AutoGraph Embed Field in Collection

Add embeddings to documents in any ArangoDB collection

Embed field in collection

POST https://<EXTERNAL_ENDPOINT>:8529/autograph/v1/embed-field-in-collection

Add embeddings to documents in any ArangoDB collection you already have. This path is independent of import, corpus build, clustering, and the {project}_CorpusGraph named graph.

Recommended path: This endpoint works independently; no import or corpus build required. Call once per (collection, field) pair. Every call recomputes the candidates from the live data, so rows that were inserted after a previous successful run are embedded on the next call.

Request

{
  "collection": "products",
  "field": "description"
}

Parameters

Parameter	Type	Required	Description	Recommended value
`collection`	string	Yes	Fully qualified logical name of an existing collection in the service database.	Your business collection name (e.g. `products`, `articles`). Must match ArangoDB naming rules.
`field`	string	Yes	Document attribute to embed. The service appends `_embedding` automatically (e.g. `description` → `description_embedding`).	A text-heavy attribute (`description`, `body`, `content`). Pass the source field name, not the embedding field.

Behavior

The service rescans the collection on every call and sorts the documents into three groups:

Candidates have a non-null source field and a missing or null <field>_embedding. An explicit null counts as not yet embedded. These are the documents that get embedded.
Skipped documents already have a non-null <field>_embedding and are left untouched.
Ineligible documents have no source value and no embedding. They are neither embedded nor counted as failed.

Source values may be string or numeric (coerced to text). Truncation follows the same rough character budget as corpus build. After a successful run, the service ensures a vector index on the embedding field and an ArangoSearch view on the source field when applicable.

Response

{
  "status": "completed",
  "message": "Embeddings generated for documents missing description_embedding.",
  "collection": "products",
  "field": "description",
  "embedding_field": "description_embedding",
  "documents_updated": 150,
  "documents_skipped": 20,
  "documents_examined": 175,
  "documents_failed": 2,
  "documents_ineligible": 3
}

Field	Type	Description
`status`	string	`"completed"`
`message`	string	Summary; may mention documents that could not be embedded
`collection`	string	Collection name
`field`	string	Source field
`embedding_field`	string	Name of the embedding attribute
`documents_updated`	integer	Documents that received embeddings in this run
`documents_skipped`	integer	Documents that already had a non-null `<field>_embedding` (unchanged by this call)
`documents_examined`	integer	The size of the collection. Reconcile it against your own store.
`documents_failed`	integer	Candidates that could not be embedded, because of an empty value or an error
`documents_ineligible`	integer	Documents with no source value and no embedding

The counts are expected to add up:

documents_examined == documents_updated + documents_skipped
                      + documents_failed + documents_ineligible

A gap is logged server-side at error level. The response still returns the counts, because the call does not abort mid-write only for a counter mismatch.

Status Code	Meaning
`200`	Success
`400`	Invalid `collection` or `field`
`401`	Authentication failed
`404`	Collection does not exist
`500`	Server error

HTTP Example

curl -X POST \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <token>" \
  -d '{"collection": "products", "field": "description"}' \
  https://<EXTERNAL_ENDPOINT>:8529/autograph/v1/embed-field-in-collection

Next Steps

Corpus Build: Learn about automatic embedding generation during builds