按问题复杂度自动分流的电商客服 Agent:简单问题直接回答;复杂问题经过多跳检索并自我纠错;查单、退改通过工具调用完成,且高风险写操作必须先经人工确认。基于 FastAPI 和 LangGraph,混合检索并附带引用,内置离线评测与 Prometheus 指标。
🐢 Open-Source Evaluation & Testing library for LLM Agents
Open-source, end-to-end platform for evaluating, observing, and improving LLM and AI agent applications. Tracing · Evals · Simulations · Datasets · Gateway · Guardrails. Self-hostable. Apache 2.0.
Agentic RAG for local and self-hosted document search: hybrid retrieval, reranking and multimodal RAG on embedded LanceDB, with Docling parsing and an MCP server
中文优先的企业 RAG 知识库:可控解析、治理、切块、混合检索、重排、引用、图谱、评测与 Dify 接入。