Member Junction
    Preparing search index...

    Class RerankerService

    Service for managing reranker instances and performing note reranking. Implements singleton pattern for efficient reranker caching.

    Key responsibilities:

    • Parse RerankerConfiguration JSON from agent settings
    • Instantiate and cache reranker instances by model ID
    • Perform two-stage retrieval (vector search -> reranking)
    • Handle graceful fallback when reranking fails

    Usage:

    const service = RerankerService.Instance;
    const config = service.parseConfiguration(agent.RerankerConfiguration);

    if (config?.enabled) {
        const result = await service.rerankNotes(
            vectorSearchResults,
            userQuery,
            config,
            contextUser
        );
    }

    Hierarchy (View Summary)

    Index

    Constructors

    constructor

    Accessors

    GlobalKey Instance

    Methods

    clearCache ClearCache GetGlobalObjectStore getReranker GetReranker parseConfiguration RerankExamples rerankNotes RerankNotes getInstance

    Constructors

    Accessors

    • get GlobalKey(): string

      Returns string

    • get Instance(): RerankerService

      Get the singleton instance of RerankerService

      Returns RerankerService

    Methods

    • Returns void

      Use ClearCache.

    • Clear the reranker cache. Useful for testing or when models are updated.

      Returns void

    • The Global Object Store is a place to store global objects that need to be shared across the application. Depending on the execution environment, this could be the window object in a browser, or the global object in a node environment, or something else in other contexts. The key here is that in some cases static variables are not truly shared because it is possible that a given class might have copies of its code in multiple paths in a deployed application. This approach ensures that no matter how many code copies might exist, there is only one instance of the object in question by using the Global Object Store.

      Returns GlobalObjectStore

    • Parameters

      • modelID: string
      • contextUser: UserInfo
      • OptionalpromptID: string

      Returns Promise<BaseReranker>

      Use GetReranker.

    • Get or create a reranker instance for the specified model. Caches instances for reuse across multiple calls.

      Parameters

      • modelID: string

        ID of the AIModel with type='Reranker'

      • contextUser: UserInfo

        User context for operations

      • OptionalpromptID: string

        The prompt a prompt-backed reranker runs. LLMReranker requires its chat prompt's ID. For DecisionReranker it is optional: without one it asks the decision prompt its model-vendor row's APIName names, as AIRerankerRunner does, or Default Decision when that names no prompt.

      Returns Promise<BaseReranker>

      Reranker instance or null if unavailable

      Builds a driver directly, so the call gets no failover and writes no MJ: AI Prompt Runs row. Use AIRerankerRunner.RunRerank, which RerankNotes uses.

    • Parse RerankerConfiguration JSON from agent settings. Returns null if configuration is missing or invalid.

      Parameters

      • configJson: string

        JSON string from AIAgent.RerankerConfiguration

      Returns RerankerConfiguration

      Parsed configuration with defaults applied, or null if disabled/invalid

    • Rerank agent examples using the configured reranker: the examples stage that RerankerConfiguration.rerankExamples turns on. It uses the same model, prompt and minRelevanceThreshold as RerankNotes. Each document is the example's input and output. With an agent run in options, it records a Rerank Examples step linked to the rerank's run, as RerankNotes records Rerank Notes.

      Like RerankNotes, this method throws when reranking fails. The calling code decides, from config.fallbackOnError, whether to fall back to the vector search results.

      Parameters

      Returns Promise<ExampleMatchResult[]>

      The examples at or above the threshold, most relevant first, each with its rerank score as similarity

      Error if reranking fails

    • Rerank notes using the configured reranker. Implements two-stage retrieval for semantic relevance ranking.

      Reranking goes through AIRerankerRunner, pinned to config.rerankerModelId, so every call writes an MJ: AI Prompt Runs row. The observability step, when there is one, links to that row through its TargetLogID.

      IMPORTANT: This service does NOT handle fallback logic. If reranking fails, this method throws. The calling code (agent class) decides, from config.fallbackOnError, whether to fall back to the original vector search results.

      Parameters

      Returns Promise<RerankServiceResult>

      Reranked notes sorted by relevance

      Error if reranking fails

    • Returns the singleton instance of the class. If the instance does not exist, it is created and stored in the Global Object Store. If className is provided it will be used as part of the key in the Global Object Store, otherwise the actual class name will be used. NOTE: the class name used by default is the lowest level of the object hierarchy, so if you have a class that extends another class, the lowest level class name will be used.

      Type Parameters

      Parameters

      • this: new () => T
      • OptionalclassName: string

      Returns T

    Settings

    Member Visibility

    On This Page

    Constructors
    Accessors
    Methods