英文原文
Glean introduces Waldo, the first agentic search model built on NVIDIA Nemotron 3 Nano, specifically post-trained for search planning.
Waldo can autonomously break down queries, decide which tools to call, which content to read, and when to stop and return results — enabling truly agentic enterprise search. Unlike traditional RAG, Waldo has proactive planning capabilities, making it suitable for knowledge-intensive enterprise scenarios.
Key differentiators from traditional RAG:
- Proactive query decomposition and planning
- Autonomous tool selection and execution
- Self-assessment of evidence sufficiency
- Multi-step reasoning with adaptive depth
中文翻译
Glean 推出 Waldo，首个 Agentic 搜索模型，基于 NVIDIA Nemotron 3 Nano 构建，专门针对搜索规划做后训练。
Waldo 能自主分解查询、决定调用哪些工具、阅读哪些内容、何时停止并返回结果，实现真正代理式企业搜索。与传统 RAG 不同，Waldo 有主动规划能力，适合知识密集型企业场景。
与传统 RAG 的关键区别：
- 主动的查询分解和规划
- 自主的工具选择和执行
- 自我评估证据充分性
- 具有自适应深度的多步推理
摘要
Glean 推出 Waldo，首个 Agentic 搜索模型，基于 NVIDIA Nemotron 3 Nano 构建，专门针对搜索规划做后训练。Waldo 能自主分解查询、决定调用哪些工具、阅读哪些内容、何时停止并返回结果。