Semantic search

Query your collections using natural language or pre-computed vectors to find semantically similar documents.

Text query mode

Provide a natural language query and the server generates the embedding automatically, then searches for similar documents.

{
  "query": "How do I configure SSL certificates?",
  "limit": 5,
  "threshold": 0.7
}

Vector query mode

Provide a pre-computed embedding vector for direct similarity search. Use this when you’ve already generated embeddings in your application.

{
  "vector": [0.1, 0.2, 0.3, "..."],
  "limit": 5,
  "threshold": 0.7
}

Search parameters

Parameter	Required	Default	Description
query	Yes*	—	Natural language search query
vector	Yes*	—	Pre-computed embedding vector
limit	No	5	Maximum results to return (1—20)
threshold	No	0.7	Minimum similarity score (0—1)
includeEmbeddings	No	false	Include embedding vectors in results

* Use either query or vector, not both.

Response format

Results include the document content, similarity score, and metadata:

{
  "results": [
    {
      "documentId": "doc-123",
      "content": "To configure SSL certificates...",
      "similarity": 0.92,
      "metadata": { "source": "docs", "page": "/ssl-setup/" }
    }
  ]
}

Tuning the threshold

The threshold parameter controls the minimum similarity score for returned results.

• Lower values (0.5) — return more results with less relevance
• Higher values (0.9) — return only highly relevant matches
• Default (0.7) — a balanced starting point for most use cases

Start with the default and adjust based on the quality and quantity of your results.

Using with agents

Assign collections to agents so they can search your knowledge bases during conversations. This enables retrieval-augmented generation (RAG), grounding agent responses in your data rather than relying solely on the model’s training data.

Next steps

• Creating collections — set up new collections and choose embedding models
• Uploading documents — add content to your collections