Developer Guide Overview

Applicable role: Developer Last updated: 2026-08-06

The platform provides OpenAI Compatible, Messages Compatible, and Gemini Native compatible interfaces. Projects that already use a compatible SDK typically only need to adjust the Base URL, API Key, and model call name.

Connection Details

Configuration Value
OpenAI Compatible Base URL https://<your-api-endpoint>
Chat Completions https://<your-api-endpoint>/v1/chat/completions
Messages Compatible Base URL https://<your-api-endpoint>
Messages https://<your-api-endpoint>/v1/messages
API Key An sk- prefixed key created in the console
Model call name Copy from the "Call Guide" section in the console

For your first API call, start with Quick Start.

Key Configurations in the Call Chain

Configuration Purpose
API Key Identifies the user and applies group, quota, rate, and source restrictions
Base URL Determines which API service receives the request
model Specifies the model; can also include a group identifier
Key-bound group The group used preferentially for plain model names
Default model configuration Provides candidate groups when the bound group does not have the target model

Examples of plain call names include deepseek-v4-flash and kimi-k3. When group identifier routing is enabled, you can also use the full call name shown in the console, such as 5MHXZWKA/deepseek-v4-flash.

Whether a model can be called also depends on the organization's groups, member user permissions, and account status. For detailed rules, see Model Availability and Group Authorization and Call Guide and Routing.

Choosing a Protocol

Protocol Endpoint Best for
OpenAI Compatible /v1/chat/completions Most SDKs, chat clients, and workflow tools
Messages Compatible /v1/messages SDKs and agent tools that use the Messages request structure
Gemini Native /v1beta/models/{model}:generateContent etc. Clients using the Gemini SDK or native Gemini API format

The three protocols differ in request fields and streaming events. Choose the protocol that your client actually supports; do not mix fields from different protocols in the same request.

Documentation Navigation

Need Document
Complete your first API request Quick Start
View parameters, authentication, and streaming output API Call Basics
Use Python, JavaScript, and other SDKs SDK Integration
Connect Claude Code, Cursor, and other tools Third-Party Tool Integration
Handle 401, 403, 404, 429, and 5xx errors Error Codes and Troubleshooting
Understand model call names and routing Call Guide and Routing
Improve Chinese output stability Chinese Output Stability
Quickly look up common questions Developer FAQ

Whether tool calling, image input, prompt caching, and structured output are available depends on the specific model and protocol. Before going live, you should validate with your target model on a small scale and prepare fallback handling for unsupported capabilities.