Model Definition

The first step on the model platform: register vendors and model instances.

Two-Layer Structure

Vendor                    Model instance
─ OpenAI                  ─ gpt-4-turbo
─ Anthropic               ─ gpt-4o
─ Google                  ─ claude-3-5-sonnet
─ Azure                   ─ gemini-1.5-pro
─ Aliyun / Tencent / ByteDance ─ qwen-max
─ Self-hosted             ─ deepseek-v3 (local)
─ OpenAI-compatible       ─ ...

Each vendor has multiple model instances under it.

6 Capability Types

Type	Use
LLM	Chat · reasoning · function call
Embedding	Text/image vectorization (knowledge bases)
Reranking	Re-order retrieval results
Image	Image generation
Audio	Speech synthesis / recognition
Video	Video generation

Default model configuration supports per-type defaults.

Core Model Fields

Field	Example
Name	`gpt-4-turbo`
Type / capability	LLM / Embedding / ...
Sampling	Temperature / TopP / max length
Context length	128K
Metadata	Description / icon / default prompt

Detail Page

In the model detail you can:

View / edit name, description, icon
Tune LLM sampling (temperature 0-2 / TopP 0-1 / max output length)
Configure Embedding dimension and normalization
Configure Reranking TopN

Next Steps

Connect SaaS APIs → Interface platform (set Base URL + API Key)
Self-host models → Model deployment
Set org default → Default model configuration