Indexing documents

Two paths are available for pushing documents into a Flexible Vector Database: index-direct for small batches and index-jobs for larger ones.

index-direct

Synchronous path. Send a batch of documents in the request body; the platform encodes, embeds, and writes them in one call. The response includes the count of indexed documents. Use this for small batches (a few hundred at a time) where you want a synchronous confirmation.

index-jobs

Asynchronous path. Submit a job; the platform returns a job id that you can poll. Use this for large batches or for scheduled reindexes that should not block the caller.

Document payload

Each document has a stable id, free-text content, optional metadata, and an optional source block. Documents with acontent_image_url field are encoded by the multimodal embedding model when the backing database supports it.

index-payload.json

1{
2  "documents": [
3    {
4      "id": "doc-1",
5      "content": "QDivZero Compute is the managed inference layer.",
6      "metadata": { "section": "compute" },
7      "source": { "text": "QDivZero Compute is the managed inference layer." }
8    },
9    {
10      "id": "doc-2",
11      "content": "Vector databases support multimodal embeddings.",
12      "content_image_url": "https://cdn.example.com/diagram.png",
13      "metadata": { "section": "vector-databases" }
14    }
15  ]
16}