Developer Guide Overview

Applicable role: Developer Last updated: 2026-07-14

The platform provides APIs that are fully compatible with the OpenAI / Anthropic / Gemini ecosystems. If your code already uses the OpenAI or Claude SDK, you usually only need to change two configuration lines (Base URL + API Key) to switch to the platform — a single integration lets you call the world's mainstream models.

1. Integration in Just Three Steps

① Get an API Key      → Created by the organization or a sub-user on the console's "API Keys" page
② Configure Base URL  → Point requests to the platform
③ Send requests       → Call using the familiar OpenAI/Anthropic format

For detailed steps, see Quickstart: Your First Call in 5 Minutes.

2. Core Concepts

Understanding the following concepts means understanding the entire call chain:

Concept	What It Is	Analogy
API Key	Your identity credential, starting with `sk-`	Access card
Base URL	The platform's API address	Server street address
Model Call Name	The value you put in the `model` field of a request	Which dish you order
Group (group)	A label dimension for models with the same name, used for disambiguation and routing	Your menu section

Regarding the model call name, v2 offers two formats:

Model name, e.g. claude-sonnet-4-6, gpt-4o: Resolved and routed by priority (Key-bound group → default routing config), compatible with the old behavior.
With group identifier, e.g. 5MHXZWKA/gpt-4o: Routed to the specified group by "calling out" the group identifier (requires group-identifier routing to be enabled).

Model names can be obtained from the console's "Call Guide", the Model Marketplace, or the /v1/models API. After enabling group-identifier routing, the Call Guide page also displays call names with group identifiers. For detailed explanations and scenario selection, see Call Guide and Routing.

Which groups / models you can call depends on the range of models available to your account (configured by the platform per organization). If you can't reach a certain model, first confirm the available range — see Administrator Manual · Regions and Model Ranges.

3. Protocol Formats

The platform is compatible with several mainstream protocols; you can choose whichever you prefer:

Protocol	Endpoint	Best For	Example Docs
OpenAI format	`/v1/chat/completions`	Using the OpenAI SDK, most frameworks	API Basics
Anthropic format	`/v1/messages`	Using the Anthropic SDK, calling Claude	API Basics
Gemini format	`/v1beta/models/*`	Using the Google Gemini SDK	API Basics

All formats call the same set of models; the only difference is the field structure of the request/response. The same model can be called using either the OpenAI format or the Anthropic format.

4. Choose Docs by Your Scenario

Your Situation	Recommended Path
First time using it, want to get it running first	Quickstart
Want to figure out exactly what to put in the `model` field	Call Guide and Routing
Have existing OpenAI/Claude code and want to migrate	SDK Integration
Using tools like Claude Code / Cursor	Third-Party Tool Integration
Want to learn about request parameters and streaming output	API Basics
Encountering errors	Error Codes and Troubleshooting
Model occasionally outputs Korean/Japanese	Claude Language Drift Notes
Want to quickly find the answer to a question	Developer FAQ

5. Advanced Usage

The following capabilities are already supported; dedicated docs are coming soon:

Prompt Caching (supported, docs in progress)
Tool Calls / Function Calling (supported, docs in progress)
Multimodal (image input) (supported, docs in progress)
Structured Output (supported by some models) (supported, docs in progress)