@convo-lang/convo-lang/src/lib/convo-rag-types.d.ts

The language of AI

import type { Conversation } from "./Conversation.js";
import type { ConvoTokenUsage, FlatConvoConversation, FlatConvoMessage } from "./convo-types.js";
export type ConvoVectorValue = string | number[] | Uint8Array;
export interface ConvoDocumentReference {
    content: string;
    id?: string;
    name?: string;
    url?: string;
    pageIndex?: number;
    pageEndIndex?: number;
    lineIndex?: number;
    lineEndIndex?: number;
    charIndex?: number;
    charEndIndex?: number;
    contentType?: string;
    vector?: string | number[] | Uint8Array;
    path?: string;
    metadata?: Record<string, any>;
}
export interface ConvoRagContext {
    params: Record<string, any>;
    tolerance: number;
    lastMessage: FlatConvoMessage;
    flat: FlatConvoConversation;
    conversation: Conversation;
}
export type ConvoRagCallback = (ragContext: ConvoRagContext) => ConvoDocumentReference | null | (ConvoDocumentReference | null)[] | Promise<ConvoDocumentReference | null | (ConvoDocumentReference | null)[]>;
export interface ConvoRagSearch {
    /**
     * The content to search against.
     */
    content?: string;
    tolerance: number;
    limit?: number;
    /**
     * If true the embeddings property of returned document should include their embeddings. In most
     * cases this is not needed and just added more overhead memory usage.
     */
    includeVector?: boolean;
    paths?: string[];
}
export interface ConvoRagSearchResultItem {
    id?: string;
    distance?: number;
    document: ConvoDocumentReference;
}
export interface ConvoRagSearchResult {
    items: ConvoRagSearchResultItem[];
    usage: ConvoTokenUsage;
}
export interface ConvoRagService {
    /**
     * Searches to for matching document references
     */
    searchAsync(search: ConvoRagSearch): Promise<ConvoRagSearchResult>;
    /**
     * Upserts new documents in to the rag service.
     */
    upsertAsync?(documents: ConvoDocumentReference[]): Promise<void>;
}