Google’s Agent Search is what Vertex AI Search is called since April 2026. It is the same product that launched in 2023 as Generative AI App Builder, and was renamed four more times in between. It builds Google-quality search, recommendations and generated answers over your own websites and documents, and serves as a ready-made RAG system for your apps and agents. The console still calls it Vertex AI Search and AI Applications, and the API is still the Discovery Engine API.
Every name, with dates
From Google’s own release notes:
| Date | Name | What Google said |
|---|---|---|
| March 2023 | Generative AI App Builder (Gen App Builder), with Enterprise Search as its search part | Preview for allowlisted customers |
| August 2023 | Vertex AI Search and Conversation | New name, and generally available |
| April 2024 | Vertex AI Agent Builder | The console showed “Agent Builder” |
| April 2025 | AI Applications | New name for console, docs and marketing; functionality and endpoints unchanged |
| October 2025 | Vertex AI Search | Docs and marketing renamed; the console stayed “AI Applications” |
| April 2026 | Agent Search | Docs renamed; the console stays “Vertex AI Search” and “AI Applications” |
So if you are looking for any of these, it’s the same product:
- In the console: Vertex AI Search, under AI Applications.
- In the API and client libraries: Discovery Engine, at
discoveryengine.googleapis.com. - In current docs and pricing: Agent Search.
On the same day in April 2026, Vertex AI itself became part of Gemini Enterprise Agent Platform. Vertex AI Studio is now Agent Studio, Vertex AI Agent Engine is Agent Runtime, and Vector Search 2.0 is Agent Retrieval.
What does it do today?
You create an app and connect it to one or more data stores:
- Search apps over websites, documents and structured data, with generated answers and summaries on the Enterprise edition.
- Recommendation apps for content and media.
- Data stores from websites, Cloud Storage, BigQuery, Google Drive, Gmail, Google Sites, Calendar and Groups, Cloud SQL, Spanner, Firestore, Bigtable and AlloyDB, or JSON through the API.
It’s also a grounding source: an app or agent can ask Agent Search for relevant passages and answer from them, which is the RAG pattern without running your own index.
The connector trap: Google’s marketing pages still mention connectors for Jira, Confluence or Salesforce. Agent Search’s own docs say third-party data sources are no longer supported there. Those connectors moved to Gemini Enterprise, Google’s app for employees, formerly Agentspace.
Where did “Agent Builder” go?
The name moved, the agent features moved with it. Building and running agents is now part of Gemini Enterprise Agent Platform: the Agent Development Kit to write agents, Agent Runtime to host them, and Agent Studio, a low-code canvas. Agent Search is the search and retrieval part.
How do I create a search app?
In the console, still labelled AI Applications:
- Open Create data store, choose Website content or Cloud Storage, and enter your sites or bucket.
- Choose the global location, name the data store and click Create.
- Open Create App, and under Site search with AI mode click Create.
- Turn on Enterprise edition features if you want generated answers.
- Name the app, choose global, select your data store and click Create.
- Watch indexing on the app’s Data page, then try queries under Preview.
From code, query it through the Discovery Engine API:
curl -X POST \
-H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:search" \
-d '{"query": "How do refunds work?", "pageSize": 10}'APP_ID is the app’s ID from the console. Google’s Python client is google-cloud-discoveryengine.
What does it cost?
Two models, and you pick one per app:
- General: pay per query, at Standard or Enterprise edition rates, with generated answers as an add-on, plus index storage per GiB.
- Configurable: a monthly commitment for query capacity and storage.
There is a monthly free trial allowance for queries, without the advanced generated answers, and some index storage is free each month. Current rates and free amounts are on Google’s pricing page.
New billing accounts have also seen a one-time “Trial credit for GenAI App Builder”. On our own account it shows up as specific-use credit valid for a year. Google’s current pages no longer describe it, and posts from developers report it only covers this product family, not the Gemini API. Check its scope under Billing → Credits before you count on it.
When is it worth it?
- Search over a website or a document collection, quickly and well, without running your own index.
- Grounding for your own agent in data that already sits in Google Cloud.
- Your data and compliance rules live in Google Cloud already.
It’s the wrong tool for source code an agent works on, which changes constantly; coding agents search live files instead. See what are embeddings?
Related guides
- GenAI app builders: Google, AWS, Microsoft and OpenAI side by side
- What is RAG?: what Agent Search does for you, in 50 lines of code
- Vertex AI: Google’s models from Claude Code, Codex or OpenCode
- AI coding costs: credits you may already have