Sakura AI Engine – Dify Plugin

日本語

A Dify model provider plugin that connects Sakura AI Engine to Dify via the OpenAI-compatible API.

Features

• LLM & Embedding — supports both large language models and text embedding models
• Predefined models — ready to use out of the box
• Custom models — add any model available on Sakura AI Engine
• Streaming — full streaming support with reasoning content (reasoning_content) handling
• OpenAI-compatible — standard API format, no extra adapters needed

Supported Models

LLM

Model	Context
gpt-oss-120b	128K
Qwen3-Coder-480B-A35B-Instruct-FP8	128K
Qwen3-Coder-30B-A3B-Instruct	128K
llm-jp-3.1-8x13b-instruct4	128K

Preview models

Model	Context
Phi-4-mini-instruct (CPU)	16K
Phi-4-multimodal-instruct	128K
Qwen3-0.6B (CPU)	32K
Qwen3-VL-30B-A3B-Instruct	128K

Text Embedding

Model	Dimensions	Context
multilingual-e5-large	1024	512
Qwen3-Embedding-4B-FP16 (Preview)	2560	8192

Quick Start

1. Install

Download the .difypkg file from Releases and install it in your Dify instance.

2. Configure

1. Go to Settings → Model Providers in Dify
2. Find Sakura AI Engine and click configure
3. Enter your API key in UUID:secret format

3. Use

Select any predefined model in your Dify app, or add a custom model by entering its model ID.

Configuration

Enter your API key in UUID:secret format — that's all you need. The plugin handles everything else automatically.

Project Structure

├── manifest.yaml                  # Plugin metadata
├── main.py                        # Entrypoint
├── provider/
│   ├── sakura-ai-engine.yaml      # Provider definition
│   └── sakura-ai-engine.py        # Credential validation
├── models/
│   ├── llm/                       # LLM models
│   │   ├── llm.py
│   │   ├── *.yaml
│   │   └── preview/*.yaml
│   └── text_embedding/            # Embedding models
│       ├── text_embedding.py
│       ├── *.yaml
│       └── preview/*.yaml
└── references/                    # API specs & docs

Requirements

• Dify ≥ 1.11.1
• Python 3.12
• Sakura AI Engine API key

Security

• Never commit your API key to the repository
• Store credentials only in Dify's credential fields
• Rotate the key immediately if exposed

License

MIT