POST /v1/embed converts chunks into vectors, the same as the SDK embed(). Each returned chunk gains an embedding field. Pass the result to /v1/store.
Request
curl -X POST http://127.0.0.1:8000/v1/embed \
-H "Content-Type: application/json" \
-d '{
"chunks": [
{ "id": "guide-0", "text": "Ragrails builds modular RAG workflows...", "source": "files/guide.pdf", "metadata": {"title": "Guide"} }
],
"provider": "voyage",
"model": "voyage-3",
"input_type": "document",
"batch_size": 64
}'
{
"inputs": 1,
"embedded": 1,
"items": [
{ "id": "guide-0", "text": "...", "embedding": [0.0132, -0.0719, 0.0241], "source": "files/guide.pdf", "metadata": {"title": "Guide"} }
],
"failed": 0,
"errors": []
}
Use
input_type: "document" when embedding content to store, and "query" when embedding a search query. Mismatching them hurts retrieval quality. Use the same model for indexing and querying.Request fields
| Field | Default | Description |
|---|---|---|
chunks | required | List of chunk dicts; each needs id and text |
provider | "voyage" | Embedding provider |
model | "voyage-3" | Model name |
input_type | "document" | "document" for indexing, "query" for retrieval |
batch_size | 64 | Chunks per embedding request |
options | null | Provider-specific options |
Response fields
| Field | Description |
|---|---|
inputs | Chunks passed in |
embedded | Chunks successfully embedded |
items | Chunk dicts with an added embedding vector |
failed | Failed chunks |
errors | Error objects |