aiwcli 0.12.6 → 0.12.8

package/dist/templates/cc-native/_cc-native/lib-ts/rlm/vector-store.ts

@@ -1,278 +1,278 @@
-/**
- * Vector store using SQLite + sqlite-vec for KNN embedding search.
- *
- * Single-file DB at ~/.claude/rlm-vectors.db.
- * Uses bun:sqlite with the sqlite-vec extension for vector similarity search.
- */
-
-import { Database } from "bun:sqlite";
-import * as sqliteVec from "sqlite-vec";
-import { RLM_VECTOR_DB_PATH, EMBED_DIMENSIONS, type VectorSearchResult } from "./types.js";
-import { logDebug, logInfo } from "./logger.js";
-
-const HOOK_NAME = "rlm_vector";
-
-export interface ChunkRow {
-  session_id: string;
-  project: string;
-  date: string;
-  segment_index: number;
-  line_start: number;
-  line_end: number;
-  topic: string;
-  chunk_text: string;
-  source_path: string;
-  embedding: Float32Array;
-}
-
-export interface VectorStats {
-  session_count: number;
-  chunk_count: number;
-}
-
-/**
- * Open the vector DB, load sqlite-vec extension, create schema, set WAL mode.
- */
-export function openVectorDb(path?: string): Database {
-  const dbPath = path ?? RLM_VECTOR_DB_PATH;
-  const db = new Database(dbPath);
-  sqliteVec.load(db);
-  db.run("PRAGMA journal_mode=WAL");
-
-  db.run(`
-    CREATE TABLE IF NOT EXISTS embedded_sessions (
-      session_id TEXT NOT NULL,
-      project TEXT NOT NULL,
-      source_mtime INTEGER NOT NULL,
-      chunk_count INTEGER NOT NULL,
-      embedded_at TEXT NOT NULL,
-      PRIMARY KEY (session_id, project)
-    )
-  `);
-
-  // vec0 virtual table for KNN search
-  db.run(`
-    CREATE VIRTUAL TABLE IF NOT EXISTS chunks USING vec0(
-      embedding float[${EMBED_DIMENSIONS}],
-      project text,
-      date text,
-      +session_id text,
-      +segment_index integer,
-      +line_start integer,
-      +line_end integer,
-      +topic text,
-      +chunk_text text,
-      +source_path text
-    )
-  `);
-
-  logDebug(HOOK_NAME, `Opened vector DB at ${dbPath}`);
-  return db;
-}
-
-/**
- * Insert chunks in a single transaction.
- */
-export function insertChunks(db: Database, chunks: ChunkRow[]): void {
-  if (chunks.length === 0) return;
-
-  const stmt = db.prepare(`
-    INSERT INTO chunks (
-      embedding, project, date,
-      session_id, segment_index, line_start, line_end,
-      topic, chunk_text, source_path
-    ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
-  `);
-
-  const tx = db.transaction(() => {
-    for (const chunk of chunks) {
-      stmt.run(
-        new Uint8Array(chunk.embedding.buffer),
-        chunk.project,
-        chunk.date,
-        chunk.session_id,
-        chunk.segment_index,
-        chunk.line_start,
-        chunk.line_end,
-        chunk.topic,
-        chunk.chunk_text,
-        chunk.source_path,
-      );
-    }
-  });
-
-  tx();
-  logDebug(HOOK_NAME, `Inserted ${chunks.length} chunks`);
-}
-
-/**
- * Mark a session as embedded (upsert).
- */
-export function markSessionEmbedded(
-  db: Database,
-  sessionId: string,
-  project: string,
-  mtime: number,
-  count: number,
-): void {
-  db.run(
-    `INSERT OR REPLACE INTO embedded_sessions (session_id, project, source_mtime, chunk_count, embedded_at)
-     VALUES (?, ?, ?, ?, ?)`,
-    [sessionId, project, mtime, count, new Date().toISOString()],
-  );
-}
-
-/**
- * Check if a session is already embedded at the given mtime.
- */
-export function isSessionEmbedded(
-  db: Database,
-  sessionId: string,
-  project: string,
-  mtime: number,
-): boolean {
-  const row = db.query(
-    `SELECT 1 FROM embedded_sessions WHERE session_id = ? AND project = ? AND source_mtime = ?`,
-  ).get(sessionId, project, mtime);
-  return row !== null && row !== undefined;
-}
-
-/**
- * Delete all chunks for a session (for re-indexing).
- */
-function isRowidResult(obj: unknown): obj is { rowid: number } {
-  return typeof obj === "object" && obj !== null && "rowid" in obj && typeof (obj as { rowid: unknown }).rowid === "number";
-}
-
-export function deleteSessionChunks(
-  db: Database,
-  sessionId: string,
-  project: string,
-): void {
-  // vec0 tables support DELETE with rowid ranges, but we need to find matching rowids first
-  const rawRows = db.query(
-    `SELECT rowid FROM chunks WHERE session_id = ? AND project = ?`,
-  ).all(sessionId, project);
-
-  const rows = (Array.isArray(rawRows) ? rawRows : []).filter(isRowidResult);
-
-  if (rows.length > 0) {
-    const tx = db.transaction(() => {
-      for (const row of rows) {
-        db.run(`DELETE FROM chunks WHERE rowid = ?`, [row.rowid]);
-      }
-    });
-    tx();
-    logDebug(HOOK_NAME, `Deleted ${rows.length} chunks for ${sessionId}`);
-  }
-
-  db.run(
-    `DELETE FROM embedded_sessions WHERE session_id = ? AND project = ?`,
-    [sessionId, project],
-  );
-}
-
-function isSearchResultRow(obj: unknown): obj is {
-  rowid: number;
-  distance: number;
-  project: string;
-  date: string;
-  session_id: string;
-  segment_index: number;
-  line_start: number;
-  line_end: number;
-  topic: string;
-  source_path: string;
-} {
-  if (typeof obj !== "object" || obj === null) return false;
-  const r = obj as Record<string, unknown>;
-  return (
-    typeof r.rowid === "number" &&
-    typeof r.distance === "number" &&
-    typeof r.project === "string" &&
-    typeof r.date === "string" &&
-    typeof r.session_id === "string" &&
-    typeof r.segment_index === "number" &&
-    typeof r.line_start === "number" &&
-    typeof r.line_end === "number" &&
-    typeof r.topic === "string" &&
-    typeof r.source_path === "string"
-  );
-}
-
-/**
- * KNN search for the closest chunks to a query embedding.
- */
-export function searchKnn(
-  db: Database,
-  queryEmbedding: Float32Array,
-  topK: number,
-  projectFilter?: string,
-): VectorSearchResult[] {
-  const queryBytes = new Uint8Array(queryEmbedding.buffer);
-
-  let sql: string;
-  let params: unknown[];
-
-  if (projectFilter) {
-    sql = `
-      SELECT rowid, distance, project, date,
-             session_id, segment_index, line_start, line_end,
-             topic, source_path
-      FROM chunks
-      WHERE embedding MATCH ? AND k = ? AND project = ?
-      ORDER BY distance
-    `;
-    params = [queryBytes, topK, projectFilter];
-  } else {
-    sql = `
-      SELECT rowid, distance, project, date,
-             session_id, segment_index, line_start, line_end,
-             topic, source_path
-      FROM chunks
-      WHERE embedding MATCH ? AND k = ?
-      ORDER BY distance
-    `;
-    params = [queryBytes, topK];
-  }
-
-  const rawRows = db.query(sql).all(...params);
-  const rows = (Array.isArray(rawRows) ? rawRows : []).filter(isSearchResultRow);
-
-  return rows.map((r) => ({
-    chunk_id: r.rowid,
-    session_id: r.session_id,
-    project: r.project,
-    segment_index: r.segment_index,
-    line_start: r.line_start,
-    line_end: r.line_end,
-    topic: r.topic,
-    date: r.date,
-    source_path: r.source_path,
-    distance: r.distance,
-  }));
-}
-
-function isCountResult(obj: unknown): obj is { cnt: number } {
-  return typeof obj === "object" && obj !== null && "cnt" in obj && typeof (obj as { cnt: unknown }).cnt === "number";
-}
-
-/**
- * Get counts of embedded sessions and chunks.
- */
-export function getStats(db: Database): VectorStats {
-  const sessionsRaw = db.query(
-    `SELECT COUNT(*) as cnt FROM embedded_sessions`,
-  ).get();
-  const chunksRaw = db.query(
-    `SELECT COUNT(*) as cnt FROM chunks`,
-  ).get();
-
-  const sessionCount = isCountResult(sessionsRaw) ? sessionsRaw.cnt : 0;
-  const chunkCount = isCountResult(chunksRaw) ? chunksRaw.cnt : 0;
-
-  return {
-    session_count: sessionCount,
-    chunk_count: chunkCount,
-  };
-}
+/**
+ * Vector store using SQLite + sqlite-vec for KNN embedding search.
+ *
+ * Single-file DB at ~/.claude/rlm-vectors.db.
+ * Uses bun:sqlite with the sqlite-vec extension for vector similarity search.
+ */
+
+import { Database } from "bun:sqlite";
+import * as sqliteVec from "sqlite-vec";
+import { RLM_VECTOR_DB_PATH, EMBED_DIMENSIONS, type VectorSearchResult } from "./types.js";
+import { logDebug, logInfo } from "./logger.js";
+
+const HOOK_NAME = "rlm_vector";
+
+export interface ChunkRow {
+  session_id: string;
+  project: string;
+  date: string;
+  segment_index: number;
+  line_start: number;
+  line_end: number;
+  topic: string;
+  chunk_text: string;
+  source_path: string;
+  embedding: Float32Array;
+}
+
+export interface VectorStats {
+  session_count: number;
+  chunk_count: number;
+}
+
+/**
+ * Open the vector DB, load sqlite-vec extension, create schema, set WAL mode.
+ */
+export function openVectorDb(path?: string): Database {
+  const dbPath = path ?? RLM_VECTOR_DB_PATH;
+  const db = new Database(dbPath);
+  sqliteVec.load(db);
+  db.run("PRAGMA journal_mode=WAL");
+
+  db.run(`
+    CREATE TABLE IF NOT EXISTS embedded_sessions (
+      session_id TEXT NOT NULL,
+      project TEXT NOT NULL,
+      source_mtime INTEGER NOT NULL,
+      chunk_count INTEGER NOT NULL,
+      embedded_at TEXT NOT NULL,
+      PRIMARY KEY (session_id, project)
+    )
+  `);
+
+  // vec0 virtual table for KNN search
+  db.run(`
+    CREATE VIRTUAL TABLE IF NOT EXISTS chunks USING vec0(
+      embedding float[${EMBED_DIMENSIONS}],
+      project text,
+      date text,
+      +session_id text,
+      +segment_index integer,
+      +line_start integer,
+      +line_end integer,
+      +topic text,
+      +chunk_text text,
+      +source_path text
+    )
+  `);
+
+  logDebug(HOOK_NAME, `Opened vector DB at ${dbPath}`);
+  return db;
+}
+
+/**
+ * Insert chunks in a single transaction.
+ */
+export function insertChunks(db: Database, chunks: ChunkRow[]): void {
+  if (chunks.length === 0) return;
+
+  const stmt = db.prepare(`
+    INSERT INTO chunks (
+      embedding, project, date,
+      session_id, segment_index, line_start, line_end,
+      topic, chunk_text, source_path
+    ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+  `);
+
+  const tx = db.transaction(() => {
+    for (const chunk of chunks) {
+      stmt.run(
+        new Uint8Array(chunk.embedding.buffer),
+        chunk.project,
+        chunk.date,
+        chunk.session_id,
+        chunk.segment_index,
+        chunk.line_start,
+        chunk.line_end,
+        chunk.topic,
+        chunk.chunk_text,
+        chunk.source_path,
+      );
+    }
+  });
+
+  tx();
+  logDebug(HOOK_NAME, `Inserted ${chunks.length} chunks`);
+}
+
+/**
+ * Mark a session as embedded (upsert).
+ */
+export function markSessionEmbedded(
+  db: Database,
+  sessionId: string,
+  project: string,
+  mtime: number,
+  count: number,
+): void {
+  db.run(
+    `INSERT OR REPLACE INTO embedded_sessions (session_id, project, source_mtime, chunk_count, embedded_at)
+     VALUES (?, ?, ?, ?, ?)`,
+    [sessionId, project, mtime, count, new Date().toISOString()],
+  );
+}
+
+/**
+ * Check if a session is already embedded at the given mtime.
+ */
+export function isSessionEmbedded(
+  db: Database,
+  sessionId: string,
+  project: string,
+  mtime: number,
+): boolean {
+  const row = db.query(
+    `SELECT 1 FROM embedded_sessions WHERE session_id = ? AND project = ? AND source_mtime = ?`,
+  ).get(sessionId, project, mtime);
+  return row !== null && row !== undefined;
+}
+
+/**
+ * Delete all chunks for a session (for re-indexing).
+ */
+function isRowidResult(obj: unknown): obj is { rowid: number } {
+  return typeof obj === "object" && obj !== null && "rowid" in obj && typeof (obj as { rowid: unknown }).rowid === "number";
+}
+
+export function deleteSessionChunks(
+  db: Database,
+  sessionId: string,
+  project: string,
+): void {
+  // vec0 tables support DELETE with rowid ranges, but we need to find matching rowids first
+  const rawRows = db.query(
+    `SELECT rowid FROM chunks WHERE session_id = ? AND project = ?`,
+  ).all(sessionId, project);
+
+  const rows = (Array.isArray(rawRows) ? rawRows : []).filter(isRowidResult);
+
+  if (rows.length > 0) {
+    const tx = db.transaction(() => {
+      for (const row of rows) {
+        db.run(`DELETE FROM chunks WHERE rowid = ?`, [row.rowid]);
+      }
+    });
+    tx();
+    logDebug(HOOK_NAME, `Deleted ${rows.length} chunks for ${sessionId}`);
+  }
+
+  db.run(
+    `DELETE FROM embedded_sessions WHERE session_id = ? AND project = ?`,
+    [sessionId, project],
+  );
+}
+
+function isSearchResultRow(obj: unknown): obj is {
+  rowid: number;
+  distance: number;
+  project: string;
+  date: string;
+  session_id: string;
+  segment_index: number;
+  line_start: number;
+  line_end: number;
+  topic: string;
+  source_path: string;
+} {
+  if (typeof obj !== "object" || obj === null) return false;
+  const r = obj as Record<string, unknown>;
+  return (
+    typeof r.rowid === "number" &&
+    typeof r.distance === "number" &&
+    typeof r.project === "string" &&
+    typeof r.date === "string" &&
+    typeof r.session_id === "string" &&
+    typeof r.segment_index === "number" &&
+    typeof r.line_start === "number" &&
+    typeof r.line_end === "number" &&
+    typeof r.topic === "string" &&
+    typeof r.source_path === "string"
+  );
+}
+
+/**
+ * KNN search for the closest chunks to a query embedding.
+ */
+export function searchKnn(
+  db: Database,
+  queryEmbedding: Float32Array,
+  topK: number,
+  projectFilter?: string,
+): VectorSearchResult[] {
+  const queryBytes = new Uint8Array(queryEmbedding.buffer);
+
+  let sql: string;
+  let params: unknown[];
+
+  if (projectFilter) {
+    sql = `
+      SELECT rowid, distance, project, date,
+             session_id, segment_index, line_start, line_end,
+             topic, source_path
+      FROM chunks
+      WHERE embedding MATCH ? AND k = ? AND project = ?
+      ORDER BY distance
+    `;
+    params = [queryBytes, topK, projectFilter];
+  } else {
+    sql = `
+      SELECT rowid, distance, project, date,
+             session_id, segment_index, line_start, line_end,
+             topic, source_path
+      FROM chunks
+      WHERE embedding MATCH ? AND k = ?
+      ORDER BY distance
+    `;
+    params = [queryBytes, topK];
+  }
+
+  const rawRows = db.query(sql).all(...params);
+  const rows = (Array.isArray(rawRows) ? rawRows : []).filter(isSearchResultRow);
+
+  return rows.map((r) => ({
+    chunk_id: r.rowid,
+    session_id: r.session_id,
+    project: r.project,
+    segment_index: r.segment_index,
+    line_start: r.line_start,
+    line_end: r.line_end,
+    topic: r.topic,
+    date: r.date,
+    source_path: r.source_path,
+    distance: r.distance,
+  }));
+}
+
+function isCountResult(obj: unknown): obj is { cnt: number } {
+  return typeof obj === "object" && obj !== null && "cnt" in obj && typeof (obj as { cnt: unknown }).cnt === "number";
+}
+
+/**
+ * Get counts of embedded sessions and chunks.
+ */
+export function getStats(db: Database): VectorStats {
+  const sessionsRaw = db.query(
+    `SELECT COUNT(*) as cnt FROM embedded_sessions`,
+  ).get();
+  const chunksRaw = db.query(
+    `SELECT COUNT(*) as cnt FROM chunks`,
+  ).get();
+
+  const sessionCount = isCountResult(sessionsRaw) ? sessionsRaw.cnt : 0;
+  const chunkCount = isCountResult(chunksRaw) ? chunksRaw.cnt : 0;
+
+  return {
+    session_count: sessionCount,
+    chunk_count: chunkCount,
+  };
+}