Deep Knowledge Exploration Template#
The Deep Knowledge Exploration Template goes one step further than the standard Knowledge Exploration Template.
It is a template for designing an advanced RAG agent that explores and reasons step by step by actively determining the reasoning flow based on the query context and deciding what knowledge and search strategy are needed.Active search strategy: The agent determines the search query, search scope, and search method on its own in each round to derive the most effective results
Improved accuracy through iterative reasoning: Instead of ending with a single search, the agent repeats exploration and evaluation across multiple rounds to improve the completeness of the answer
Support for complex queries: Suitable for analytical, comparative, and synthesis-based questions that require cross-referencing multiple documents
Flexible control of exploration: The number of search rounds, allowed time, and search scope can be adjusted depending on the situation
For simple Q&A or FAQ-style use cases where the answer can be found in a single document, we recommend the Knowledge Exploration Template. For complex questions that require cross-referencing multiple documents or comprehensive analysis, we recommend the Deep Knowledge Exploration Template.
Differences from the Knowledge Exploration Template#
The existing Knowledge Exploration Template operates in a simple flow of single search → response generation for a user query. In contrast, the Deep Knowledge Exploration Template allows the agent to combine multiple search skills and perform iterative exploration over several rounds, then synthesize the collected evidence to generate the final answer.| Item | Knowledge Exploration | Deep Knowledge Exploration |
|---|
| Search method | Single search with one query | The agent re-determines the search strategy in each round |
| Search skills used | Knowledge search | Automatically combines multiple skills such as semantic chunk search, keyword-based file search, context expansion, and web search |
| Depth of reasoning | Uses search results directly | Repeats exploration and reasoning over multiple rounds |
| Suitable question types | Simple Q&A, FAQ | Analytical and comparative questions that require synthesizing multiple documents |
| Response time | Fast | Relatively slower depending on the depth of exploration |
| Token cost | Low | Increases in proportion to the number of rounds |
