hikma-engine/dist/persistence/db/vector.js

Code Knowledge Graph Indexer - A sophisticated TypeScript-based indexer that transforms Git repositories into multi-dimensional knowledge stores for AI agents

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });
exports.storeVector = storeVector;
exports.vectorSearch = vectorSearch;
exports.semanticSearch = semanticSearch;
exports.batchStoreEmbeddings = batchStoreEmbeddings;
const error_handling_1 = require("../../utils/error-handling");
const error_handling_2 = require("../../utils/error-handling");
async function storeVector(client, table, column, recordId, embedding) {
    const db = client.getDb();
    const embeddingBlob = Buffer.from(new Float32Array(embedding).buffer);
    const sql = `UPDATE ${table} SET ${column} = ? WHERE id = ?`;
    try {
        db.prepare(sql).run(embeddingBlob, recordId);
    }
    catch (error) {
        throw new error_handling_1.DatabaseOperationError('SQLite', 'storeVector', (0, error_handling_2.getErrorMessage)(error), error);
    }
}
async function vectorSearch(client, table, column, queryEmbedding, limit = 10, threshold) {
    const db = client.getDb();
    const queryBlob = Buffer.from(new Float32Array(queryEmbedding).buffer);
    let sql = `
    SELECT id,
           vec_distance_cosine(${column}, ?) as similarity,
           *
    FROM ${table}
    WHERE ${column} IS NOT NULL
  `;
    const params = [queryBlob];
    if (threshold !== undefined) {
        const distanceThreshold = 1 - threshold;
        sql += ` AND vec_distance_cosine(${column}, ?) <= ?`;
        params.push(queryBlob, distanceThreshold);
    }
    sql += ` ORDER BY similarity ASC LIMIT ?`;
    params.push(limit);
    try {
        const results = db.prepare(sql).all(...params);
        return results.map((row) => ({
            id: row.id,
            similarity: 1 - row.similarity,
            data: row,
        }));
    }
    catch (error) {
        throw new error_handling_1.DatabaseOperationError('SQLite', 'vectorSearch', (0, error_handling_2.getErrorMessage)(error), error);
    }
}
async function semanticSearch(client, queryEmbedding, limit = 5) {
    const [files, functions, commits, pullRequests] = await Promise.all([
        vectorSearch(client, 'files', 'content_embedding', queryEmbedding, limit),
        vectorSearch(client, 'functions', 'signature_embedding', queryEmbedding, limit),
        vectorSearch(client, 'commits', 'message_embedding', queryEmbedding, limit),
        vectorSearch(client, 'pull_requests', 'title_embedding', queryEmbedding, limit),
    ]);
    return { files, functions, commits, pullRequests };
}
async function batchStoreEmbeddings(client, table, column, records) {
    const db = client.getDb();
    const stmt = db.prepare(`UPDATE ${table} SET ${column} = ? WHERE id = ?`);
    let success = 0;
    let failed = 0;
    const errors = [];
    const updateMany = db.transaction((recordList) => {
        for (const record of recordList) {
            try {
                const embeddingBlob = Buffer.from(new Float32Array(record.embedding).buffer);
                stmt.run(embeddingBlob, record.id);
                success++;
            }
            catch (error) {
                failed++;
                const errorMsg = `Failed to store embedding for ${record.id}: ${(0, error_handling_2.getErrorMessage)(error)}`;
                errors.push(errorMsg);
            }
        }
    });
    try {
        updateMany(records);
    }
    catch (error) {
        throw new error_handling_1.DatabaseOperationError('SQLite', 'batchStoreEmbeddings', (0, error_handling_2.getErrorMessage)(error), error);
    }
    return { success, failed, errors };
}