Summary

Super Agent is an enterprise-grade AI agent platform for Java developers, offering comprehensive capabilities beyond simple API calls. It addresses intelligent dialogue, document Q&A, web search, RAG, tool integration (MCP), and skill extensibility, providing deep engineering insights for building robust AI systems and excelling in technical interviews.

How It Works

The system employs a layered architecture with a dialogue orchestration core and a three-layer executor system, prioritizing deterministic responses (ambiguity, RAG) before resorting to open-ended ReAct Agents. Key innovations include dual-channel hybrid retrieval (vector/keyword) with RRF fusion, Parent-Child chunk aggregation, and a Neo4j-driven document structure graph. Advanced RAG components like knowledge routing, query rewriting, evidence control, and hallucination short-circuiting are engineered.

Quick Start & Requirements

Requires Java (JDK), Spring Boot, and middleware: PostgreSQL+PGVector, Elasticsearch, Neo4j, MySQL, Redis, Kafka, MinIO. Apache Tika handles document parsing. Frontend uses Vue 3/Vite. Setup involves deploying backend services and configuring API keys for external LLM/search providers (e.g., Aliyun DashScope, Tavily). Specific setup guides are referenced but not directly provided.

Highlighted Details

• Three-Layer Executor System: Intelligently routes requests to ambiguity clarification, RAG Q&A, or ReAct Agent based on intent.
• Advanced RAG Pipeline: Features multi-stage knowledge routing, query rewriting, dual-channel hybrid retrieval (vector/keyword) with RRF fusion, and Parent-Child chunk aggregation.
• Extensibility: Supports dynamic tool integration via MCP and declarative skill expansion through SKILL.md files.
• Neo4j Document Structure Graph: Builds a document hierarchy graph for structural navigation.
• Observability & Safety: Includes full-link tracing, management console, evidence budget control, and no-evidence short-circuiting for hallucination prevention.

Maintenance & Community

The project is primarily author-driven, with community support via WeChat for questions and learning materials. Updates are announced via the "阿星不是程序员" WeChat account. No explicit community forums like Discord/Slack are mentioned.

Licensing & Compatibility

The specific open-source license is not stated in the README, preventing definitive assessment of commercial use compatibility or usage restrictions.

Limitations & Caveats

The unspecified license is a significant blocker for commercial adoption. Setup requires considerable effort due to numerous middleware dependencies. Direct links to online demos and detailed documentation are referenced but not included. The system relies on external LLM/search services, necessitating API key configuration and incurring potential costs.