For Vertex AI Agents release notes, see the Dialogflow release notes.
You can see the latest product updates for all of Google Cloud on the Google Cloud page, browse and filter all release notes in the Google Cloud console, or programmatically access release notes in BigQuery.
To get the latest product updates delivered to you, add the URL of this page to your feed reader, or add the feed URL directly.
December 09, 2024
Vertex AI Agent Builder: Grounding is available in more languages (GA with allowlist)
The grounded generation API supports more than 35 languages.
This feature is available to select Google Cloud customers (GA with allowlist). For general information about grounding, see Generate grounded answers with RAG. For available languages, see Languages.
Vertex AI Agent Builder: Additional inputs for generating grounded answers (GA with allowlist)
You can specify a language code and a latitude-longitude value when making calls to the grounded generation API.
If the language can't be determined from the query, then the language code is used to set the language for the answer. If the language code is not present, then the latitude-longitude value is used to set the language.
The latitude-longitude value is also used to answer location-related queries, such as "restaurants near me".
This feature is available to select Google Cloud customers (GA with allowlist). For more information, see Generate grounded answers with RAG.
December 04, 2024
Vertex AI Search: Boost controls for media recommendations (Public preview)
Boost controls are used to affect the order in which recommendations are listed. Boost controls use filters on string and boolean values in the schema data to determine what media content to boost or bury. The boost value (-1 to 1) determines whether the content should be placed lower (buried) or higher (boosted) in the list of recommendations returned.
Boost controls are attached to serving configs and applied to recommend method calls.
The boost feature is in public preview and is available through the API. For more information about the feature, see Boost and bury media recommendations.
Vertex AI Search: gemini-1.5-flash-002-high-fidelity model (Public preview)
The gemini-1.5-flash-002-high-fidelity model is available for grounded answer generation with RAG. This model is based on the gemini-1.5-flash-002 model and has been further tuned to address context-based question and answering tasks. This model is suitable for specialized industries, such as financial services, healthcare, and insurance.
This model is available in Public preview.
For more information, see High fidelity models.
December 03, 2024
Vertex AI Search: Index and refresh web pages using sitemap (Public preview)
If advanced website indexing is enabled in your data store, you can submit and use sitemaps and sitemap indexes to index and refresh the web pages in your data store. This feature supports only XML sitemaps and sitemap indexes.
This feature is in public preview and is available through the API. For more information about the feature, see Index and refresh web pages using sitemaps.
November 26, 2024
Vertex AI Search: Check ingested data quality for media recommendations (GA)
You can check the quality of your ingested data for media recommendations through the Google Cloud console. These checks are not blocking but can suggest ways that your data can be improved. This feature is Generally available (GA).
Previously, this check was only available through API method calls.
For more information, see Check data quality for media recommendations.
October 31, 2024
Vertex AI Search: Stream answers (GA with allowlist)
The answer streaming method can return generated answers in sequential parts. This reduces the perception of latency. As the end users read the first part of the answer, the subsequent parts of the answer are being generated.
The answer streaming method also includes many of the features of the original answer method.
This feature is Generally available to select Google customers (GA with allowlist). For more information, see Stream answers.
October 25, 2024
Vertex AI Search: Get grounding scores for answers with summaries and follow-ups (GA)
The answer method can return aggregated grounding scores for answers and individual grounding scores for claims.
This feature is Generally available (GA). For more information, see Return grounding support scores.
Vertex AI Search: Return only well-grounded answers with summaries and follow-ups (GA)
With the answer method, you can choose to filter out poorly-grounded answers. There are two filter levels: choose to return only answers with high grounding scores (at the risk of losing some helpful answers) or choose a lower filter to get more answers.
This feature is Generally available (GA). For more information, see Show only well-grounded answers.
Vertex AI Search: Advanced autocomplete (Public preview)
Use advanced autocomplete to enable autocomplete on blended search apps. Also, advanced autocomplete supports:
- Access control
- Language boosting
- Rich suggestions, which return document suggestions or recent search suggestions
For more information, see Configure advanced autocomplete. This feature is in Public preview.
October 17, 2024
Vertex AI Search: CMEK for US and EU (GA) and CMEK with EKM and HSM (GA with allowlist)
Customer-managed encryption keys (CMEK) are Generally available (GA) in the US and the EU. You no longer need to be added to an allowlist to use CMEK. If you store your data in a US or EU multi-region data store, you can provide your own encryption key to protect your data at rest.
Using external key manager (EKM) or hardware security module (HSM) with CMEK is in GA with allowlist.
For information, see Customer-managed encryption keys.
October 14, 2024
Vertex AI Search: Answers with summaries and follow-ups for blended search apps (GA with allowlist)
The answer method can be used to query blended search apps. You can apply the answer method to blended search apps in the same way that you apply the method to search apps that are connected to only one data store.
This feature is Generally available to select Google customers (GA with allowlist). For more information, see Get answers and follow-ups.
October 01, 2024
Vertex AI Agent Builder: Dynamic retrieval for grounded results (GA with allowlist)
Dynamic retrieval lets you choose when to turn off grounding with Google Search. This is useful when a prompt doesn't require an answer grounded in Google Search and the supported models can provide an answer based on their own knowledge without grounding. Dynamic retrieval helps you manage latency, quality, and cost more effectively.
This feature is available to select Google Cloud customers (GA with allowlist). For more information, see Dynamic retrieval.
September 25, 2024
Vertex AI Search: gemini-1.5-flash-002/answer_gen/v1 model
The gemini-1.5-flash-002/answer_gen/v1 model is available for answer generation. This model is based on the gemini-1.5-flash-002 model and has been further tuned to address question and answering tasks.
For more information, see Answer generation model versions and lifecycle.
Vertex AI Search: Update to the preview model
The preview model for answer generation has been updated to gemini-1.5-pro-002 from gemini-1.5-pro-001.
For more information, see Answer generation model versions and lifecycle.
September 18, 2024
Vertex AI Agent Builder: Redirection URI for grounded results (GA)
When you use Grounding with Google Search, the grounded result contains a redirection URI that leads you to the publisher's URI. This redirection URI remains accessible for up to 30 days after the grounded result is generated.
This feature is Generally available (GA). For more information, see Generate grounded answers with RAG.
September 17, 2024
Vertex AI Search: Firestore and Cloud SQL import (GA)
Importing data from Firestore and Cloud SQL is Generally available.
For more information, see Import from Firestore and Import from Cloud SQL.
September 11, 2024
Vertex AI Search: Natural language query filters (Public preview)
For queries on structured data stores, the natural language queries can be reformulated as filters and a residual query. For example, "Find a coffee shop serving banana bread" becomes
"query": "banana bread", "filter": "type": ANY(\"cafe\").
The natural-language query understanding feature only applies to generic apps that use structured data stores.
This feature is in Public preview. For more information, see Filter with natural language understanding.
Vertex AI APIs: Updated model for ranking and reranking documents for RAG
The ranking API model is upgraded. This underlying model significantly improves the relevance of top-ranked documents and provides more nuanced scores. For more information about ranking documents, see Rank and rerank documents with RAG.
August 28, 2024
Vertex AI Search: Turn off schema auto-detect for structured data
By default, schema auto-detect dynamically adds new properties to the schema when the property fields are detected on data import.
However, you can turn off the dynamic feature so that only data that corresponds to fields already in the schema get imported. This is good approach for not-so-clean data because you can choose not to import extraneous data that isn't part of your defined schema and that you don't want in your structured data store.
For more information, see About providing your own schema as a JSON object.
Vertex AI Search: Datetime and geolocation detection for structured data
By default, when structured data is imported, fields that are detected in datetime and geolocation format are assigned those types in the schema.
However, you can turn off datetime and geolocation detection so that in the schema the datetime fields are set to type string and the geolocation fields are set to type object.
For more information, see About providing your own schema as a JSON object.
August 23, 2024
Vertex AI Search: Connect Google Cloud Storage datasets to Vertex AI Search (Public preview)
You can create Vertex AI Search data stores that periodically sync with data in Cloud Storage datasets. You can choose how often you want to update your data stores: every day, every 3 days, or every 5 days.
Synchronizing Cloud Storage data to Vertex AI Search is available in Public preview. For more information, see Import from Cloud Storage.
August 16, 2024
Vertex AI Search: Search tuning (GA)
Search tuning for unstructured data stores is Generally available (GA). You can upload training files to tune the model for your search app.
Search tuning supports Data Residency; you can tune data stores in the US and EU multi-regions as well as global data stores.
For information, see Improve search results with search tuning.
August 06, 2024
Vertex AI Search: Layout parser GA
The layout parser for Vertex AI Agent Builder is Generally available. The layout parser transforms documents in various formats into structured representations. It makes content like paragraphs, tables, lists, and structural elements like headings, page headers, and footers easily accessible.
For more information, see Layout parser.
Vertex AI Search: Generative answers performance improvements
Generative answers have been updated with performance improvements.
- Re-ranking for generative answers has been updated to decrease response latency.
- Detection of adversarial queries has been updated for improved accuracy.
July 25, 2024
Vertex AI Search: Domain verification (GA)
Domain verification for advanced website indexing using domain association is Generally available (GA). You can use domain association to associate your Vertex AI Search data store to the specified domain. This is useful when you're not the owner of the specified domain or when you don't have access to the Google Search console needed to verify the domain.
For more information, see Verify website domains.
July 23, 2024
Vertex AI Search: Widget uses new method for generative answers
The search widget now uses the search and answer methods together, instead of the older search with summaries for Search with an answer and the converse method for Search with follow-ups.
The answer method is expected to improve the quality of the results.
For general information about the answer method, see Get answers and follow-ups.
July 19, 2024
Vertex AI Search: Multi-step retrieval for answer (GA)
For the answer method, multi-step retrieval using multi-step (ReAct) reasoning is Generally available (GA).
For information about this feature, see Query rephrasing and Search and answer (specify maximum steps).
July 17, 2024
Vertex AI Search: Evaluate search quality (Public preview)
Evaluate the search quality of your generic search applications using sample query sets. This lets you assess your search engine's performance, understand potential biases or shortcomings in ranking algorithms, and compare historical evaluation results to understand the impact of changes in your search configuration.
For more information, see Evaluate search quality. This feature is in Public preview.
July 15, 2024
Vertex AI Search: Rotation of CMEK keys, which protect data stores (Private preview)
Customer-managed encryption keys (CMEK) for data stores associated with search apps can be rotated.
Don't rotate keys for data stores associated with recommendations apps. Also, if you rely on analytics, don't rotate keys.
Key rotation is available in Private preview. For information about rotating CMEK keys to protect Vertex AI Agent data stores, see Customer-managed encryption keys.
July 10, 2024
Vertex AI Search: Edit the schema for structured data on import (Public preview)
When you create a data store by importing structured data from BigQuery or Cloud Storage, you can review and edit the schema before you import the data. This saves time over the alternative method of importing the data first and subsequently editing the schema.
This feature is available in Public preview and applies to generic and media data stores. To try this feature for healthcare data stores, contact your Google account team and ask for access to the Private preview.
Vertex AI Search: Bring your own schema for media data stores (Public preview)
Previously, all media data stores had to follow a JSON schema for media predefined by Google. However, now you can use your own JSON schema for media data, provided that you map fields in your schema to the key properties: category, media_available_time, media_duration, title, and uri.
This feature is in Public preview.
Vertex AI Search: Media app creation (Public preview)
Media data stores can be created directly from the Data Stores page.
This is an alternative to the method where you create a media data store as part of the app creation workflow.
This feature is available in Public preview.
July 03, 2024
Vertex AI Search: On July 6, text-bison@001/answer_gen/v1 is discontinued
As of July 6, 2024, model version text-bison@001/answer_gen/v1 is discontinued.
If you specify text-bison@001/answer_gen/v1 by name in your search requests, replace text-bison@001/answer_gen/v1 with a newer model or with stable.
For more information, see Answer generation model versions and lifecycle.
Vertex AI Search: gemini-1.5-flash-001/answer_gen/v1 for answer generation
Model version gemini-1.5-flash-001/answer_gen/v1 is the stable model for generating answers in Vertex AI Search.
For more information, see Answer generation model versions and lifecycle.
Vertex AI Search: You can't use the Folder option to upload structured data from Cloud Storage
When creating a data store for structured or media data, you must use the File option when importing from a Cloud Storage bucket. Choosing the Folder option results in an error, "Schema preview failed. Requested entity was not found."
To work around this issue, use the File option and upload one file from the folder. After you've created the data store, import the folder contents from the Documents tab of the data store.
July 01, 2024
Vertex AI Search: Filter search results by relevance (Public preview)
Each document returned by a search query is given an estimated level of relevance to the query. When you make a query through an API call, you can set a relevance threshold.
Setting a high relevance threshold can greatly reduce the number of documents returned by a query. You can experiment with low, medium, and high thresholds to find the right level for your users.
Filter by relevance is available in Public preview.
For more information, see Filter searches by document-level relevance.
Vertex AI Search: Healthcare search using natural language query with generative AI answers (GA with allowlist)
Healthcare data search using natural language query with generative AI answer is Generally available to select Google customers (GA with allowlist).
For more information, see Search using natural language query with generative AI answer.
June 27, 2024
Vertex AI Search: Connect BigQuery datasets to Vertex AI Search (Public preview)
You can create Vertex AI Search data stores that periodically sync with data in BigQuery datasets. You can choose how often you want to update your data stores: every day, every 3 days, or every 5 days.
Synchronizing BigQuery data to Vertex AI Search is available in Public preview.
For more information, see Import from BigQuery.
June 24, 2024
Vertex AI Search: Check ingested data quality for media recommendations (Public preview)
You can check the quality of your ingested data for media recommendations.
By running the Public preview requirements:checkRequirement method, you find out if your data store meets the minimum quality requirements for your recommendations app. If your data doesn't meet the minimum threshold for the key metrics for your model and objective, you receive a warning about the issues. Address the issues and rerun the check.
For more information, see Check data quality for media recommendations.
June 21, 2024
Vertex AI Search: Answers with summaries and follow-ups (GA)
The answer API improves on the search with summary and search with follow-ups features. For example, it better handles complex queries and provides customization of answer styles.
The answer API is Generally available (GA). However, the multi-step retrieval functionality remains in Public preview.
For more information, see Get answers and follow-ups.
Vertex AI Search: The answer method can skip irrelevant answers
The answer method can be set to generate an answer only if at least one of the results is deemed relevant.
If you choose to ignore low relevant content and if all the results are deemed irrelevant or almost irrelevant, then the answer method doesn't generate an answer. Instead, a fallback message replaces the answer.
For more information, see Show only relevant answers.
Vertex AI Search: Add structured data for advanced website indexing (Public preview)
If advanced website indexing is enabled in your data store, you can use structured data, such as schema.org data, to enrich your indexing.
For more information, see Use structured data for advanced site indexing.
Vertex AI Search: Generate grounded answers (GA with allowlist)
You can add system instructions as preambles to your prompts. System instructions govern the behavior of the model and modify the output accordingly. For example, you can add a persona to the generated answer or instruct the model to format the output text a certain way.
For more information, see Generate grounded answers.
Vertex AI Search: The generated answer message doesn't contain the name field for synchronous and sessionless queries
The name field is only included in the answer response for session queries and for asynchronous queries. These are stateful and context-aware queries.
If a query is a synchronous and stateless query, the name field is no longer included in the generated answer message.
For more information about the answer method, see Get answers and follow-ups.
Vertex AI Search: Choose when to enable autocomplete
You can choose to enable autocomplete as soon as possible instead of waiting a couple of days for sufficiently good autocomplete data. If you choose to make autocomplete available sooner, at first, you won't get suggestions for all queries and some suggestions might be of poor quality.
For more information, see Enable autocomplete in Update autocomplete settings.
June 14, 2024
Vertex AI Search: Boost search results
Boosting search results for media apps and for generic search apps that contain unstructured and website data is Generally available.
For more information, see Boost search results.
Vertex AI Search: Set language codes for data stores (Public preview)
Setting a language code for a data store can improve the quality of the extractive segments and extractive answers returned in search results. Language codes for data stores are supported in public preview.
For information about the language code field for data stores, see the DataStore resource.
Vertex AI Search: Specify a language code in search request (Public preview)
Setting a language code in a search query can improve the quality of the search results. Language codes in search queries are supported in public preview.
For information about the language code field in search, see the servingConfigs.search method.
June 05, 2024
Vertex AI Search: Generate grounded answers (GA with allowlist)
Generating grounded answers is Generally available to select Google Cloud customers (GA with allowlist).
As part of your Retrieval Augmented Generation (RAG) experience, generate grounded answers based on Google Search, inline text, or the content in your Vertex AI Search data store. You can generate answers in a single turn or over multiple turns. For more information, see Generate grounded answers.
When you use Google Search as a grounding source, you connect your Gemini large language model (LLM) to the most up-to-date information on the internet. You must display a Google Search entry point when grounding with Google Search. For more information, see Use Google Search entry point.
May 31, 2024
Vertex AI Search: Document ranking API (GA)
The ranking API takes a list of documents and reranks those documents based on how relevant the documents are to a query. This is a stateless API that does not require you to index documents in advance.
The ranking API is Generally available (GA).
For more information, see Rank and rerank documents.
May 14, 2024
Vertex AI Search: Check grounding (GA)
The check grounding API is Generally available (GA).
The check grounding API determines how grounded a piece of text is in a given set of facts. The API returns support scores and citations.
Filler and introductory statements can be deemed as not requiring attribution. No scores or citations are provided for those statements.
Additionally, as an experimental feature, the API also generates contradicting citations that show which facts contradict the text and how strongly.
For more information, see Check grounding and the check API.
April 29, 2024
Vertex AI Search: Order healthcare search results (Public preview)
When you search over FHIR resource types that contain unstructured text, you can order your search results according to their relevance to your query. For more information, see Order healthcare search results.
Vertex AI Search: Boost search results (Public preview)
Boosting search results for media apps and for generic search apps that contain unstructured and website data is available in Public preview. For more information, see Boost search results.
Vertex AI Search: Add structured data for advanced website indexing (Public preview)
If advanced website indexing is enabled in your data store, you can use structured data, such as Google-inferred page dates, meta tags, and PageMap content, to enrich your indexing.
For more information, see Use structured data for advanced site indexing and Example use case using a Google-inferred page date.
Vertex AI Search: gemini-1.0-pro-002/answer_gen/v1 for answer generation
Model version gemini-1.0-pro-002/answer_gen/v1 is available for generating answers in Vertex AI Search. For more information, see Answer generation model versions and lifecycle.
April 24, 2024
Vertex AI Agent Builder: Renamed in the console and documentation
The Google Cloud console and the documentation at cloud.google.com have been updated to show the current product name for Vertex AI Agent Builder. On the console, look for "Agent Builder".
You might see the old name (Vertex AI Search and Conversation) in some places—for example, in videos.
April 09, 2024
Vertex AI Search: Document chunking support for more search types (Public preview)
When document chunking is turned on for an unstructured data store, search summaries and search with follow-ups are supported in Public preview.
For information, see Chunk documents for RAG.
Vertex AI Search: Document ranking API (Public preview)
The ranking API takes a list of documents and reranks those documents based on how relevant the documents are to a query. This is a stateless API that does not require you to index documents in advance.
For more information, see Rank and rerank documents.
Vertex AI Search: Check grounding (Public preview)
The check grounding API is available as a Public preview feature.
The check grounding API determines how grounded a piece of text is in a given set of facts. Perfect grounding requires that every statement in the text can be attributed to one or more of the given facts. The API returns support scores and citations.
Additionally, as an experimental feature, the API also generates contradicting citations that show which facts contradict the text and how strongly.
For more information, see Check grounding and the check API.
Vertex AI Search: Answers with summaries and follow-ups (Public preview)
The answer API improves on the search with summary and search with follow-ups features. For example, it better handles complex queries, can do multi-step retrieval, and provides customization of answer styles.
The answer API is supported in Public preview.
For more information, see Get answers and follow-ups.
Vertex AI Search: FHIR data streaming ingestion (Private preview)
Select the import frequency for your healthcare FHIR data. You can either perform a one-time batch import or set up a streaming import. Streaming import is available as a Private preview feature.
For more information, see Create a healthcare search data store.
Vertex AI Search: Autocomplete support for healthcare search (Public preview)
Autocomplete is available as a Public preview feature for healthcare data search. The autocomplete configuration uses a canonical medical data source to generate autocomplete suggestions for healthcare data stores.
For more information, see Configure autocomplete.
Vertex AI Search: Connect Google Drive to Vertex AI Search (GA)
Syncing Google Drive data to Vertex AI Search is available in GA. For more information about creating a Google Drive data store, see Sync from Google Drive.
Vertex AI Search: Connect multiple search apps to the same data store (GA)
Connecting more than one generic search app to a single data store is supported in GA. With this capability, you can create multiple apps that search across the same data without having to ingest that data multiple times.
Vertex AI Search: Blended search (GA)
Blended search, where you can search across multiple data stores using a single search app, is available in GA. For more information about blended search, see About connecting multiple data stores.
Vertex AI Search: Connect Spanner, Cloud SQL, Firestore, and Bigtable to Vertex AI Search (Public preview)
Importing data from Spanner, Cloud SQL, Firestore, and Bigtable to Vertex AI Search is available in Public preview. For more information about creating a Google Drive data store, see Create a search data store.
Vertex AI Search: Media search (GA)
Vertex AI Search for media is Generally available (GA).
You can create media search apps on media data stores. You can connect the media search app to an existing media data store or create a new one. You can also use document metadata to filter search queries of your media content.
Vertex AI Search: Additional languages supported for media search
Vertex AI Search for media is supported in nine languages: Arabic, English, French, German, Hindi, Korean, Japanese, Portuguese, and Spanish.
For more information, see Languages.
Vertex AI Search: Search-as-you-type for media apps
Search results are returned after each character instead of after the full query is entered. Search-as-you-type is ideal for search apps with awkward input devices such as television remotes. You can enable search-as-you-type through the widget UI as well as through the API.
For more information, see Get search-as-you-type results for a media app.
March 28, 2024
Vertex AI Search: Autocomplete updates
Autocomplete is Generally available (GA) for the US and EU multi-regions as well for the global region.
Autocomplete supports access transparency. This means that, when access transparency is enabled, if Google personal access your autocomplete data, this is recorded in the Access Transparency logs.
For information about autocomplete, see Configure autocomplete, and for information about Access Transparency, see Enable Access Transparency in Vertex AI Search.
Vertex AI Search: Structured data stores (GA)
Use of data stores containing structured data is Generally available (GA). Additionally, two new field value types are allowed for structured data stores: geolocation and datetime.
For information about structured data stores, see Structured data in Prepare for ingesting and Schemas: auto-dectection versus providing you own.
Vertex AI Search: Boost search results (Public preview)
Boosting search results using custom numerical attributes and according to freshness is available in Public preview.
For more information, see Boost search results.
Vertex AI Search: Extractive segments and relevance scores (GA)
Extractive segments and relevance scores for extractive segments are GA.
For more information, see Extractive segments.
Vertex AI Search: Document chunking and parsing improvements (Public preview)
The following improvements have been introduced for document chunking and parsing:
- Adjacent chunks: When returning chunks in search responses, you can return chunks from immediately before and after the relevant chunk in the source document. Doing so can improve context and accuracy.
- Page span: Chunk metadata in search responses includes the span of pages where the chunk appeared in the source document.
- List chunks: List all chunks from a specific source document.
- Get chunks: Get a specific chunk.
- Get processed documents in JSON: Get a parsed document or a chunked document in JSON format.
- Bring your own chunks (Preview with allowlist): Upload data that you've already chunked. Contact your Google account team if you're interested in trying this feature.
For more information, see Parse and chunk documents.
Vertex AI Search: Media recommendations analytics (GA)
Analytics for media recommendations are GA. You can view analytics for your media recommendations apps in the Google Cloud console.
For more information, see View analytics.
March 15, 2024
Vertex AI Search: Sync from Google Drive (Preview with allowlist)
Connecting to Google Drive as a data source for Vertex AI Search is available as a Preview with allowlist feature. For more information, see Sync from Google Drive.
March 12, 2024
Vertex AI Search: Vertex AI Search for healthcare (GA)
Vertex AI Search for healthcare is Generally available (GA). Healthcare search lets you query healthcare records stored in FHIR data stores. For more information, see Vertex AI Search. With healthcare search, you can:
Vertex AI Search: Specify a parser for unstructured content (Public preview)
You can control how documents are parsed when they are uploaded to Vertex AI Search. Parser specification is available in Public preview.
Vertex AI Search provides a digital parser (GA), an OCR parser for PDFs (Public preview), and a layout parser (Public Preview). During data store creation for generic search apps with unstructured data, you can set a default parser for the data store and an override parser for specific file types.
For more information, see Parse documents.
Vertex AI Search: Turn on document chunking (Public preview)
To use Vertex AI Search for retrieval-augmented generation (RAG) for LLMs, you can turn on document chunking when creating a data store. Document chunking is available in Public preview.
When document chunking is turned on, your documents are split into chunks when you ingest documents into your data store, and your search app can return chunks of data in search results instead of full documents. Using chunked data for RAG increases relevance for LLM answers and reduces computational load for LLMs. Document chunking is in Public preview. For more information, see Chunk documents for RAG.
Vertex AI Search: Connect ServiceNow as a data source (Private preview)
You can connect ServiceNow as a third-party data source for Vertex AI Search. For more information, see Connect a third-party data source.
March 05, 2024
Vertex AI Search: Watch time duration objective for media recommendations apps
When you create a media recommendations app, you can select watch duration per session as a business objective. Optimizing for watch duration per session maximizes the duration of media consumption.
For more information, see Watch duration per session.
February 28, 2024
Vertex AI Search: Add metadata to your web index (Public preview)
If advanced website indexing is enabled in your data store, you can add metadata to the data store schema to enrich your indexing.
For more information, see Add metadata for advanced site indexing.
Vertex AI Search: Automatic web page refresh (Public preview)
With advanced website indexing, Vertex AI Search performs conditional, automatic refresh.
For more information, see Refresh web pages.
Vertex AI Search: Apply tuned search to some queries (Public preview)
You can specify whether you want a query to use the tuned search model or the non-tuned search model. This is particularly helpful for testing the difference between the two versions of the model.
Previously, the tuned search model was enabled (or disabled) for all queries against the data store.
For more information, see Test tuned search and use it for individual search queries.
Vertex AI Search: Access controlled data sources (Public preview)
Access control for BigQuery, Cloud Storage, and Confluence data is available in Public preview. This feature allows you to limit the data that users can view in your search app's results. Google uses your identity provider to identify the end user performing a search and determine if they have access to the documents that are returned as results. Google Identity and third-party identity provider federation are supported.
For more information, see Use data source access control.
Vertex AI Search: Blended search (Public preview)
Blended search, where multiple data stores can be connected to a single generic search app, is available in Public preview. This feature allows you to use one generic search app to search across multiple sources and types of data.
For more information, see About connecting multiple data stores.
Vertex AI Search: Search analytics (GA)
Search analytics are GA for global data stores. For data stores in US and EU multi-regions, viewing analytics is in Public Preview.
For more information, see View analytics.
February 26, 2024
Vertex AI Search: Use Terraform to create search apps
You can use Terraform to create search apps for your Vertex AI Search.
For information, see Create a search app.
February 15, 2024
Vertex AI Search: Stable Gemini Pro answer generation model
gemini-pro@001/answer_gen/v1 is available as a stable, generally available model for answer generation. For information about all available models for answer generation, see Specify the summarization model.
January 31, 2024
Vertex AI Search: CMEK for US and EU is GA
Customer-managed encryption keys (CMEK) are available in the US and the EU as GA with allowlist.
If you store your data in a US or EU multi-region data store, you can provide your own encryption key to protect your data at rest.
For information, see Customer-managed encryption keys.
Vertex AI Search: Check grounding in Preview with allowlist
The CheckGrounding API determines how grounded a piece of text is in a given set of facts. Perfect grounding requires that every statement in the text can be attributed to one or more of the given facts. The API returns an overall score of 0 to 1, indicating how grounded the text is, along with citations to the appropriate given facts for each statement.
See Check grounding.
Vertex AI Search and Conversation: Use Terraform to create data stores
You can use Terraform to create data stores for your Vertex AI Search and Conversation apps. The data stores are created empty; you then ingest the data through the console or an API call.
For information, see, for example, Create a search data store.
Vertex AI Search: Gemini Pro for search summaries
You can now choose Gemini Pro as a model for generating search summaries.
For more information, see Specify the summarization model.
Vertex AI Search: Updates to autocomplete
Autocomplete is available for your search apps in the US and EU multi-regions as Public preview.
Autocomplete removes unsafe and offensive terms in eight languages in addition to English (
en).For more information, see Autocomplete features.
December 21, 2023
Vertex AI Search: Access controlled data sources in Preview with allowlist
Access control for BigQuery, Cloud Storage, and Confluence data is available in Preview with allowlist. This feature allows you to limit the data that users can view in your search app's results. Google uses your identity provider to identify the end user performing a search and determine if they have access to the documents that are returned as results. Google Identity and third-party identity provider federation are supported.
Vertex AI Search: Bring your own parsed documents in Preview with allowlist
Importing pre-parsed unstructured documents into Vertex AI Search data stores is available in Preview with allowlist. For example, instead of importing a raw PDF document, you can parse the PDF yourself and import the parsing result instead. This lets you to import your documents in a structured way, ensuring that search and answer generation have full information about the document's layout and elements.
See Bring your own parsed document. For more information, contact your Google account team.
Vertex AI Search: User events joined with documents asynchronously for generic apps
For generic search and recommendation apps, user events are joined with their associated documents asynchronously. For media apps, user events continue to be joined synchronously.
For more about user event import, see Import historical user events.
December 13, 2023
Vertex AI Search: Media search (Preview with allowlist)
You can create media search apps on media data stores. You can connect the media search app to an existing media data store or create a new one. You can also use document metadata to filter search queries of your media content.
Media search is available as a Preview with allowlist capability. To use this capability, contact your Google account team.
See About media documents and data stores and Filter media search.
Vertex AI Search: Company name field for search apps
When you create a search app, you supply the name of your company or organization. This can improve the quality of the responses for summarization and search with follow-ups.
See Create a search app.
Vertex AI Search: Upgrade to advanced website indexing
You can update an existing data store to advanced website indexing.
Vertex AI Search: Support for Access Transparency is GA
Access Transparency supports Vertex AI Search in GA.
Vertex AI Search: Additional languages supported
The search, snippets, and extractive answers features are supported in 25 languages.
Summarization is supported in most languages.
See Languages.
Vertex AI Search: Estimate the size of your web data
You can call an API to get an estimate of the size of your web data prior to importing it into a data store. With this information, you can estimate your monthly storage costs using the Vertex AI Search and Conversation pricing page or the Google Cloud Pricing Calculator.
Vertex AI Search: Customized Summaries is GA
Customized summaries are available in GA. For search with answers and search with follow-ups, you can provide free-form natural-language instructions to customize the summaries that are returned with your results. Customize summaries through the console for search widget results and through the API for other search requests.
For more information, see Get customizable summaries.
Vertex AI Search: Filter search with follow-ups
You can filter your queries for search with follow-ups, both for website data and for structured data with metada.
Vertex AI Search: Search tuning (Public preview)
Search tuning is available as a Public preview feature. You can upload additional training data to tune the model for your search app.
For information, see Improve search results with search tuning.
Vertex AI Search: Use multiple serving configs for media recommendations apps
For media recommendations apps, you can create, edit, and delete serving configs. Serving configs contain recommendation demotion and diversity configurations that are used when generating results. When you make recommendations calls to a media app, you can specify different serving configs to get different levels of demotion and diversity in recommendations from the same app.
For more information, see Create and manage serving configs.
Vertex AI Search: Citations with metadata for summaries (Public preview)
When citations are turned on for summaries, the response includes citation metadata. This includes metadata about the sources that are being referenced and information about which sources apply to specific sentences in the summary.
See Get citations.
Vertex AI Search: OCR processing (Public preview)
OCR processing is available in Public preview. If you have non-searchable PDFs (scanned PDFs or PDFs with text inside images), you can turn on optical character recognition
(OCR) processing during data store creation. This allows Vertex AI Search to extract elements such as text blocks and tables. If you instead have searchable PDFs that contain many tables, you might also consider turning on OCR processing with useNativeText set to true and enhancedDocumentElements set to table.
Vertex AI Search: Blended search (Preview with allowlist)
Blended search, where multiple data stores can be connected to a single generic search app, is available in Preview with allowlist. This feature allows you to use one generic search app to search across multiple sources and types of data.
November 15, 2023
Vertex AI Search: Autocomplete denylist (Preview with allowlist)
Importing an autocomplete denylist is available as a preview with allowlist feature. To use this feature, contact your Google account team.
For information about autocomplete denylists, see Use an autocomplete denylist.
November 14, 2023
Vertex AI Search: Additional languages supported
Extractive answers are now supported in the following languages:
- Arabic
- Chinese (Simplified)
- Japanese
See Languages.
November 09, 2023
Vertex AI Search: New model for search summarization
A better model for generating search summaries has been launched. This underlying model improves the quality of search summaries and their grounding in the provided document corpus. You might see some differences in summary output after this update.
For more information about search summaries, see Get search summaries.
Vertex AI Search: Confidence scores are changed to relevance scores (Preview with allowlist)
Confidence scores are renamed to relevance scores. Scores are returned in the relevanceScore field. Previously, they were returned in the confidenceScore field.
This feature is in preview with allowlist. For more information about relevance scores, see Get snippets and extracted content.
November 06, 2023
Vertex AI Search: Multi-region support for US and EU locations is GA
The US multi-region and the EU multi-region APIs are generally available (GA).
For more information about multi-regions including limitations, see Vertex AI Search locations.
October 27, 2023
Vertex AI Search: Create media recommendations in Vertex AI Search
You can now create apps for media recommendations in Vertex AI Search. Media recommendations include content such as videos, news, and music. For more information, see Vertex AI Search.
Important: If you are using Discovery for Media for media recommendations, you need to switch to the media recommendations capability of Vertex AI Search. All of the existing data and models that you created with Discovery for Media will automatically appear in the Vertex AI Search and Conversation console, with the models appearing as apps. For more information, see Migrate from Discovery for Media to media recommendations.
October 13, 2023
Vertex AI Search: Customer-managed encryption key integration for the EU
Customer-managed encryption keys (CMEK) is available in the EU as an allowlisted preview feature.
If you store your data in an EU multi-region data store, you can provide your own encryption key to protect your data at rest.
For information, see Customer-managed encryption keys.
October 09, 2023
Vertex AI Search and Conversation: Renamed in the console and documentation
The Google Cloud console and the documentation at cloud.google.com have been updated to show the current product name for Vertex AI Search and Conversation. On the console, look for "Search and Conversation".
You might see the old name (Generative AI App Builder) in some places—for example, in the API reference.
September 29, 2023
Vertex AI Search (Enterprise Search): Customer-managed encryption key integration
Customer-managed encryption keys (CMEK) is available as an allowlisted preview feature.
If you store your data in a US multi-region data store, you can provide your own encryption key to protect your data at rest.
For information, see Customer-managed encryption keys.
Vertex AI Search (Enterprise Search): Search tuning
Search tuning is available as an allowlisted preview feature. You provide additional training data in the form of query and segment pairs. We use this data to tune the model for your app.
For information, see Improve search results with search tuning.
Vertex AI Search (Enterprise Search): VPC Service Controls are GA
Virtual Private Cloud Service Controls support for Enterprise Search is generally available (GA).
For more information, see Supported products and limitations in the VPC Service Controls documentation. For general information about VPC Service Controls, see Overview of VPC Service Controls.
Vertex AI Search (Enterprise Search): Data location
Vertex AI Search may be configured for data location pursuant to the "Data Location" section of the Service Specific Terms.
For information about data residency in Vertex AI Search, see Enterprise Search locations.
Vertex AI Search (Enterprise Search): Support for Access Transparency
Access Transparency supports Vertex AI Search in preview.
For more information, see Enable Access Transparency in Enterprise Search.
Vertex AI Search (Enterprise Search): Citations for search with follow-ups
Citations indicate from which search results specific sentences in the summary are taken.
For more information, see Configure the summary.
Vertex AI Search (Enterprise Search): Ignore adversarial queries and non-summary seeking queries for search with follow-ups
Ignore adversarial queries can stop generation of summaries that are unsafe or violate policy.
Non-summary seeking queries stop generation of summaries that aren't helpful for some queries.
For more information, see Configure the summary.
Vertex AI Search (Enterprise Search): Additional languages supported
Search, snippets, and other features are now supported in the following languages:
- Arabic
- Chinese (Simplified)
- Greek
- Hebrew
- Japanese
- Korean
- Polish
- Russian
See Languages.
September 22, 2023
Vertex AI Search (Enterprise Search): Third-party data connectors
You can set up your Vertex AI Search data stores to sync with data from Jira, Confluence, or Salesforce.
This feature is in private preview. To try this feature, contact your Google account team to find out if you qualify.
For more about setting up a connection to third-party data, see Create an Enterprise Search data store.
September 20, 2023
Manually refresh your web pages
Call the recrawlUris method to manually refresh specific web pages in a data store with Advanced website indexing turned on. You can check the status of the recrawl operation by polling the operations.get method.
September 19, 2023
Vertex AI Search (Enterprise Search): Turn Enterprise edition on or off
You can turn Enterprise edition features on or off for existing apps.
For more about Enterprise edition, see About advanced features.
September 15, 2023
Vertex AI Search (Enterprise Search): Languages for summarization
Summarization is supported in the following languages in addition to English:
- German (
de-DE) - Spanish (
es-ES) - Italian (
it-IT) - French (
fr-FR) - Dutch (
nl-NL) - Portuguese (
pt-BR) - Swedish (
sv-SE)
See Languages.
Vertex AI Search (Enterprise Search): Adjacent segments for preview with allowlist
When getting extractive segments, you can also get up to 3 segments from immediately before and after the relevant segment. Adjacent segments can add context and accuracy to the relevant segment. Turning on adjacent segments can increase latency.
Adjacent segments is in preview with allowlist. Contact your Google account team to try out adjacent segments.
See Extractive segments.
Vertex AI Search (Enterprise Search): Customizable summaries for preview with allowlist
When you request summaries, you can customize them by providing natural-language instructions. You can request customizations such as such as the length and level of detail, style of output (such as "simple"), language of output, focus of answer, and format (such as tables, bullets, and XML).
Customizable summaries are in preview with allowlist. Contact your Google account team if you're interested in trying this feature.
Vertex AI Search (Enterprise Search): ISO compliance
Vertex AI Search meets ISO 27001, ISO 27017, ISO 27018, and ISO 27701 compliance standards.
September 08, 2023
Vertex AI Search (Enterprise Search): Image search is GA
Image search is now generally available (GA).
Vertex AI Search (Enterprise Search): Page numbers for extractive segments
Page numbers can be returned with extractive segments. Page numbers indicate where an answer was extracted from in a document.
For more about extractive segments, see Get snippets and extracted content.
August 29, 2023
Vertex AI Search and Conversation is the new product name for Generative AI App Builder.
Generative AI App Builder: GA
Gen AI App Builder is publicly and generally available (GA).
Separation of data stores and apps
Data stores and apps are separate entities. In the console, you can see your apps on the Apps page and your data stores on the Data Stores page. You can create a data store and attach it to an app during app creation, or create a new data store during app creation.
Apps and data stores have a one-to-one relationship. Each app is associated with one data store; they can't be disconnected after you attach them. This change does not affect the functionality of your existing apps.
Enterprise Search: Search with follow-ups is GA
Previously, the search with follow-ups feature was called multi-turn search.
Search with follow-ups is now generally available (GA).
Search with follow-ups can be applied to websites if advanced website indexing is enabled and to unstructured data.
In addition, to improve consistency between search with an answer and search with follow-ups, the response from the conversations.converse method v1 provides the summary object and no longer provides the reply and references objects. The response from the v1beta version of the method remains unchanged.
SafeSearch is available with the search with follow-ups feature.
For general information about search with follow-ups, see Search with follow-ups.
Enterprise Search: Improvements to snippets, extractive answers, and extractive segments
Snippet status is now returned along with the snippet.
Extractive answers include the document page number where the answer was found.
Up to 10 extractive segments can be returned for a search result.
Enterprise Search: Multi-region support for US and EU locations
When you create a data store, you can specify global, the US multi-region, or the EU multi-region.
For more information including limitations associated with multi-region use, see Enterprise Search locations.
Enterprise Search: Languages
More features in more languages are supported for Enterprise Search.
See Languages.
Enterprise Search: Verify website domains
New requirement to verify your domain ownership for any websites in your data stores with advanced website indexing turned on.
Enterprise Search: Confidence scores
Confidence scores for extractive segments are available in preview with allowlist. Scores are based on the similarity of the query to the extracted segment.
See Extractive segments.
Enterprise Search: Serving controls using the API are allowlisted GA
Boost, filter, synonym, and redirect serving controls affect search results returned through API method calls.
For more information, see Configure serving controls.
Enterprise Search: Related questions
Related questions are available as an allowlisted, preview feature for search with follow-ups.
For information, see Related questions.
August 18, 2023
Enterprise Search: Multi-turn, conversational search UI
Multi-turn, conversational search from within the Google Cloud console is available in preview.
For information, see Configure search results for unstructured data and Search with multi-turn.
Enterprise Search: Languages
Search results and snippets are supported in:
- Danish (
da-DK) - Dutch (
nl-NL) - Hindi (
hi-IN) - Portuguese (
pt-BR) - Swedish (
sv-SE)
Search results and snippets are also supported in English (en-US), French (fr-FR), Spanish (es-ES), German (de-DE), and Italian (it-IT).
For more information, see Languages.
August 09, 2023
Enterprise Search: Support for VPC Service Controls
VPC Service Controls supports Enterprise Search in preview.
For more information, see Supported products and limitations in the VPC Service Controls documentation. For general information about VPC Service Controls, see Overview of VPC Service Controls.
August 04, 2023
Enterprise Search: Summaries are GA
Summaries for searches are now generally available (GA). These are a short summarization of the top one or more search results returned in a search response.
For more information, see Get search summaries.
Enterprise Search: Citations are GA
Summary citations are now generally available (GA). Citations indicate from which search results specific sentences in the summary are taken.
For more information, see Get citations.
Enterprise Search: Ignore adversarial and non-summary seeking queries are GA
The Ignore adversarial and non-summary seeking queries features are now generally available (GA).
For more information, see Ignore adversarial queries and Ignore non-summary seeking queries.
Enterprise Search: Languages
Search and snippets results are supported in French (fr-FR), in addition to English (en-US), Spanish (es-ES), German (de-DE), and Italian (it-IT).
For more information, see Languages.
August 02, 2023
Enterprise Search: Use custom embeddings
Bringing your own custom vector embeddings to Enterprise Search is supported in preview. If you've created your own embeddings on your data, you can upload them to Enterprise Search and use them when querying.
For more information, see Use custom embeddings.
July 20, 2023
Enterprise Search: Citations
Summary citations are now available in preview. Citations indicate from which search results specific sentences in the summary are taken.
For more information, see Get citations.
Enterprise Search: Ignore adverserial and non-summary seeking queries
You can now configure search requests so that adversarial queries and non-summary seeking queries do not include a summary in the response. This feature is in preview.
For more information, see Ignore adversarial queries and Ignore non-summary seeking queries.
Enterprise Search: Personalize
We have renamed the "Personalize" feature of Generative AI App Builder to "Recommendations". This is a naming change only. There is no change to product functionality.
July 10, 2023
Enterprise Search: Snippets no longer provide page numbers
Snippets are returned as brief extracts of text of uniform length. They no longer provide page numbers. If you previously used snippets as inputs for large language models to generate responses and summaries, we recommend using extractive answers or extractive segments. If you need page numbers in your extracts, we recommend using extractive answers, which provide page numbers where available.
Enterprise Search: Snippets with hit highlighting
Snippets in search responses are available with hit highlighting for rendering in UIs.
For more information, see Use snippets and extracted content.
Enterprise Search: Languages
Search and snippets results are supported in English (en-US), Spanish (es-ES), German (de-DE), and Italian (it-IT).
For more information, see Languages.
June 30, 2023
Enterprise Search: Image search
Search for images on your website. Provide an image (Base64 encoded PNG, JPG, or BMP) as the query and use an API method to return similar images.
For more information, see Search for images using a website search engine.
Enterprise Search: Extractive answers
Extractive answers are available in preview. An extractive answer is verbatim text that is returned with each search result. This text is extracted directly from the search result document and is typically displayed near the top of web pages to provide an end user with a brief answer that is contextually relevant to their query.
For more information, see Use snippets and extracted content.
Enterprise Search: Extractive segments
Extractive segments are available in preview. An extractive segment is verbatim text that is returned with each search result and is usually more verbose than an extractive answer. Extractive segments can be displayed as an answer to a query and be used in post-processing tasks and as input for large language models to generate answers or new text.
For more information, see Use snippets and extracted content.
Enterprise Search: OCR processing
Optical character recognition (OCR) processing is available, improving parsing and segmentation of PDF files at data ingestion. This enables Gen App Builder to take into account structures such as paragraphs and tables when ingesting your documents, providing more accurate search results, snippets, and summarizations. To request this functionality, contact your Google Cloud account team.
For more information, see OCR processing.
June 05, 2023
Enterprise Search: Specify domains for the widget
To use the widget externally, you must provide a list of domains where the widget appears.
If, in an earlier version, you created a widget with API key authentication, that widget might not be supported.
To ensure that the widget continues working, you must reconfigure it: Follow the instructions on Add the widget to a webpage to specify one or more domains for your widget and to generate a fresh snippet for the widget.
Enterprise Search: No API keys
Generated API keys are no longer used by the widget. You can enable Public Access instead.
Enterprise Search: Multi-turn search
Multi-turn search is available in allowlisted preview. Multi-turn search enables follow up questions in context. For information, see Search with multi-turn.
Enterprise Search: PPTX, DOCX, and TXT formats
Enterprise Search supports search over PPTX, DOCX, and TXT documents as well as HTML and PDF documents. Support for the PPTX, DOCX, and TXT formats is available in preview.
For general information about unstructured data, see Unstructured data.
Personalize
Like Enterprise Search and Infobot, Personalize is a component of Gen App Builder. With Personalize, you build a state-of-the-art recommendation engines based on your own data. The recommendation engine uses AI to suggest documents that are similar to the document that the user is currently viewing.
Personalize is available in preview.
For information, see Get started with Personalize.
Enterprise Search: Purge data
The addition of the Purge Data button on the Documents page makes it easier to delete data from a data store. For information, see Delete data from a data store.
Enterprise Search: Collect feedback from users
You have the option to collect feedback (thumbs up or thumbs down) about the quality of the search results provided through the widget. Users who don't like the results can also select, from a list, the reason for their dislike. Feedback collection is available in preview.
For information about feedback collection, see Configure widget feedback.
Enterprise Search: Schema editing
Schemas for structured data stores can be viewed and updated from within the Google Cloud console. Schema editing is available in preview.
For information about schema viewing and editing, see View the schema definition for structured data, Update a schema for structured data, and Schemas: auto-detecting versus providing your own.
Enterprise Search: HIPAA compliance
Enterprise Search is ready to support HIPAA compliance.
Enterprise Search: Analytics
On the analytics dashboard, you can compare metrics for two time periods. The analytics dashboard is available in preview.
For information, see View analytics.
Enterprise Search: Re-import after changing the indexable setting on a field
It is no longer necessary to re-import data after changing an indexable field setting. See Configure search attributes.
April 28, 2023
Enterprise Search: Image search
Search for images on your website. Provide a text string as the query and use an API method to return images. For information, see Search for images using a website search engine.
Enterprise Search: Purge data
The addition of the documents.purge method makes it easier to delete data.
For information, see Delete data from a datastore.
Enterprise Search: Advanced autocomplete
Experimental feature: The advanced autocomplete document data model uses Google's large language models to generate high-quality autocomplete suggestions. For information, see Experimental: Advanced autocomplete document data model.
April 14, 2023
Enterprise Search
Search engines for unstructured data can be created through the Google Cloud console. For information, see Console instructions: Create a search engine and ingest unstructured data.
March 29, 2023
Generative AI App Builder is available for preview to those on the allowlist.