Schema validation fails on FirestoreRetriever

[sisbell]

I am trying to use a retriever defined below

export const suggestionsRetriever, = defineFirestoreRetriever({
    name: 'suggestionsRetriever',
    firestore,
    collection: 'options',
    contentField: 'data',
    vectorField: 'embedding',
    embedder: textEmbeddingGecko,
    distanceMeasure: 'EUCLIDEAN',
});

const suggestions = await run(
    'Retrieve options',
    async () => {
        const docs = await retrieve({
            retriever: suggestionsRetriever,
            query: input.query,
            options: {
                limit: 5,
            },
        });
        return docs;
    }
);

The paths in firestore are options/{id}/data and options/{id}/embedding .

Data is a map of field/values (not a string)

The retriever does properly match but it is unable to return the content.

The error seems to indicate that it must be a string and have a media property. How do I configure the query to return non-string data?

Error: INVALID_ARGUMENT: Schema validation failed. Parse Errors:

- documents.0.content.0.text: must be string
- documents.0.content.0: must have required property 'media'
- documents.0.content.0: must match a schema in anyOf

Provided data:

{
  "documents": [
    {
      "content": [
        {
          "text": {//This is my data node, which I would like to return in the query
            "firestoreId": "s33a-83f1-e053-d3sad0ad83c",
            "propertyId": "5ddsa3f1c9-8a",
             // Other fields
          }
        }
      ],
      "metadata": {}
    },

[sisbell]

The following is how I would do a search without a GenKit defineFirestoreRetriever. It properly returns the data field.

async function getVectorSearchResultsData(snapshot: firestore.VectorQuerySnapshot): Promise<any[]> {
    const resultsData = snapshot.docs.map(result => {
        const documentData = result.data();
        return documentData.data;
    });

    return resultsData;
}

const coll = firestore.collection('options');
const q = await createEmbedding(input.query);
const preFilteredVectorQuery: VectorQuery = coll
    .findNearest("embedding", q, {
        limit: 5,
        distanceMeasure: "EUCLIDEAN",
    });

const vectorQueryResults = await preFilteredVectorQuery.get();
const resultsData = await getVectorSearchResultsData(vectorQueryResults);
console.log(JSON.stringify(resultsData, null, 2));