@chainlink/mcp-server/dist/vectordb/schema.js

Prototype MCP Server for CLL

"use strict";
/**
 * @fileoverview Vector database schema definitions and interfaces
 *
 * Defines the structure for document storage in the vector database using
 * Apache Arrow schema format. Supports configurable embedding dimensions
 * for different providers and includes metadata for document management.
 */
Object.defineProperty(exports, "__esModule", { value: true });
exports.CCIPDocTableSchema = exports.EMBEDDING_DIMENSIONS = void 0;
exports.createCCIPDocTableSchema = createCCIPDocTableSchema;
const apache_arrow_1 = require("apache-arrow");
/** Metadata structure for document entries */
const metadataStruct = new apache_arrow_1.Struct([
    new apache_arrow_1.Field("sourceDocId", new apache_arrow_1.Utf8(), true),
    new apache_arrow_1.Field("sourceDocFilename", new apache_arrow_1.Utf8(), true),
    new apache_arrow_1.Field("sourceDocHash", new apache_arrow_1.Utf8(), true),
    new apache_arrow_1.Field("product", new apache_arrow_1.Utf8(), true), // ccip, functions, etc.
]);
/**
 * Embedding dimensions by provider
 *
 * Defines the vector dimensions for different embedding providers.
 * These must match the actual dimensions output by each provider's models.
 *
 * @constant EMBEDDING_DIMENSIONS
 */
exports.EMBEDDING_DIMENSIONS = {
    /** OpenAI text-embedding-3-small dimensions */
    openai: 1536,
    /** Ollama nomic-embed-text dimensions */
    ollama: 768,
};
/**
 * Create a vector field with configurable dimensions
 *
 * @param dimensions - Number of dimensions for the embedding vectors
 * @returns Apache Arrow Field configured for vector storage
 */
function createVectorField(dimensions) {
    return new apache_arrow_1.Field("vector", new apache_arrow_1.FixedSizeList(dimensions, new apache_arrow_1.Field("item", new apache_arrow_1.Float32())), false);
}
/**
 * Create schema for CCIP document table with configurable vector dimensions
 *
 * The schema includes all necessary fields for document storage:
 * - Unique chunk identifier
 * - Metadata (source, product)
 * - Vector embedding (configurable dimensions)
 * - Chunk type and content
 * - Content hash for deduplication
 *
 * @param embeddingDimensions - Number of dimensions for embedding vectors
 * @returns Apache Arrow Schema for document table
 */
function createCCIPDocTableSchema(embeddingDimensions) {
    return new apache_arrow_1.Schema([
        new apache_arrow_1.Field("chunkId", new apache_arrow_1.Utf8(), false),
        new apache_arrow_1.Field("metadata", metadataStruct, false),
        createVectorField(embeddingDimensions),
        new apache_arrow_1.Field("chunkType", new apache_arrow_1.Utf8(), false),
        new apache_arrow_1.Field("chunkText", new apache_arrow_1.Utf8(), false),
        new apache_arrow_1.Field("chunkHash", new apache_arrow_1.Utf8(), false),
    ]);
}
/**
 * Default schema for backwards compatibility (OpenAI dimensions)
 *
 * @constant CCIPDocTableSchema
 */
exports.CCIPDocTableSchema = createCCIPDocTableSchema(exports.EMBEDDING_DIMENSIONS.openai);
//# sourceMappingURL=schema.js.map