Knowledge Base
Knowledge Settings
Configure chunk sizes, similarity thresholds, top-K retrieval limits, and re-crawl frequencies.
Overview
In Project > Settings > Knowledge, engineers can fine-tune every hyperparameter of the RAG pipeline.
Retrieval Parameters
| Parameter | Default | Recommended Range | Description |
|---|---|---|---|
| Top-K Chunks | 4 | 3 – 6 | Number of passages injected into prompt |
| Similarity Threshold | 0.72 | 0.65 – 0.80 | Minimum score required to accept chunk |
| Hybrid Search Alpha | 0.7 | 0.5 – 0.8 | Weighting between vector (1.0) and keyword (0.0) |
Chunking & Embedding Options
- Chunk Size: 512 tokens
- Chunk Overlap: 64 tokens
- Preserve Code Blocks: True (keeps functions intact)
Crawler & Refresh Policies
- Max Crawl Pages: 2,500 pages per project (configurable on Enterprise plans).
- Auto-Sync Frequency: Daily / Weekly / Off.