for the console, provider configuration, API key creation and management, billing, and usage
Public route	Data plane	`https://ai.inf.space`	`/v1/*` API entry point for chat, images, audio, and other capabilities

All /v1/* calls use https://ai.inf.space; the same host also serves browser management pages. For private deployment, replace it with your own data-plane domain. See Routes and entry points for the complete SDK Base and endpoint list.

Some endpoints are master/control-plane paths on the same host: OCR /v1/recognize, vision segmentation /v1/vision-segment/*, public model lookup /v1/public/models/lookup, organization pricing lookup /v1/pricing/lookup, and the ASR/TTS aliases /v1/transcribe and /v1/synthesize. High-volume data-plane endpoints such as chat, images, and audio (/v1/audio/*) also use ai.inf.space; each capability page identifies the required path.

Unified capabilities

Inference Space exposes a set of unified capabilities. Each capability maps to a model category and a group of /v1/* endpoints:

Capability ID	Category	Description
`llm`	Chat models	Text chat, reasoning, and tool calls through OpenAI-compatible and native Anthropic interfaces
`image`	Image generation and editing	Text-to-image, image-to-image, and multi-image compositing
`asr`	Speech recognition	Speech-to-text
`tts`	Speech synthesis	Text-to-speech
`ocr`	OCR	Image/document text recognition
`vision-segment`	Vision segmentation (SAM3)	Image/video segmentation

The available model families, model IDs, and prices are shown in the console. We recommend managing model IDs as configuration so you can switch versions smoothly as the console changes. Query available models and pricing through the Model discovery and catalog endpoint.

Authentication

All API calls use gateway API key authentication. Create a key with the gk_ prefix in the console at https://ai.inf.space, then pass it in an HTTP header:

Authorization: Bearer gk_xxxxxxxxxxxxxx

API keys are authorized by capability scope (ai:chat for chat, plus ai:image / ai:asr / ai:tts / ai:ocr / ai:vision-segment, or the ai:* wildcard). See Authentication and API keys.

Compatibility and Base

Inference Space provides two compatible interfaces for LLMs, sharing the same account, key, and billing:

Interface	Base	Main endpoints
OpenAI-compatible	Domestic `https://ai.inf.space/v1`; overseas `https://ai.inf.space/v1`	`POST /v1/chat/completions`, `POST /v1/responses`
Anthropic-compatible	Domestic `https://ai.inf.space`; overseas `https://ai.inf.space`	`POST /v1/messages`
Image	Domestic `https://ai.inf.space`; overseas `https://ai.inf.space`	`POST /v1/images/generations`, `POST /v1/images/edits`

OpenAI-compatible requests require an explicit provider; otherwise they return 400. This supports Inference Space routing across multiple providers. See Quickstart.

Next steps

Quickstart: Make your first request with one curl command.
Authentication and API keys: API keys, Bearer headers, capability scopes, and authentication error codes.
Model discovery and catalog: Query available models and current prices.
Claude Messages API, GPT-Image 2 image generation and editing: Capability-specific API details.

Inference Space Overview