An interactive console for testing multi-Knowledge-Base Q&A.
The Playground is an interactive debugging panel for Knowledge Bases. Select one or more Knowledge Bases, type a question, and watch the retrieved Chunks and the model's answer in real time.
What you can do
| Feature | Description |
|---|---|
| Aggregated multi-KB Q&A | Select up to 15 Knowledge Bases at once; the platform routes retrievals automatically. |
| Model switching | Compare Qwen-series models, DeepSeek, and others side by side. |
| Citation highlighting | Citations in the answer link back to the original Chunk. |
| Knowledge source visualization | Shows hit-document cards and per-source counts. |
Steps
1
Open the Playground
Sign in to the console and click Playground in the left navigation.
2
Select Knowledge Bases
In the left panel, click Add Knowledge Base and check the ones you want to search. You can pick multiple.
3
Ask a question
Type a question in the input box at the bottom and press Enter. The platform retrieves relevant Chunks first, then asks the LLM for an answer.
4
Review the result
The answer area shows:
- The model-generated answer text.
- Citation markers (such as [1], [2]) — click to expand the source.
- The right panel lists the matched Chunks with relevance scores.
Tips
- Check chunk quality — see whether the returned Chunks contain the correct answer. If recall is low, tune your Chunking Strategy.
- Compare models — switch models to compare answer quality and pick the best fit for your workload.
- Multi-KB priorities — for multi-KB setups, configure per-KB weights in Knowledge Retrieval.
Compared with the Knowledge Base detail page
| Entry point | Scope | Best for |
|---|---|---|
| Playground | Multiple KBs | Cross-KB Q&A. |
| KB detail page → Knowledge Retrieval | Single KB | Testing recall on one Knowledge Base. |
| KB detail page → Knowledge Q&A | Single KB | Testing Q&A on one Knowledge Base. |