Agent Search on Gemini Enterprise Agent Platform
Build Google-quality search for your own data in hours, not months
Agent Search on Gemini Enterprise Agent Platform (formerly Vertex AI Search) helps developers build Google-quality search experiences for websites, structured and unstructured data. It also provides an out-of-the-box grounding system and DIY grounding APIs for building generative AI agents and apps.
Overview
What is Agent Search?
Agent Search is a Google Search quality information retrieval and answer generation system that can be a component of any generative AI application that uses your enterprise data.
There are two key opportunities for enterprises to use Agent Search:
The first is to improve the quality of search experiences across your intranet and customer facing websites. With Agent Search, you can go from frustrating keyword matching to modern conversational search experiences similar to Google’s new generative search experience. This can be as easy as adding a search widget to your webpage.
The second opportunity is to improve the quality of your generative AI applications by grounding them in your enterprise data using Agent Search. Here Gemini Enterprise Agent Platform serves as an out-of-the-box system for retrieval augmented generation or RAG.
Can I use Agent Search as a RAG system?
Yes, you can. Today, there is a lot of excitement about RAG, an architecture that combines LLMs with a data retrieval system, or in other words, a search engine. By grounding LLM responses in your company's own data, it ensures improved accuracy, reliability, and relevance, something that's critical for real-world business applications. You could build your own retrieval augmented generation-based Search but this can be a highly complex process. Agent Search functions as an out-of-the-box RAG system for information retrieval. Under the hood with Agent Search, we’ve simplified the end-to-end search and discovery process of managing ETL, OCR, chunking, embedding, indexing, storing, input cleaning, schema adjustments, information retrieval, and summarization to just a few clicks. This makes it super easy for you to build RAG-powered apps using Agent Search as your retrieval engine.
Does Agent Search have industry specific offerings?
Yes, Agent Search has specialized offerings tuned for unique industry requirements like searching product catalogs, media libraries, and clinical data repositories. Commerce search on Gemini Enterprise for Customer Experience offers retailers the ability to improve the search, product recommendations, and browsing experience on their channels. Agent Search for media offers media and entertainment companies the ability to provide more personalized content recommendations powered by generative AI, increasing consumer time spent on their platforms, which can lead to higher engagement, revenue, and retention. Agent Search for healthcare and life sciences is a medically tuned search that improves patient and provider experience.
Can I use Agent Search to build my own RAG system?
Developing a well-functioning RAG system for DIY grounding can be complex. To address this, Agent Platform offers a comprehensive set of APIs that help developers create bespoke DIY solutions and maintain them. These APIs expose the underlying components of Agent Search's out-of-the-box RAG system, empowering developers to address custom use cases or serve customers who want granular control. These include the Document AI Layout Parser API, ranking API, grounded generation API, and check grounding API.
I want to create my own vector database using embeddings from scratch. Does Agent Platform offer vector search?
Agent Search lets organizations and developers set up search engines out of the box. These search engines offer adequate customization for most enterprise needs and even offer automatic fine-tuning for embeddings. In some cases, you may have custom embeddings, and Agent Search works fine with your own embeddings. However, more advanced developers who need direct control of a highly performant vector database to power niche use cases like recommendations and ad serving can use Vector Search, the vector database used by Agent Search as a component for their use cases. We’ve recently updated Vector Search’s user experience so developers can create and deploy indexes without coding. We’ve also significantly reduced indexing latency from hours to minutes for smaller datasets.
Why should I choose Agent Search?
Agent Search makes it significantly easier for you to build high-quality, AI-powered search experiences into your applications. It is built on Google’s deep expertise and decades of experience in semantic search and so provides more relevant search results. This improves the quality of information retrieval for apps that use your enterprise data. Customization options let you tailor the search experience to your specific needs, while robust enterprise-grade features take care of scalability, privacy, and governance. For more specialized use cases, Agent Search offers vertical specific offerings for retail, media, healthcare, and DIY vector search capabilities.
How does Agent Search offer Google-quality results?
Agent Search is underpinned by a variety of Google Search technologies, including semantic search, which helps deliver more relevant results than traditional keyword-based search techniques by using natural language processing and machine learning techniques to infer relationships within the content and intent from the user’s query input. Agent Search also benefits from Google’s expertise in understanding how users search and factors in content relevance to order displayed results.
How can I get access to Agent Search?
Agent Search is now generally available. You can access via the Google Cloud Console. Please don't hesitate to contact your Google Cloud sales team for assistance or access to preview features.
What generative AI features does Agent Search offer?
Agent Search is powered by foundation models. This means you can offer your customers multi-turn (the ability to easily ask follow-up questions), multimodal (search using images in addition to text), immersive search experiences that are similar to Google's search generative experience. Your customers or employees can view crisp summaries on top of search results with citations and links to data sources that help in knowledge discovery.
Is my data privacy protected while using Agent Search?
Yes, when you use Agent Search from Google Cloud, your data is secure in your cloud instance. Google does not access or use your data to train models or for any other purpose you have not explicitly authorized. Agent Search also meets specific industry compliance standards like HIPAA, ISO 27000-series, and SOC -1/2/3. We’re expanding support for access transparency to provide customers with awareness of Googler administrative access to their data. Virtual Private Cloud Service Controls prevent customers or employees from infiltrating or exfiltrating data. We are also offering Customer-managed Encryption Keys (CMEK) in Preview, allowing customers to encrypt their core content with their own encryption keys.
How does Agent Search ensure results are relevant?
All search results from Agent Search are grounded to your enterprise data or applications you have provided access to. Google Cloud offers grounding out of the box for search results in applications built using Agent Search. Further, Agent Search offers citations and links for summaries generated, which means information presented can be verified by users. You have full control in determining what data sources are used and you can even program responses for off-topic questions.
How does Agent Search maintain freshness of results?
Agent Search can connect to your first-party, Google, and third-party applications through Gemini Enterprise Agent Platform extensions and data connectors. Agent Platform's extensions help in ingesting data and drive transactions on the users' behalf while data connectors ingest data with read-only access to key applications like Jira, Confluence, and Salesforce. Together, Agent Platform extensions and data connectors ensure your data is fresh across your search engines.
How It Works
Your organization may have terabytes of data; and organizing it to be easily found can be one of the most challenging problems to solve. You could also have a public-facing website and need high-quality search for your customers. For both cases, you can use Agent Search to create search engines. Watch this video and discover how to make an internal search app with minimal coding and minimal setup.
Your organization may have terabytes of data; and organizing it to be easily found can be one of the most challenging problems to solve. You could also have a public-facing website and need high-quality search for your customers. For both cases, you can use Agent Search to create search engines. Watch this video and discover how to make an internal search app with minimal coding and minimal setup.
Delivering generative AI-powered search
Get started with Agent Search
Agent Search brings together the power of deep information retrieval, state-of-the-art natural language processing, and the latest in large language processing to understand user intent and return the most relevant results for the user.
This tutorial explains how to create three search apps, one for each type of data: website, structured data, and unstructured data.
Watch the video on the right for an overview and head over to the product documentation link below for detailed instructions.
How-tos
Get started with Agent Search
Agent Search brings together the power of deep information retrieval, state-of-the-art natural language processing, and the latest in large language processing to understand user intent and return the most relevant results for the user.
This tutorial explains how to create three search apps, one for each type of data: website, structured data, and unstructured data.
Watch the video on the right for an overview and head over to the product documentation link below for detailed instructions.
DIY with Vector Search and embeddings
Build a recommendation engine with Vector Search
Find similar things in seconds, even with billions of items. Vector Search unlocks powerful semantic matching for recommendations, chatbots, and more. Let's see how to build a recommendation engine with Vector Search:
- Generate embeddings: Create a numerical representation (embedding) of your items to capture their semantic relationships. You can do this externally or use Gemini Enterprise Agent Platform's generative AI.
- Upload to Cloud Storage: Store your embeddings in Cloud Storage for Vector Search to access.
- Connect to Vector Search: Link your embeddings to Vector Search to perform nearest neighbor search.
- Create and deploy index: Build an index from your embeddings and deploy it to an endpoint for querying.
- Query for recommendations: Use the index endpoint to query for approximate nearest neighbors, finding items semantically similar to your query.
- Evaluate and adjust: Assess the results and refine the algorithm's parameters or scaling as needed to ensure accuracy and performance.
How-tos
Build a recommendation engine with Vector Search
Find similar things in seconds, even with billions of items. Vector Search unlocks powerful semantic matching for recommendations, chatbots, and more. Let's see how to build a recommendation engine with Vector Search:
- Generate embeddings: Create a numerical representation (embedding) of your items to capture their semantic relationships. You can do this externally or use Gemini Enterprise Agent Platform's generative AI.
- Upload to Cloud Storage: Store your embeddings in Cloud Storage for Vector Search to access.
- Connect to Vector Search: Link your embeddings to Vector Search to perform nearest neighbor search.
- Create and deploy index: Build an index from your embeddings and deploy it to an endpoint for querying.
- Query for recommendations: Use the index endpoint to query for approximate nearest neighbors, finding items semantically similar to your query.
- Evaluate and adjust: Assess the results and refine the algorithm's parameters or scaling as needed to ensure accuracy and performance.
Agent Search for healthcare
A medically-tuned Google search experience on healthcare data
Searching data in healthcare can be a difficult task due to the complexities of medical terminology and data standardization.
Agent Search uses its medical tuning to find relevant information from structured and unstructured patient records. It understands medical abbreviations like "abx" and can answer questions with MedLM to provide generative AI answers grounded on patient data. The product integrates with Healthcare Data Engine for a seamless experience.
How-tos
A medically-tuned Google search experience on healthcare data
Searching data in healthcare can be a difficult task due to the complexities of medical terminology and data standardization.
Agent Search uses its medical tuning to find relevant information from structured and unstructured patient records. It understands medical abbreviations like "abx" and can answer questions with MedLM to provide generative AI answers grounded on patient data. The product integrates with Healthcare Data Engine for a seamless experience.
Enable Google-quality search on your website
Boost customer engagement with generative AI powered search
- Create a site index: This is done simply by adding your site URL. Your index is available right away to search if you don’t need generative answers. If you need generative answers, you will need to verify your domain ownership first.
- Connect to a search app: Connect your site index to a new search app, where you will be able to manage the search experience. Make sure to turn LLM features on if you intend to use generative answers.
- Configure your search experience: Set up the right configurations that will define your search experience such as choosing between getting search results only, or being able to receive generative answers.
- Test & refine the search: Preview search results for various queries, and refine your search based on your needs. You can for example add metadata based on your site’s html, boost results based on publication date or other information, filter based on metadata or url patterns.
- Deploy the search to your site: you can choose to deploy using our out of the box widget as an HTML component to add to your site, or to directly integrate using the API.
How-tos
Boost customer engagement with generative AI powered search
- Create a site index: This is done simply by adding your site URL. Your index is available right away to search if you don’t need generative answers. If you need generative answers, you will need to verify your domain ownership first.
- Connect to a search app: Connect your site index to a new search app, where you will be able to manage the search experience. Make sure to turn LLM features on if you intend to use generative answers.
- Configure your search experience: Set up the right configurations that will define your search experience such as choosing between getting search results only, or being able to receive generative answers.
- Test & refine the search: Preview search results for various queries, and refine your search based on your needs. You can for example add metadata based on your site’s html, boost results based on publication date or other information, filter based on metadata or url patterns.
- Deploy the search to your site: you can choose to deploy using our out of the box widget as an HTML component to add to your site, or to directly integrate using the API.
