Skip to content

RetrieverAgent

Description: A RetrieverAgent is an agent that will take a query, rewrite it for better similarity vector search, and then perform the vector search on the document store. The agent will dynamically load and embed files, so it is not performant for loading large bodies of files, but performs very well for small to medium-sized document stores (hundreds to thousands of documents) which are updating frequently.

PropertyPatternTypeDeprecatedDefinitionTitle/Description
- implementationNoconstNo-RetrieverAgent
- max_num_resultsNointegerNo-Max Num Results
- question_transformerNoReference[QuestionTransformer]No-QuestionTransformer
- document_retrieverNoReference[DocumentRetriever]No-DocumentRetriever
- document_rerankerNoReference[DocumentReranker]No-DocumentReranker
- result_summarizerNoReference[ResultSummarizer]No-ResultSummarizer
+ nameNostringNo-Name
+ descriptionNostringNo-Description
- loader_root_locationNoCombinationNo-Loader Root Location
- loader_patternNoCombinationNo-Loader Pattern
- document_managerNoCombinationNo--
- apuNoReference[APU]No-APU

1. Property implementation

Typeconst
RequiredNo

Description: RetrieverAgent

Specific value: "RetrieverAgent"

2. Property max_num_results

Title: Max Num Results

Typeinteger
RequiredNo
Default10

Description: The maximum number of results to consider.

3. Property question_transformer

Title: QuestionTransformer

Type[Reference[QuestionTransformer]](/docs/components/questiontransformer/overview)
RequiredNo
Default{"implementation": "MultiQuestionTransformer"}

Description: Overview of QuestionTransformer components

4. Property document_retriever

Title: DocumentRetriever

Type[Reference[DocumentRetriever]](/docs/components/documentretriever/overview)
RequiredNo
Default{"implementation": "SimilarityMemoryRetriever"}

Description: Overview of DocumentRetriever components

5. Property document_reranker

Title: DocumentReranker

Type[Reference[DocumentReranker]](/docs/components/documentreranker/overview)
RequiredNo
Default{"implementation": "RAGFusionReranker"}

Description: Overview of DocumentReranker components

6. Property result_summarizer

Title: ResultSummarizer

Type[Reference[ResultSummarizer]](/docs/components/resultsummarizer/overview)
RequiredNo
Default{"implementation": "ResultSummarizer"}

Description: Overview of ResultSummarizer components

7. Property name

Title: Name

Typestring
RequiredYes

Description: The name of the document store to use.

8. Property description

Title: Description

Typestring
RequiredYes

Description: A detailed description of the the retriever including all necessary information for the calling agent to decide to call this agent, i.e. file type or location or etc…

9. Property loader_root_location

Title: Loader Root Location

Typecombining
RequiredNo
Additional properties[Any type: allowed]
Defaultnull

Description: A URL specifying the root location of the loader.

Any of(Option)
item 0
item 1

9.1. Property item 0

Typestring
RequiredNo

9.2. Property item 1

Typenull
RequiredNo

10. Property loader_pattern

Title: Loader Pattern

Typecombining
RequiredNo
Additional properties[Any type: allowed]
Default"**/*"

Description: The search pattern to use when loading files.

Any of(Option)
item 0
item 1

10.1. Property item 0

Typestring
RequiredNo

10.2. Property item 1

Typenull
RequiredNo

11. Property document_manager

Typecombining
RequiredNo
Additional properties[Any type: allowed]
Defaultnull
Any of(Option)
DocumentManager
item 1

11.1. Property DocumentManager

Title: DocumentManager

Type[Reference[DocumentManager]](/docs/components/documentmanager/overview)
RequiredNo

Description: Overview of DocumentManager components

11.2. Property item 1

Typenull
RequiredNo

12. Property apu

Title: APU

TypeReference[APU]
RequiredNo
Default{"implementation": "ConversationalAPU"}

Description: The APU is the main interface for the Agent to interact with the LLM. The APU provides a set of capabilities that encapsulate LLM functionality and creates a clear separation between business logic and the underlying LLM implementation.

To learn more, check out our blog article APU: What is it and how does it work?.