What is the difference between Vertex AI Vector Search and Vertex AI Search?

Vertex AI Vector Search (this guide) is a low-level approximate nearest neighbor (ANN) service — you manage embeddings, indexes, and queries directly. It is optimized for high-throughput similarity search on pre-computed embeddings. Vertex AI Search is a higher-level managed search service with built-in document processing, web crawling, and semantic search. Use Vertex AI Vector Search when you control embedding generation; use Vertex AI Search when you want Google to handle document ingestion and ranking.

Why is there no text stored alongside the vectors?

Vertex AI Vector Search stores only vector embeddings — there is no text field in the index. The text field in Gateco search results will always be empty for this connector. To associate document content with policy-filtered results, Gateco uses sidecar metadata (the default metadata_resolution_mode: sidecar), which reads from Gateco's own registered resource metadata. Ensure you ingest documents via the Gateco API (or retroactive registration) so the sidecar metadata is populated.

What does the public_endpoint_domain field look like?

It is the hostname shown on your deployed index endpoint's detail page — something like 123456789.us-central1-abc.vdb.vertexai.goog. Pass only the hostname without https://. Do not use the management API endpoint (us-central1-aiplatform.googleapis.com) — that is only used for health checks.

Connect Vertex AI Vector Search to Gateco

Vertex AI Vector Search (formerly Vertex AI Matching Engine) is Google Cloud's managed ANN service for high-scale vector similarity search. Gateco connects to Vertex AI Vector Search as a Tier 2 vector-only connector — adding identity-aware retrieval and RAG access control to your Vertex AI indexes. Note that keyword, hybrid, and grep search are not available because Vertex AI Vector Search does not store text alongside vectors.

Vertex AI Vector Search stores embeddings only — no text content. Keyword, hybrid, and grep search modes are not supported. Use sidecar metadata resolution (the default) to associate document content with policy-gated results.

Step 1 — Enable Vertex AI API and create an index

bash

gcloud services enable aiplatform.googleapis.com --project=YOUR_PROJECT_ID

# Create a Vector Search index (underscore in display name required, not hyphens)
gcloud ai indexes create \
  --display-name=lcf_index \
  --description="Gateco vector search index" \
  --metadata-schema-uri=gs://google-cloud-aiplatform/schema/matchingengine/metadata/nearest_neighbor_search_1.0.0.yaml \
  --region=us-central1 \
  --project=YOUR_PROJECT_ID

Index display names cannot contain hyphens. Use underscores (e.g. lcf_index, not lcf-index). This is a Vertex AI API constraint.

Step 2 — Deploy the index to an endpoint

In GCP Console, navigate to Vertex AI → Vector Search → Index Endpoints.
Click Create endpoint.
Select Public endpoint (required for Gateco).
After the endpoint is created, click Deploy index and select your index.
Note the Deployed Index ID (no hyphens allowed).
Wait for the deployment status to become DEPLOYED (~5–10 minutes).

Step 3 — Create a service account

bash

gcloud iam service-accounts create gateco-vertex \
  --display-name="Gateco Vertex AI" \
  --project=YOUR_PROJECT_ID

gcloud projects add-iam-policy-binding YOUR_PROJECT_ID \
  --member="serviceAccount:gateco-vertex@YOUR_PROJECT_ID.iam.gserviceaccount.com" \
  --role="roles/aiplatform.user"

gcloud iam service-accounts keys create gateco-vertex-key.json \
  --iam-account=gateco-vertex@YOUR_PROJECT_ID.iam.gserviceaccount.com

Step 4 — Get the public endpoint domain

In GCP Console, go to Vertex AI → Vector Search → Index Endpoints.
Click your endpoint.
Under "Public endpoint domain", copy the hostname (e.g. 123456789.us-central1-abc.vdb.vertexai.goog).
Do not include https:// — Gateco adds the protocol automatically.

Step 5 — Add the connector in Gateco

Navigate to Connectors → Add connector → Vertex AI Vector Search.
Enter your GCP project ID, region, endpoint ID, deployed index ID, public endpoint domain, and service account JSON.
Click Test connection.
Click Save.

Field	Example	Description
`project_id`	`my-gcp-project`	`GCP project ID`
`region`	`us-central1`	`Region where the index endpoint is deployed`
`index_endpoint_id`	`1234567890123456789`	`Numeric index endpoint ID (from GCP Console URL)`
`deployed_index_id`	`my_index_deployed`	`Deployed index ID (no hyphens)`
`public_endpoint_domain`	`123456789.us-central1-abc.vdb.vertexai.goog`	`Public endpoint hostname only (no https://)`
`service_account_json`	`{"type":"service_account",...}`	`Single-line JSON key from Step 3`
`distance_measure`	`COSINE`	`Distance metric: COSINE (default), DOT_PRODUCT_DISTANCE, or SQUARED_L2_DISTANCE`

Troubleshooting

Error	Cause	Fix
`PERMISSION_DENIED`	`Service account lacks aiplatform.user role`	`Run the gcloud add-iam-policy-binding command from Step 3`
`NOT_FOUND: index endpoint not found`	`Wrong endpoint ID or region`	`Copy the numeric ID from the GCP Console URL, verify the region matches`
`deployed index ID contains hyphens`	`Vertex API rejects hyphenated deployed index IDs`	`Use underscores in the deployed index ID`
`Score is 0 for all results`	`distance_measure mismatch`	`Check your index's configured distance type and set distance_measure to match in the connector config`