IntelliSearch V3.1: Unifying Search, Empowering Action

IntelliSearch began as a simple search agent based on the MCP (Model Context Protocol) protocol, with the vision of evolving into a lightweight, decoupled, and extensible Agentic Infrastructure and ecosystem foundation. It integrates agent topology, multi-dimensional internal context memory and external document management, dynamic external tool scheduling and environmental interaction mechanisms, as well as multi-agent communication protocols, providing developers with a framework that balances ease of use and flexibility.

IntelliSearch-v3.1

IntelliSearch-v3.1 (SJTU AI - Intelligent Search) is the second model release in the IntelliSearch agent series. Through the MCP protocol, it achieves integration of multi-dimensio nal, multi-source high-quality information sources and tools, while providing a simple sequential context memory module that significantly expands the boundaries and exploration capabilities of language models. IntelliSearch integrates numerous high-quality MCP tools, including:

Search Tools:

• Web Search (Google Search, Zhipu AI Search, Web Content Parser)
• GitHub Search - Repository, code, user, Issue, and PR search
• Academic Search (Google Scholar, arXiv latest papers)
• Geographic Information Search (Amap API - route planning, geocoding, POI search)
• Bilibili Video Search
• Douban Movie/Book/Review Search
• 12306 Train Information Query
• WeChat Official Account Article Search
• Local Semantic Search (RAG - supports PDF, TXT, MD, DOCX)
• SAI Memos Knowledge Base Search

Operation Tools:

• Browser Automation (Playwright - web navigation, interaction, content extraction)
• File System Operations (create, read, write, delete, supports CSV/PDF/JSON)
• Python Code Execution (IPython backend - state persistence, result capture)
• Terminal Command Execution (timeout control, output capture)
• Basic Tool Kit (date/time, UUID, random numbers, and other utilities)

Developer Guide

See DEV_SETUP for details

IntelliSearch-v3.1 BackBone

To support the evolution of IntelliSearch-v3.1 into more personalized and flexible agent module designs, we implemented a version-level project refactoring and update (IntelliSearch-v3.1 BackBone). This aims to build a lightweight yet efficient layered agent module design, providing infrastructure support for upper-level applications.

Design Philosophy

Adopts a layered architecture design that clearly separates system responsibilities into the following layers:

• Core Layer (core/): Defines abstract base classes and data models

  • BaseAgent: Abstract base class for all Agents
  • AgentFactory: Agent factory pattern implementation
  • AgentRequest/AgentResponse: Unified request/response models

• Agent Layer (agents/): Concrete Agent implementations

  • MCPBaseAgent: Main Agent with integrated MCP tools

• Memory Layer (memory/): Conversation context management & external knowledge base component management

  • BaseMemory: Memory abstraction interface
  • SequentialMemory: Linear context management implementation

• Tools Layer (tools/): Tool invocation interfaces based on MCP protocol communication & environment simulation interfaces

  • MCPBase: MCP tool communication component
  • MultiServerManager: MCP server lifecycle management

• UI Layer (ui/): Unified CLI user interface components