Feature: Retrieval Trace View for RAG Debugging
Is your feature request related to a problem? Please describe.
When a workflow returns an incorrect or incomplete answer, it can be difficult to determine where the failure occurred.
For knowledge-base applications, users often need visibility into:
- Which chunks were retrieved
- Which chunks were selected after reranking
- Which sources were ultimately provided to the model
Without this information, debugging retrieval quality becomes largely trial-and-error.
Describe the solution you'd like
Add an optional "Retrieval Trace" view for workflow executions.
For each query, display:
- Retrieved chunks
- Retrieval scores
- Reranked results
- Final context sent to the LLM
- Source attribution
This could be available in execution logs or a dedicated debugging panel.
Describe alternatives you've considered
External observability tools can provide partial visibility, but they often lack workflow-specific retrieval information and require additional setup.
Additional context
This feature would make it significantly easier to debug chunking strategies, retrieval quality, reranking performance, and knowledge-base coverage issues.
Source: labring/FastGPT