Playground KDB

Use Playground KDB to test how well your Knowledge DB retrieves relevant content before connecting it to an agent.

Open Playground

1. Go to Knowledge DB.
2. Select a KDB from the list.
3. Click the vertical ellipsis at KDB card and select Playground action to open the test screen.

Playground Layout

At the top, you can review KDB metadata such as:

• KDB name and status
• Vector DB
• Number of files
• Total size
• Embedding model
• Chunking size

Main areas:

• Search input at the bottom
• Search Results table in the center
• Retrieval controls at the top-right

Retrieval Controls

Configure retrieval quality before searching:

• Top-K: Number of initial chunks to retrieve.
• Reranker: Enable to re-rank retrieved chunks for better relevance.
• Top-N: Number of final chunks shown after reranking.

Recommended start:

• Top-K: 10
• Reranker: On
• Top-N: 5

Run a Search Test

1. Enter a test question in the input box (for example: testDb).
2. Press Enter.
3. Review returned rows in Search Results.

Understand Search Results

Each row shows:

• #: Rank position
• Content: Retrieved text chunk
• Document: Source file identifier
• Reference: Source title/section
• Relevance Score: Similarity score and visual bar

Higher relevance score means better match to your query.

Tuning Tips

• If results are too broad: lower Top-K or Top-N.
• If important text is missing: increase Top-K.
• If ranking quality is weak: keep Reranker enabled.
• If chunks look fragmented: adjust chunking settings in KDB configuration and re-test.

Exit Playground

Click the close icon (X) at the top-right to leave Playground and return to KDB list.