Agent Hub Model Management

In the VJSP intelligent development ecosystem, Models serve as the capability engine of Agents, directly determining their upper limits in code understanding, generation, dialogue, tool invocation, and other functionalities.

⚠️ Important Prerequisite
Models cannot be used independently. They must be bound to an Agent before they can be invoked within a plugin.
A single Agent can bind multiple models and automatically switch between them based on their assigned roles (e.g., using Model A for Chat and Model B for Edit).

VJSP supports two types of model sources:

• System-preconfigured Models (ready-to-use out of the box)
• Locally Customized Models (privately deployed, configured via config.yaml)

System-Preconfigured Models

The platform comes with a set of official system models that require no additional setup and are ready for immediate use. These models are maintained by VJSP.

Features

• Pre-assigned Roles: Each system model already has appropriate roles (such as chat, edit, autocomplete) pre-configured to ensure optimal performance across different scenarios.

• Automatic Capability Detection: The system automatically detects whether a model supports advanced features like tool_use (tool calling) or image_input (image input).

• Security Isolation: All API requests to system models are routed through the VJSP secure proxy.

💡 Tip: You can view currently available system models in your plugin’s model selector and switch between them at any time. For higher customization (e.g., private deployment, multi-model collaboration), please refer to the Local Model config.yaml Reference Documentation.

Pre-installed System Models

Agent Hub Model Usage Guide

Agent Hub is the official shared platform for high-quality model configurations, rule sets, and prompt templates contributed by the community. You can browse and select models suited to your development scenarios at Agent Hub.

Models form the core foundation of the entire Agent interaction experience and provide several specialized capabilities:

• Chat: Supports interactive communication around code and offers detailed technical guidance.
• Edit: Handles complex code transformation and refactoring tasks.
• Apply: Precisely executes targeted code modification operations.
• Autocomplete: Provides real-time code suggestions during development.
• Embedding: Converts code into vector representations to enable semantic retrieval.
• Reranker: Ranks retrieved results based on semantic relevance to improve retrieval accuracy.

AgentHub

Recommended Models

High-Quality Models Categorized by Functionality

Model Function	Recommended Model	Description
Coder	Qwen3-Coder-30B-A3B-Instruct-VJSP	Specializes in handling and understanding development requirements related to the VJSP full-stack framework—such as framework component invocation, front-end/back-end interaction logic, business feature implementation, configuration file writing, and full-stack deployment strategies. It is not a general-purpose coding model but is deeply focused on full-stack coding and adaptation tasks within the VJSP development workflow, providing foundational coding and engineering implementation capabilities for VJSP projects.
	Qwen3-Coder-30B-A3B-Instruct	Specializes in processing and understanding software engineering–related code and development context—such as full-repository code, technical documentation, API definitions, refactoring requirements, and automation scripts. It is not a general-purpose language model but is deeply optimized for developer workflows involving code generation and agent tasks, delivering efficient, long-context coding and reasoning capabilities for enterprise-grade development scenarios (e.g., full-repo analysis, code generation, tool calling).
VL	Qwen2.5-VL-32B-Instruct	Specializes in processing and understanding image data relevant to software engineering—such as web page screenshots, mobile UI mockups, flowcharts, form prototypes, and system error interfaces. It is not a general-purpose image recognition model but is specifically designed for visual semantic parsing tasks within developer workflows, providing foundational perception and reasoning capabilities for VJSP-series agents (e.g., Visual Agent, Form Generator, Testing Assistant).
Reranker	VJSP-Reranker	VJSP Reranker is a specialized AI model designed for fine-grained re-ranking. It performs deep semantic relevance evaluation and reorders candidate results returned from initial retrieval methods (e.g., keyword matching or vector similarity search). It does not handle initial filtering but acts as a “second intelligent filter” to ensure the most relevant and contextually appropriate answers or code snippets appear at the top.
Embedding	VJSP-Embedding	An AI model specifically engineered for high-precision semantic representation, multilingual support, and efficient retrieval scenarios.
Apply	VJSP-Apply	A lightweight AI model optimized specifically for rapid code merging functionality.

Using Official VJSP Models

• Browse the existing model list and view system model cards on the Agent Hub page.
• View model details: Click any model card to access its detail page and review its description.
• Click “Use Now” and select the Agent you wish to associate it with.

Prerequisite: You must be logged into the system. If not, please log in first.
You can only associate models with Agents you have created yourself.