Create vectors
The input field accepts one item or a batch. Retain the position of each input: a response index must map to your source document.
Do not mix vectors produced by different models or dimensions in one index without an explicit migration plan.
Embedding request
bash
curl --request POST https://cicora.ai/api/v1/embeddings \
--header "Authorization: Bearer $CICORA_API_KEY" \
--header 'Content-Type: application/json' \
--data '{
"model": "openai/text-embedding-3-small",
"input": ["First support note", "Second support note"],
"encoding_format": "float"
}'Indexing
- Normalize text the same way for indexing and search.
- Keep model id and dimension with each index.
- Split long documents into meaningful chunks before embedding.
Search and evaluation
Use cosine similarity or the metric required by your vector engine, then optionally apply rerank to a small candidate set.
Evaluate quality on real queries regularly. High numeric similarity does not guarantee a useful user answer.