Skip to main content
Version: 2.0

Query API Definition

Organizations often struggle to extract relevant information from large datasets and get meaningful answers to complex queries. The Vectara Query API offers a powerful and flexible solution for performing sophisticated searches across one or more corpora and significantly enhance the accuracy and relevance of search results.

The Query API lets you perform a query while defining its parameters that specify the query text, pagination details, metadata filters, and other search settings that enable application builders to tailor their queries to specific use cases.

After you index data into one or more corpora, you can run queries and display the results. This page provides a detailed reference for how to run queries and also describes some of Vectara's capabilities in metadata filtering, reranking, Retrieval Augmented Generation (RAG), and hybrid search.

Query Types​

The Vectara REST API 2.0 has different types of queries for you use depending on your search needs. Depending on the type, queries enable you to define parameters that control the behavior of the query and summarization:

  • Search Parameters: Filter data by metadata, apply lexical weighting, add additional context about the data, and rerank the results
  • Summarization Parameters: Choose model and prompt, and response settings like length and factual scoring, and even more nuanced model parameters
  • Stream Response: Optionally have the summarized response stream in real time.

The exact request format depends on the specific query type that you want to use.

tip

Check out our interactive API Reference that you experiment with these query types. You can also perform a simple single corpus query, advanced single corpus query, and a multiple corpora query.

Multiple Corpora Query​

The /v2/query endpoint allows you to perform Retrieval Augmented Generation (RAG) across one or more corpora in your account. You send a POST request in the body that specifies the following:

  • query - Contains your query text
  • stream_response - Indicates whether to stream the response in real-time (true) or to send a complete summary at the end of processing the request (false)
  • search - Specifies the search parameters
  • generation - Specifies the summarization parameters and generation_preset_name. Table
    summarization has a specific preset: vectara-summary-table-query-ext-dec-2024-gpt-3-5 or vectara-summary-table-query-ext-dec-2024-gpt-4o.
    • model_parameters - Specifies advanced summarization configuration parameters such as llm_name and max_tokens.
    • citations - Specifies the citation settings for the generator to use, including style, url_pattern, and text_pattern.
    • enable_factual_consistency_score - Indicates whether to return the factual consistency score with query results.
  • save_history - Indicates whether to save the query to query history.
  • intelligent_query_rewriting - Indicates whether to enable intelligent query rewriting.

Excluding this generation field disables summarization. The generation preset contains the name, description, llm_name, prompt_template, and other fields make up the preset.

This query type is useful when you want to query all your data sources at once.

Simple Single Corpus Query​

Send a simplified GET request to /v2/corpora/:corpus_key/query for querying a single corpus that specifies the following:

  • q: Contains the query string
  • limit: Specifies the maximum number of results
  • offset: Specifies the starting position in results

This query types provides a lightweight way to search a single corpus.

Advanced Single Corpus Query​

Send a POST request to /v2/corpora/:corpus_key/query to query a specific corpus with more advanced capabilities. This query type allows for detailed customization, including the ability to summarize table data when querying tables.

The request body is similar to the Query Corpora type and specifies the same parameters:

  • query - Contains your query text
  • stream_response - Indicates whether to stream the response in real-time or to send a complete summary at the end of processing the request
  • search - Specifies the search parameters
  • generation - Specifies the summarization parameters and generation_preset_name.
    • model_parameters - Specifies advanced summarization configuration parameters such as llm_name and max_tokens.
    • citations - Specifies the citation settings for the generator to use, including style, url_pattern, and text_pattern.
    • enable_factual_consistency_score - Indicates whether to return the factual consistency score with query results.
  • save_history - Indicates whether to save the query to query history.
  • intelligent_query_rewriting - Indicates whether to enable intelligent query rewriting.

Excluding this generation field disables summarization. The generation preset contains the name, description, llm_name, prompt_template, and other fields make up the preset.

This advanced type provides additional search filtering and customization options compared to the simple GET method.

The query response message encapsulates a single query result. It is a subdocument provided at indexing time. The text is the subdocument text, the score indicates how well the text answers the query (higher scores are better).

The metadata list holds any subdocument-level metadata that was stored with the item at indexing time. The corpus_key indicates which corpus the result came from: recall that a single query can execute against multiple corpora. While it's most often the case that a query is run against a single corpus, it's sometimes useful to run against several in parallel.

Finally, the document_index points at a specific document within the enclosing response set's document array. This is useful for retrieving the document id and document-level metadata.

Corpus Key Definition​

If you want to query a specific corpus or corpora, include the unique corpus_key in the path of the request such as v2/corpora/:corpus_key.

When creating a new corpus, you have the flexibility to specify a custom corpus_key that follows a naming convention of your choice. This allows you to assign easily identifiable keys to your corpora, making it easier to manage and reference them in your application.

caution

As part of the migration from API 1.0 to 2.0, all existing corpora have been assigned a new corpus_key based on their original name and corpus_id. The corpus_key is created by combining the name of the corpus (with underscores replacing spaces) and the original numeric ID.

Query Definition​

A single query consists of a query, which is specified in plain text. For example, "Where can I buy the latest iPhone?". Optionally, the query context provides additional information that the system may use to refine the results. For example, "The Apple store near my house is closed due to Covid."

Within the search object, add custom_dimensions weights (Pro or Enterprise), metadata_filter and set the lexical_interpolation (formerly lambda in the REST API v1.0). Setting to 0 disables exact and Boolean text matching, while a value of 1 disables neural retrieval. Users often see best results by setting this value somewhere between 0.01 and 0.1, and we typically recommend users start experimentation with a 0.025.

The semantics parameter indicates whether to consider a query against this corpus as a query or a response. The offset field controls the starting position within the list of results, while limit dictates how many results are returned. Thus, setting offset=5 and limit=20 would return twenty results beginning at position five. These fields are mainly used to provide pagination.

The context_configuration object lets you specify whether you want a specific number of characters or sentences before or after the matching document part.

Finally, the reranking configuration enables reranking of results, to further increase relevance in certain scenarios. For details about our Multilingual, Maximal Marginal Relevance (MMR), and User Defined Function rerankers, see Rerank Search Results.

Query Request and Response​

The corpus_key specifies the ID of the corpus being searched. The metadata_filter allows specifying a predicate expression that restricts the search to a part of the corpus. The filter is written in a simplified SQL dialect and can reference metadata that was marked as filterable during corpus creation.

note

See the Filter Expressions Overview for a description of their syntax, and Corpus Administration to learn how referenceable metadata is specified during corpus creation.

By default, Vectara only uses its neural/semantic retrieval model, and does not attempt to use keyword matching. To enable hybrid search with a mix of both keyword and neural results, edit the lexical_interpolation value.

If the corpus specifies custom dimensions (Pro or Enterprise), weights can be assigned to each dimension as well.

Finally, it's possible to override the semantic interpretation of the query string. Usually, the default settings for the corpus are sufficient. In more advanced scenarios, it's desirable to force it to be treated as a query, or, more rarely, as a response.

Reranking Configuration​

The reranker object enables the reranking of query results, to further increase relevance in certain scenarios. The cutoff property lets you specify a minimum score threshold for search results to be included after reranking. The limit property of this reranker object allows you to have more granular control over the number of results returned after reranking. For more details about these properties, see Rerank Search Results:

  • Specify the type as customer_reranker and reranker_name as Rerank_Multilingual_v1 to use the Multilingual Reranker v1, also known as Slingshot.
  • Specify the type as mmr to use the Maximal Marginal Relevance (MMR) Reranker. This reranker lets you specify a diversity_bias value between 0.0 and 1.0.
  • Specify the type as userfn to use the User Defined Function Reranker.
  • Specify the type as chain to use the Chain Reranker.
  • To use Knee Reranking, configure the chain reranker to first use the Vectara Multilingual Reranker (Slingshot). Then specify the type as userfn and user_function as knee() to enable Knee Reranking in the chain.
  • If you do not want to use a reranker, set the type to none.

Query Summarization Request - Retrieval Augmented Generation​

To use Retrieval Augmented Generation (RAG), which Vectara also refers to as "Grounded Generation" -- our groundbreaking way of producing generative summaries on top of your own data -- you can submit a generation that attempts to answer the end-user's question, citing the results as references. For more information, read about Retrieval Augmented Generation.

The generation object enables you to tailor the results of the query summarization. Growth users can specify the max_summarized_results, response_language, and enable_factual_consistency_score.

"generation": {
    "generation_preset_name": "vectara-summary-ext-v1.2.0",
    "max_used_search_results": 5,
    "prompt_template": "[\n  {\"role\": \"system\", \"content\": \"You are a helpful search assistant.\"},\n  #foreach ($qResult in $vectaraQueryResults)\n     {\"role\": \"user\", \"content\": \"Given the $vectaraIdxWord[$foreach.index] search result.\"},\n     {\"role\": \"assistant\", \"content\": \"${qResult.getText()}\" },\n  #end\n  {\"role\": \"user\", \"content\": \"Generate a summary for the query '\''${vectaraQuery}'\'' based on the above results.\"}\n]\n",
    "max_response_characters": 300,
    "response_language": "eng",
    "model_parameters": {
      "max_tokens": 0,
      "temperature": 0,
      "frequency_penalty": 0,
      "presence_penalty": 0
    },

Users also have access to advanced summarization customization options.

Generation Presets​

The generation-preset-name field in generation object specifies the prompt template to use. Generation presets bundle several properties that configure generation for the request, providing more flexibility in how parameters are set. The preset includes the prompt_template, the llm_name, and other settings like max_tokens and temperature.

"generation-preset-name": "vectara-summary-ext-v1.3.0"

To view available generation presets, use the List Generation Presets API.

note

The generation_preset_name field replaces the prompt_name field that was previously in the generation object. The prompt_name field is now deprecated but still supported for backward compatibility.

Mockingbird: Enhanced RAG Performance​

For users seeking superior RAG performance, Vectara offers Mockingbird, our advanced LLM specifically designed for RAG tasks.

To use Mockingbird for your RAG tasks, specify mockingbird-1.0-2024-07-16 in the generation_preset_name field the generation object, like in this example:

{
  "generation": {
    "generation_preset_name": "mockingbird-1.0-2024-07-16",
    "max_used_search_results": 5,
    "response_language": "eng",
    "enable_factual_consistency_score": true
  }
}

Mockingbird is particularly beneficial for enterprise applications requiring high-quality summaries and structured outputs. For more details on Mockingbird's capabilities and performance, see the Mockingbird LLM section.

Factual Consistency Score​

The Factual Consistency Score, based on a more advanced version of Hughes Hallucination Evaluation Model (HHEM), enables you to evaluate the likelihood of an AI-generated summary being factually correct based on search results. This calibrated score can range from 0.0 to 1.0. A higher scores indicates a greater probability of being factually accurate, while a lower score indicates a greater probability of hallucinations. It also supports English, German, and French (eng, deu, fra) as the response_language.

In your summarization request, set the enable_factual_consistency_score field to true. The Factual Consistency Score returns a calibrated value in the factual_consistency_score field of the summary message. The score field contains the value between 0.0 and 1.0.

For example, a score of 0.95 suggests a 95% likelihood that the summary is free of hallucinations and would align with the original content. A lower score of 0.40 indicates a 40% chance which would be probably much less factually accurate. We suggest starting with a setting of 0.5 as an initial guideline for cutoffs between good and bad.

Citation Format in Summary​

When generating a summary, Vectara enables users to format the style of citations object with one of the following formats:

  • numeric (default) - Citations appear as numbers [1], [2], [N], and so on.
  • none - No citations appear in the summary.
  • html - Citations appears as a URL: <a href="https://my.doc/foo">[N]</a>
  • MARKDOWN - Citations appears in Markdown: [N](https://my.doc/foo)

If set to html or markdown, you must customize the citation using both of the url_pattern and text_pattern fields to enable dynamic citation generation. Both of these parameters can access all part and document level metadata fields.

For example, the url_pattern field can specify {doc.id} and {part.page} metadata as https://mypdf.doc/foo/{doc.id}#page={part.page}. The text_pattern field specifies the document and part metadata name in curly braces. For example, use {doc.title} and the final result appears as Title.

To use citations, you must specify one of the following summarizers in generation_preset_name:

  • mockingbird-1.0-2024-07-16 - (Vectara's Mockingbird LLM)
  • vectara-summary-ext-24-05-sml - (gpt-3.5-turbo)
  • vectara-summary-ext-24-05-med-omni - (gpt-4o)
  • vectara-summary-ext-24-05-med - (gpt-4.0)
  • vectara-summary-ext-24-05-large - (gpt-4.0-turbo)
note

For more information, see the documentation about selecting summarizers.

Default Citation Behavior​

  • If text_pattern is not specified, it defaults to the numerical position of the result ([1], [2], [N].).
  • The url_pattern does not have a default, so this field must be explicitly defined.

Citation Example​

In this example, you want Vectara to say as seen in [Document-Title] with a link to the specific page:

{
  "citations": {
    "style": "MARKDOWN",
    "url_pattern": "{doc.id}#page={section.page}",
    "text_pattern": "as seen in {doc.title}"
  }
}

The response will look something like this:

In the Metropolitan Transportation Authority (MTA) rules, it is prohibited to
destroy, mark, soil, paint, draw, inscribe, or place graffiti on any facility
or conveyance of the authority [as seen in Rules of Conduct and
Fines](https://new.mta.info/document/36821#page=3).

Disable query summarization​

To disable summarization, exclude the generation object from a query.

Advanced Summarization Customization Options​

Vectara also provides more powerful summarization capabilities for tailoring summarizations to specific application and user needs. If they need to change generation beyond what the preset specifies, you can override most parameters in a query. For example:

{
  "generation": {
    "generation-preset-name": "vectara-summary-ext-v1.3.0",
    "max_tokens": 300,
    "temperature": 0.7
  }
}

To provide even more customization beyond, you can override certain parameters in your query. This enables you to use a preset as a starting point while you tailor specific aspects of the generation.

The generation_preset_name allows you to specify one of our available summarizers. Use generation_preset_name and prompt_template to override the default prompt with a custom prompt. Your use case might require a chatbot to be more human like, so you decide to create a custom response format that behaves more playfully in a conversation or summary.

The max_response_characters lets you control the length of the summary, but note that it is not a hard limit like with the max_tokens parameter. The model_parameters object provides even more fine-grained controls for the summarizer model:

  • llm_name specifies the name of the LLM model to use for summarization. If specified, it overrides the model behind generation_preset_name.
  • max_tokens specifies a hard limit on the number of characters in a response. This value supercedes the responseChars parameter in the summary object.
  • temperature indicates whether you want the summarization to not be creative at all 0.0, or for the summarization to take more creative liberties as you approach the maximum value of 1.0.
  • frequency_penalty provides even more granular control to help ensure that the summarization decreases the likelihood of repeating words. The values range from 0.0 to 1.0
  • presence_penalty provides more control over whether you want the summary to include new topics. The values also range from 0.0 to 1.0.

By leveraging these advanced capabilities, application builders can fine-tune the behavior and output style of the summarizer to align with your unique application requirements.

REST 2.0 URL​

Query API Endpoint Address​

Vectara exposes a REST endpoint at the following URL to search content from a corpus: https://api.vectara.io/v2/query

The API Reference shows the full Query REST definition.