Embeddings & Qdrant

Embedding generation

• Uses configured embedding model/provider; chunk text via worker when possible.
• Sanitizes text to avoid control chars; fallback to main-thread chunking if worker unavailable.

Sync to Qdrant

• Provide host and optional apiKey; ensure collection vector size matches embedding dimensions.
• App can hint/update vector size; monitor sync status in Embeddings view.
• Sampling: load embeddings from Qdrant for visualization; paginate if large.
• Recommendation: pair generation + judge workflows with Qdrant retrieval to improve synthetic quality via context and similarity checks.

Dimension mismatch tips

• Mismatch errors mean the collection size differs from the embedding model output.
• Fix by recreating/updating the collection or switching embedding model to match.