codesearch-mcp 0.1.0-alpha → 0.2.0-alpha

package/dist/src/config-store.js

@@ -35,6 +35,7 @@ var __importStar = (this && this.__importStar) || (function () {
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.CONFIG_DEFAULTS = exports.ALL_KEYS = exports.ENV_FILE = exports.SCHEMA_FILE = exports.CONFIG_FILE = exports.CONFIG_DIR = void 0;
 exports.readConfigJson = readConfigJson;
+exports.detectFormerOllamaConfig = detectFormerOllamaConfig;
 exports.migrateRawConfig = migrateRawConfig;
 exports.parseConfig = parseConfig;
 exports.buildJsonSchema = buildJsonSchema;
@@ -48,6 +49,7 @@ const path = __importStar(require("path"));
 const os = __importStar(require("os"));
 const zod_1 = require("zod");
 const constants_1 = require("./constants");
+const config_error_1 = require("./config-error");
 exports.CONFIG_DIR = path.join(os.homedir(), '.config', 'codesearch-mcp');
 exports.CONFIG_FILE = path.join(exports.CONFIG_DIR, 'config.json');
 exports.SCHEMA_FILE = path.join(exports.CONFIG_DIR, 'config.schema.json');
@@ -66,10 +68,10 @@ const jsonSchema = zod_1.z.object({
         name: zod_1.z.string().describe('Database name; defaults to rag_<os-username>, auto-created on first run').optional(),
     }).optional(),
     embedding: zod_1.z.object({
-        provider: zod_1.z.enum(['ollama', 'openai-compatible', 'voyage', 'infinity']).describe('Embedding backend (default ollama)').optional(),
-        baseURL: zod_1.z.string().describe('Backend base URL. ollama: http://localhost:11434 · OpenAI: https://api.openai.com/v1 · Infinity/TEI/vLLM: http://localhost:7997 · Voyage: https://api.voyageai.com/v1').optional(),
+        provider: zod_1.z.enum(['openai-compatible', 'voyage', 'infinity']).describe('Embedding backend (default infinity)').optional(),
+        baseURL: zod_1.z.string().describe('Backend base URL. Infinity/TEI/vLLM/Ollama-via-openai-compatible: http://localhost:7997 (Infinity default) · OpenAI: https://api.openai.com/v1 · Voyage: https://api.voyageai.com/v1').optional(),
         apiKey: zod_1.z.string().describe('API key. Use env:VAR_NAME to read from ~/.config/codesearch-mcp/.env or the environment (recommended). Empty for local servers.').optional(),
-        model: zod_1.z.string().describe('Embedding model (default bge-m3)').optional(),
+        model: zod_1.z.string().describe('Embedding model (default jinaai/jina-embeddings-v2-base-code)').optional(),
         dimensions: zod_1.z.number().int().describe('Vector width override; required for models whose native width exceeds 2000 (Matryoshka reduction). Omit to derive from the model/registry/probe.').optional(),
         batchSize: zod_1.z.number().int().describe('Texts per batch embedding request (default 32)').optional(),
     }).optional(),
@@ -87,16 +89,18 @@ const jsonSchema = zod_1.z.object({
     }).optional(),
     reranker: zod_1.z.object({
         enabled: zod_1.z.boolean().describe('Enable cross-encoder reranking; requires reranker.model').optional(),
-        provider: zod_1.z.enum(['openai-compatible']).describe('Rerank backend. Only openai-compatible /rerank servers (Infinity, TEI, vLLM) — Ollama has no rerank API.').optional(),
+        provider: zod_1.z.enum(['openai-compatible']).describe('Rerank backend. Only openai-compatible /rerank servers (Infinity, TEI, vLLM).').optional(),
         baseURL: zod_1.z.string().describe('Rerank server base URL (defaults to embedding.baseURL — Infinity serves both).').optional(),
         apiKey: zod_1.z.string().describe('API key, or env:VAR_NAME. Empty for local servers.').optional(),
         model: zod_1.z.string().describe('Reranker model name (e.g. qllama/bge-reranker-v2-m3)').optional(),
         url: zod_1.z.string().describe('Full URL of the rerank endpoint (default derived from provider + baseURL)').optional(),
-        overfetch: zod_1.z.number().int().min(1).max(20).describe('Candidate multiplier before reranking (limit × N, default 3)').optional(),
+        overfetch: zod_1.z.number().int().min(1).max(20).describe('Candidate multiplier before reranking (limit × N, default 5)').optional(),
+        candidateFloor: zod_1.z.number().int().min(1).describe('Minimum candidate pool fed to the reranker regardless of limit; fetchLimit = min(max(limit × overfetch, candidateFloor), 200) (default 80)').optional(),
     }).optional(),
     logging: zod_1.z.object({
         level: zod_1.z.enum(['debug', 'info', 'warn', 'error']).describe('File log level (default error)').optional(),
         dir: zod_1.z.string().describe('Base log directory; "false" to disable file logging').optional(),
+        searchMetrics: zod_1.z.boolean().describe('Emit one structured search-tuning line (info level) per search, independent of level (default false)').optional(),
     }).optional(),
     workspaceRoot: zod_1.z.string().describe('Full path to the directory containing all your projects').optional(),
 });
@@ -127,8 +131,10 @@ const FIELD_TYPES = {
     'reranker.model': 'string',
     'reranker.url': 'string',
     'reranker.overfetch': 'number',
+    'reranker.candidateFloor': 'number',
     'logging.level': 'string',
     'logging.dir': 'string',
+    'logging.searchMetrics': 'boolean',
     'workspaceRoot': 'string',
 };
 exports.ALL_KEYS = Object.keys(FIELD_TYPES);
@@ -146,7 +152,7 @@ exports.ALL_KEYS = Object.keys(FIELD_TYPES);
  */
 exports.CONFIG_DEFAULTS = {
     'db.port': 5433,
-    'embedding.provider': 'ollama',
+    'embedding.provider': 'infinity',
     'embedding.apiKey': '',
     'embedding.model': constants_1.DEFAULT_EMBED_MODEL,
     'embedding.batchSize': 32,
@@ -158,12 +164,14 @@ exports.CONFIG_DEFAULTS = {
     'search.similarityThreshold': 0.6,
     'search.defaultLimit': 5,
     'search.maxLimit': 20,
-    'reranker.enabled': false,
+    'reranker.enabled': true,
     'reranker.provider': 'openai-compatible',
     'reranker.apiKey': '',
-    'reranker.model': '',
-    'reranker.overfetch': 3,
+    'reranker.model': 'mixedbread-ai/mxbai-rerank-xsmall-v1',
+    'reranker.overfetch': 5,
+    'reranker.candidateFloor': 80,
     'logging.level': 'error',
+    'logging.searchMetrics': false,
 };
 function readConfigJson() {
     if (!fs.existsSync(exports.CONFIG_FILE))
@@ -183,14 +191,61 @@ function readConfigJson() {
         throw new Error(`Invalid JSON in ${exports.CONFIG_FILE}: ${e.message}\n` +
             `Fix the file manually or delete it and run: rag-setup`);
     }
-    return parseConfig(migrateRawConfig(parsed), exports.CONFIG_FILE);
+    return parseConfig(parsed, exports.CONFIG_FILE);
+}
+const OLLAMA_MIGRATION_REMEDY = 'The "ollama" embedding provider has been removed. Point Ollama through the ' +
+    '"openai-compatible" provider instead — set embedding.provider to "openai-compatible" ' +
+    'and embedding.baseURL to "http://localhost:11434/v1" — via `rag-config edit` (or by ' +
+    'editing the config file directly) or by re-running `rag-setup`. After switching, rebuild ' +
+    'the index with `rag-index --reindex`.';
+/** Strip a trailing `:tag` (Docker-style) from a model name, if present. */
+function stripModelTag(model) {
+    return model.replace(/:[^:]*$/, '');
+}
+/**
+ * Detect a raw (pre-migration) config object that still selects the removed
+ * `ollama` embedding provider — explicitly, via a legacy top-level `ollama`
+ * block, or implicitly (no `provider` set, but `baseURL`/`model` point at
+ * Ollama's defaults) — and throw a `ConfigError` with migration guidance
+ * instead of letting it validate under a silently reinterpreted provider.
+ * Runs before `migrateRawConfig` so a legacy `ollama.*` block is never
+ * converted into an `embedding` block first. Pure — never mutates `raw`.
+ */
+function detectFormerOllamaConfig(raw) {
+    if (raw == null || typeof raw !== 'object')
+        return;
+    const obj = raw;
+    const embedding = obj.embedding;
+    if (embedding != null && typeof embedding === 'object') {
+        const e = embedding;
+        if (e.provider === 'ollama') {
+            throw new config_error_1.ConfigError(`Config selects the "ollama" embedding provider, which is no longer supported. ${OLLAMA_MIGRATION_REMEDY}`);
+        }
+        if (e.provider === undefined) {
+            if (typeof e.baseURL === 'string') {
+                let port;
+                try {
+                    port = new URL(e.baseURL).port;
+                }
+                catch { /* unparsable — not a signal */ }
+                if (port === '11434') {
+                    throw new config_error_1.ConfigError(`Config's embedding.baseURL targets Ollama's default port (11434) with no embedding.provider set. ${OLLAMA_MIGRATION_REMEDY}`);
+                }
+            }
+            if (typeof e.model === 'string' && constants_1.FORMER_OLLAMA_MODELS.includes(stripModelTag(e.model))) {
+                throw new config_error_1.ConfigError(`Config's embedding.model ("${e.model}") was formerly registered for Ollama, with no embedding.provider set. ${OLLAMA_MIGRATION_REMEDY}`);
+            }
+        }
+        return;
+    }
+    if (obj.ollama != null && typeof obj.ollama === 'object') {
+        throw new config_error_1.ConfigError(`Config has a legacy "ollama" configuration block. ${OLLAMA_MIGRATION_REMEDY}`);
+    }
 }
 /**
  * Migrate a legacy raw config object to the current shape, in place of a
  * versioned migration system. Pure — returns a new object, never mutates input.
  *
- * - Legacy `ollama.host`/`ollama.model` → `embedding.{baseURL,model}` with
- *   `provider: 'ollama'` (only when `embedding` is not already present).
  * - Removed translation keys (`search.autoTranslate`, `search.translateModel`)
  *   are dropped so they neither validate-fail nor take effect.
  * - Legacy `reranker.provider: 'ollama'` (a dead value — Ollama has no rerank
@@ -198,21 +253,14 @@ function readConfigJson() {
  *   dropped so it re-derives to `/rerank`. Otherwise the tightened enum would
  *   fail validation on existing configs and lock the user out.
  *
- * `ollama` itself is left for the schema to strip. Exported for unit testing.
+ * A legacy top-level `ollama` block reaches here only when `detectFormerOllamaConfig`
+ * did not already reject it (i.e. an `embedding` block is also present); it is
+ * dropped rather than merged. Exported for unit testing.
  */
 function migrateRawConfig(raw) {
     if (raw == null || typeof raw !== 'object')
         return raw;
     const obj = JSON.parse(JSON.stringify(raw));
-    const legacyOllama = obj.ollama;
-    if (!obj.embedding && legacyOllama && typeof legacyOllama === 'object') {
-        const embedding = { provider: 'ollama' };
-        if (typeof legacyOllama.host === 'string')
-            embedding.baseURL = legacyOllama.host;
-        if (typeof legacyOllama.model === 'string')
-            embedding.model = legacyOllama.model;
-        obj.embedding = embedding;
-    }
     delete obj.ollama;
     if (obj.search && typeof obj.search === 'object') {
         delete obj.search.autoTranslate;
@@ -227,12 +275,19 @@ function migrateRawConfig(raw) {
     return obj;
 }
 /**
- * Validate a parsed config object against the Zod schema. Unknown keys (e.g.
- * the "$schema" editor reference) are stripped. Throws with a caller-friendly
- * message on validation failure. Pure — no file I/O.
+ * Validate a raw (pre-migration) config object against the Zod schema: first
+ * `detectFormerOllamaConfig` (a `ConfigError` on a former-Ollama config takes
+ * priority over both migration and the regular Zod enum rejection), then
+ * `migrateRawConfig`, then the schema itself. Unknown keys (e.g. the "$schema"
+ * editor reference) are stripped. Throws with a caller-friendly message on
+ * validation failure. The single entry point for both `readConfigJson` and
+ * `PUT /api/config`, so both apply identical rules with an identical message.
+ * Pure — no file I/O.
  */
 function parseConfig(raw, source = exports.CONFIG_FILE) {
-    const result = jsonSchema.safeParse(raw);
+    detectFormerOllamaConfig(raw);
+    const migrated = migrateRawConfig(raw);
+    const result = jsonSchema.safeParse(migrated);
     if (!result.success) {
         const issues = result.error.issues.map(i => `  ${i.path.join('.')}: ${i.message}`).join('\n');
         throw new Error(`Invalid config in ${source}:\n${issues}\n` +

package/dist/src/config.d.ts

@@ -1,6 +1,5 @@
-export declare class ConfigError extends Error {
-    constructor(message: string);
-}
+import { ConfigError } from './config-error';
+export { ConfigError };
 /**
  * Resolve a config value that may reference an environment variable. A value of
  * the form `env:VAR_NAME` resolves to `process.env.VAR_NAME`; any other string
@@ -26,7 +25,7 @@ export declare class Config {
      * locking the user out of `rag-web`/`rag-mcp`.
      */
     get embedding(): {
-        provider: "ollama" | "openai-compatible" | "voyage" | "infinity";
+        provider: "openai-compatible" | "voyage" | "infinity";
         baseURL: string;
         apiKey: string;
         model: string;
@@ -42,6 +41,7 @@ export declare class Config {
         model: string;
         url: string;
         overfetch: number;
+        candidateFloor: number;
     };
     get indexer(): {
         chunkSize: number;
@@ -57,6 +57,7 @@ export declare class Config {
     };
     get logging(): {
         level: "error" | "debug" | "info" | "warn";
+        searchMetrics: boolean;
         dir?: string | undefined;
     };
     get workspaceRoot(): string;

package/dist/src/config.js

@@ -38,13 +38,8 @@ exports.resolveEnvRef = resolveEnvRef;
 const os = __importStar(require("os"));
 const zod_1 = require("zod");
 const config_store_1 = require("./config-store");
-class ConfigError extends Error {
-    constructor(message) {
-        super(message);
-        this.name = 'ConfigError';
-    }
-}
-exports.ConfigError = ConfigError;
+const config_error_1 = require("./config-error");
+Object.defineProperty(exports, "ConfigError", { enumerable: true, get: function () { return config_error_1.ConfigError; } });
 /**
  * Resolve a config value that may reference an environment variable. A value of
  * the form `env:VAR_NAME` resolves to `process.env.VAR_NAME`; any other string
@@ -58,24 +53,23 @@ function resolveEnvRef(value, env = process.env) {
     const name = value.slice(4).trim();
     const resolved = name ? env[name] : undefined;
     if (resolved == null || resolved === '') {
-        throw new ConfigError(`Config references environment variable "${name}" (as "env:${name}") but it is not set. ` +
+        throw new config_error_1.ConfigError(`Config references environment variable "${name}" (as "env:${name}") but it is not set. ` +
             `Define ${name} in ${config_store_1.ENV_FILE} or export it in the environment.`);
     }
     return resolved;
 }
 const DEFAULT_BASE_URLS = {
-    'ollama': 'http://localhost:11434',
     'openai-compatible': 'https://api.openai.com/v1',
     'voyage': 'https://api.voyageai.com/v1',
     'infinity': 'http://localhost:7997',
 };
 function defaultBaseURL(provider) {
-    return DEFAULT_BASE_URLS[provider] ?? DEFAULT_BASE_URLS.ollama;
+    return DEFAULT_BASE_URLS[provider] ?? DEFAULT_BASE_URLS.infinity;
 }
 // Reranking is served only by OpenAI-compatible `/rerank` endpoints (Infinity,
-// TEI, vLLM, Jina/Cohere-shape). Ollama has no rerank API, so there is a single
-// endpoint shape. Normalize a trailing `/embeddings` or `/v1` to the server root
-// so a baseURL written with an OpenAI-style suffix still yields `<root>/rerank`.
+// TEI, vLLM, Jina/Cohere-shape) — a single endpoint shape. Normalize a trailing
+// `/embeddings` or `/v1` to the server root so a baseURL written with an
+// OpenAI-style suffix still yields `<root>/rerank`.
 function deriveRerankUrl(baseURL) {
     const root = baseURL.replace(/\/+$/, '').replace(/\/(embeddings|v1)$/, '');
     return `${root}/rerank`;
@@ -89,7 +83,7 @@ const resolvedSchema = zod_1.z.object({
         database: zod_1.z.string().min(1),
     }),
     embedding: zod_1.z.object({
-        provider: zod_1.z.enum(['ollama', 'openai-compatible', 'voyage', 'infinity']).default(config_store_1.CONFIG_DEFAULTS['embedding.provider']),
+        provider: zod_1.z.enum(['openai-compatible', 'voyage', 'infinity']).default(config_store_1.CONFIG_DEFAULTS['embedding.provider']),
         baseURL: zod_1.z.string().min(1),
         apiKey: zod_1.z.string().default(config_store_1.CONFIG_DEFAULTS['embedding.apiKey']),
         model: zod_1.z.string().min(1).default(config_store_1.CONFIG_DEFAULTS['embedding.model']),
@@ -104,6 +98,7 @@ const resolvedSchema = zod_1.z.object({
         model: zod_1.z.string().default(config_store_1.CONFIG_DEFAULTS['reranker.model']),
         url: zod_1.z.string().min(1),
         overfetch: zod_1.z.number().int().min(1).max(20).default(config_store_1.CONFIG_DEFAULTS['reranker.overfetch']),
+        candidateFloor: zod_1.z.number().int().min(1).default(config_store_1.CONFIG_DEFAULTS['reranker.candidateFloor']),
     }),
     indexer: zod_1.z.object({
         chunkSize: zod_1.z.number().int().positive().default(config_store_1.CONFIG_DEFAULTS['indexing.chunkSize']),
@@ -120,6 +115,7 @@ const resolvedSchema = zod_1.z.object({
     logging: zod_1.z.object({
         level: zod_1.z.enum(['debug', 'info', 'warn', 'error']).default(config_store_1.CONFIG_DEFAULTS['logging.level']),
         dir: zod_1.z.string().optional(),
+        searchMetrics: zod_1.z.boolean().default(config_store_1.CONFIG_DEFAULTS['logging.searchMetrics']),
     }),
     workspaceRoot: zod_1.z.string().default(''),
 });
@@ -143,7 +139,7 @@ class Config {
     constructor() {
         const raw = (0, config_store_1.readConfigJson)();
         if (!raw)
-            throw new ConfigError('Config not found. Run rag-setup.');
+            throw new config_error_1.ConfigError('Config not found. Run rag-setup.');
         const embProvider = (raw.embedding?.provider ?? config_store_1.CONFIG_DEFAULTS['embedding.provider']);
         const embBaseURL = raw.embedding?.baseURL ?? defaultBaseURL(embProvider);
         const rerankBaseURL = raw.reranker?.baseURL ?? embBaseURL;
@@ -172,6 +168,7 @@ class Config {
                 model: raw.reranker?.model,
                 url: rerankUrl,
                 overfetch: raw.reranker?.overfetch,
+                candidateFloor: raw.reranker?.candidateFloor,
             },
             indexer: {
                 chunkSize: raw.indexing?.chunkSize,
@@ -188,6 +185,7 @@ class Config {
             logging: {
                 level: raw.logging?.level,
                 dir: raw.logging?.dir,
+                searchMetrics: raw.logging?.searchMetrics,
             },
             workspaceRoot: raw.workspaceRoot,
         });
@@ -195,7 +193,7 @@ class Config {
             const issues = result.error.issues
                 .map(i => `  ${i.path.join('.')}: ${i.message}`)
                 .join('\n');
-            throw new ConfigError(`Invalid config:\n${issues}\nRun rag-setup or: rag-config set <key> <value>`);
+            throw new config_error_1.ConfigError(`Invalid config:\n${issues}\nRun rag-setup or: rag-config set <key> <value>`);
         }
         this.data = result.data;
     }

package/dist/src/constants.d.ts

@@ -1,4 +1,4 @@
-export type EmbeddingProvider = 'ollama' | 'openai-compatible' | 'voyage' | 'infinity';
+export type EmbeddingProvider = 'openai-compatible' | 'voyage' | 'infinity';
 /**
  * pgvector's HNSW / IVFFlat index hard limit. The effective embedding dimension
  * must not exceed this; larger cloud models must reduce via the Matryoshka
@@ -23,7 +23,13 @@ export interface ModelInfo {
  * time (see `resolveDimension`).
  */
 export declare const MODEL_REGISTRY: Record<string, ModelInfo>;
-export declare const DEFAULT_EMBED_MODEL = "bge-m3";
+export declare const DEFAULT_EMBED_MODEL = "jinaai/jina-embeddings-v2-base-code";
+/**
+ * Former Ollama registry model names, kept only so config validation can
+ * recognise an implicit (provider-less) config that still targets Ollama —
+ * see `detectFormerOllamaConfig` in `config-store.ts`.
+ */
+export declare const FORMER_OLLAMA_MODELS: readonly ["bge-m3", "nomic-embed-text", "mxbai-embed-large"];
 /** Advisory dimension for a model, or `undefined` when the model is unknown. */
 export declare function registryDimension(model: string): number | undefined;
 /**

package/dist/src/constants.js

@@ -1,6 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.DEFAULT_EMBED_MODEL = exports.MODEL_REGISTRY = exports.MAX_EMBED_DIM = void 0;
+exports.FORMER_OLLAMA_MODELS = exports.DEFAULT_EMBED_MODEL = exports.MODEL_REGISTRY = exports.MAX_EMBED_DIM = void 0;
 exports.registryDimension = registryDimension;
 exports.resolveDimension = resolveDimension;
 /**
@@ -16,9 +16,6 @@ exports.MAX_EMBED_DIM = 2000;
  * time (see `resolveDimension`).
  */
 exports.MODEL_REGISTRY = {
-    'bge-m3': { provider: 'ollama', dimensions: 1024 },
-    'nomic-embed-text': { provider: 'ollama', dimensions: 768 },
-    'mxbai-embed-large': { provider: 'ollama', dimensions: 1024 },
     'text-embedding-3-small': { provider: 'openai-compatible', dimensions: 1536, dimConfigurable: true },
     'text-embedding-3-large': { provider: 'openai-compatible', dimensions: 3072, dimConfigurable: true, maxIndexable: 2000 },
     'text-embedding-ada-002': { provider: 'openai-compatible', dimensions: 1536 },
@@ -28,7 +25,13 @@ exports.MODEL_REGISTRY = {
     'BAAI/bge-m3': { provider: 'infinity', dimensions: 1024 },
     'jinaai/jina-embeddings-v2-base-code': { provider: 'infinity', dimensions: 768 },
 };
-exports.DEFAULT_EMBED_MODEL = 'bge-m3';
+exports.DEFAULT_EMBED_MODEL = 'jinaai/jina-embeddings-v2-base-code';
+/**
+ * Former Ollama registry model names, kept only so config validation can
+ * recognise an implicit (provider-less) config that still targets Ollama —
+ * see `detectFormerOllamaConfig` in `config-store.ts`.
+ */
+exports.FORMER_OLLAMA_MODELS = ['bge-m3', 'nomic-embed-text', 'mxbai-embed-large'];
 /** Advisory dimension for a model, or `undefined` when the model is unknown. */
 function registryDimension(model) {
     return exports.MODEL_REGISTRY[model]?.dimensions;

package/dist/src/help.js

@@ -29,7 +29,7 @@ function printHelp(command) {
         case 'setup':
             console.log('Usage: rag-setup');
             console.log('');
-            console.log('Interactive setup: configure DB, Ollama, and workspaceRoot.');
+            console.log('Interactive setup: configure DB, embedding backend, and workspaceRoot.');
             console.log('Run once before using rag-index or rag-mcp.');
             console.log('');
             console.log('Options:');

package/dist/src/interfaces/IDatabase.d.ts

@@ -24,4 +24,21 @@ export interface IDatabase {
     ensureChunkTable(dimensions: number): Promise<void>;
     /** Drop and recreate `code_chunk` at `dimensions` (dimension change). */
     recreateChunkTable(dimensions: number): Promise<void>;
+    /**
+     * Return all file paths referenced in any of the four index tables for `project`.
+     * Union of code_chunk.filePath, code_symbol.filePath, code_call.fromFile,
+     * code_dependency.fromFile. calleeFile and toFile are excluded (may belong to other projects).
+     */
+    listKnownFiles(project: string): Promise<string[]>;
+    /**
+     * Delete rows for `filePaths` from all four index tables in one transaction.
+     * code_call rows where the file is either fromFile OR calleeFile are removed.
+     * code_dependency rows where the file is either fromFile OR toFile are removed.
+     * No-op when `filePaths` is empty.
+     */
+    deleteFiles(project: string, filePaths: string[]): Promise<void>;
+    /** Return all project names referenced in any of the four index tables. */
+    listIndexedProjects(): Promise<string[]>;
+    /** Delete all rows for `name` from the four index tables in one transaction. */
+    deleteProject(name: string): Promise<void>;
 }

package/dist/src/interfaces/ILogger.d.ts

@@ -7,4 +7,11 @@ export interface ILogger {
     info(msg: string, data?: Record<string, unknown>): void;
     warn(msg: string, data?: Record<string, unknown>): void;
     error(msg: string, data?: Record<string, unknown>): void;
+    /**
+     * Structured telemetry emitted at `info` **independent of the configured log
+     * level** — so an operator can collect it in production without the full
+     * `debug` firehose. No-op when file logging is disabled (never writes to
+     * stdout, which the MCP transport owns).
+     */
+    metric(msg: string, data?: Record<string, unknown>): void;
 }

package/dist/src/services/AstChunker.d.ts

@@ -13,6 +13,17 @@ export declare class AstChunker implements IChunker {
     private readonly chunkSize;
     constructor(config: Config);
     chunk(code: string, opts: ChunkOpts): Promise<ChunkPiece[]>;
+    private produce;
+    /**
+     * True iff every significant line (blank and comment lines stripped) belongs
+     * to an `import` / re-export statement and at least one is an import or
+     * re-export keyword — so the chunk carries no value/type declaration and is
+     * never a search answer. Any `export const|let|var|function|class|interface|
+     * enum|namespace|default|declare|async function|type X = …` disqualifies
+     * exclusion. Pure. Distinguishes `export { … } from` / `export type { … }`
+     * (re-export → drop) from `export type X = …` (declaration → keep).
+     */
+    static isImportOrReexportOnly(content: string): boolean;
     /** Character-splitter path for a whole file; line numbers from splitter metadata. */
     private splitWholeFile;
     private chunkByAst;

package/dist/src/services/AstChunker.js

@@ -27,6 +27,15 @@ class AstChunker {
         });
     }
     async chunk(code, opts) {
+        const pieces = await this.produce(code, opts);
+        // Single post-chunk pass: drop pieces whose only significant content is
+        // import / re-export statements. Both chunk sources — AST "gap" regions and
+        // the whole-file character path for barrel files — can emit these, so we
+        // filter once over the final pieces. Dependency edges are untouched (they
+        // come from analyzeProject → code_dependency, not from chunk text).
+        return pieces.filter(p => !AstChunker.isImportOrReexportOnly(p.content));
+    }
+    async produce(code, opts) {
         if (opts.isSchemaFile) {
             return this.splitWholeFile(code, this.schemaSplitter);
         }
@@ -35,6 +44,41 @@ class AstChunker {
         }
         return this.chunkByAst(code, opts.symbols);
     }
+    /**
+     * True iff every significant line (blank and comment lines stripped) belongs
+     * to an `import` / re-export statement and at least one is an import or
+     * re-export keyword — so the chunk carries no value/type declaration and is
+     * never a search answer. Any `export const|let|var|function|class|interface|
+     * enum|namespace|default|declare|async function|type X = …` disqualifies
+     * exclusion. Pure. Distinguishes `export { … } from` / `export type { … }`
+     * (re-export → drop) from `export type X = …` (declaration → keep).
+     */
+    static isImportOrReexportOnly(content) {
+        const lines = content
+            .replace(/\/\*[\s\S]*?\*\//g, '') // strip block comments
+            .split('\n')
+            .map(l => l.trim())
+            .filter(l => l !== '' && !l.startsWith('//') && !l.startsWith('*'));
+        if (lines.length === 0)
+            return false;
+        const IMPORT = /^import\b/;
+        const REEXPORT = /^export\s+(\*|type\s+\{|\{)/; // export * | export type { | export {
+        const DISQUALIFY = /^export\s+(default|async\b|function\b|const\b|let\b|var\b|class\b|abstract\b|interface\b|enum\b|namespace\b|declare\b|type\s+[A-Za-z_$])/;
+        const CONTINUATION = /^(\}?\s*from\s+['"][^'"]+['"];?|[{}]|(type\s+)?[A-Za-z_$][\w$]*(\s+as\s+[A-Za-z_$][\w$]*)?,?|['"][^'"]*['"];?)$/;
+        let keywords = 0;
+        for (const line of lines) {
+            if (DISQUALIFY.test(line))
+                return false;
+            if (IMPORT.test(line) || REEXPORT.test(line)) {
+                keywords++;
+                continue;
+            }
+            if (CONTINUATION.test(line))
+                continue;
+            return false; // unrecognized line → real code, keep the chunk
+        }
+        return keywords >= 1;
+    }
     /** Character-splitter path for a whole file; line numbers from splitter metadata. */
     async splitWholeFile(code, splitter) {
         const docs = await splitter.createDocuments([code]);

package/dist/src/services/DatabaseService.d.ts

@@ -27,6 +27,10 @@ export declare class DatabaseService implements IDatabase {
     recreateChunkTable(dimensions: number): Promise<void>;
     readIndexMeta(): Promise<IndexMeta | null>;
     writeIndexMeta(meta: IndexMeta): Promise<void>;
+    listKnownFiles(project: string): Promise<string[]>;
+    deleteFiles(project: string, filePaths: string[]): Promise<void>;
+    listIndexedProjects(): Promise<string[]>;
+    deleteProject(name: string): Promise<void>;
     /** Guard against SQL injection — the dimension is interpolated into DDL. */
     private assertDim;
     private ensureDatabaseExists;

package/dist/src/services/DatabaseService.js

@@ -36,7 +36,9 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.DatabaseService = void 0;
 const node_postgres_1 = require("drizzle-orm/node-postgres");
 const pg_1 = require("pg");
+const drizzle_orm_1 = require("drizzle-orm");
 const schema = __importStar(require("../db/schema"));
+const db_1 = require("../db");
 class DatabaseService {
     config;
     pool;
@@ -198,6 +200,50 @@ class DatabaseService {
              dimensions = EXCLUDED.dimensions,
              "indexedAt" = now()`, [meta.provider, meta.model, meta.dimensions]);
     }
+    async listKnownFiles(project) {
+        const rows = await this.query(`SELECT "filePath" AS fp FROM code_chunk      WHERE "serviceName" = $1
+       UNION
+       SELECT "filePath" AS fp FROM code_symbol     WHERE project = $1
+       UNION
+       SELECT "fromFile" AS fp FROM code_call       WHERE project = $1
+       UNION
+       SELECT "fromFile" AS fp FROM code_dependency WHERE project = $1`, [project]);
+        return rows.map(r => r.fp);
+    }
+    async deleteFiles(project, filePaths) {
+        if (filePaths.length === 0)
+            return;
+        // PostgreSQL limits ~32 767 bind params per statement; OR queries bind the
+        // array twice (fromFile + calleeFile), so cap each batch at 10 000 entries.
+        const BATCH = 10_000;
+        await this.transaction(async (tx) => {
+            for (let i = 0; i < filePaths.length; i += BATCH) {
+                const batch = filePaths.slice(i, i + BATCH);
+                await tx.delete(db_1.codeChunk).where((0, drizzle_orm_1.and)((0, drizzle_orm_1.eq)(db_1.codeChunk.serviceName, project), (0, drizzle_orm_1.inArray)(db_1.codeChunk.filePath, batch)));
+                await tx.delete(db_1.codeSymbol).where((0, drizzle_orm_1.and)((0, drizzle_orm_1.eq)(db_1.codeSymbol.project, project), (0, drizzle_orm_1.inArray)(db_1.codeSymbol.filePath, batch)));
+                await tx.delete(db_1.codeCall).where((0, drizzle_orm_1.and)((0, drizzle_orm_1.eq)(db_1.codeCall.project, project), (0, drizzle_orm_1.or)((0, drizzle_orm_1.inArray)(db_1.codeCall.fromFile, batch), (0, drizzle_orm_1.inArray)(db_1.codeCall.calleeFile, batch))));
+                await tx.delete(db_1.codeDependency).where((0, drizzle_orm_1.and)((0, drizzle_orm_1.eq)(db_1.codeDependency.project, project), (0, drizzle_orm_1.or)((0, drizzle_orm_1.inArray)(db_1.codeDependency.fromFile, batch), (0, drizzle_orm_1.inArray)(db_1.codeDependency.toFile, batch))));
+            }
+        });
+    }
+    async listIndexedProjects() {
+        const rows = await this.query(`SELECT "serviceName" AS name FROM code_chunk
+       UNION
+       SELECT project AS name FROM code_symbol
+       UNION
+       SELECT project AS name FROM code_call
+       UNION
+       SELECT project AS name FROM code_dependency`);
+        return rows.map(r => r.name);
+    }
+    async deleteProject(name) {
+        await this.transaction(async (tx) => {
+            await tx.delete(db_1.codeChunk).where((0, drizzle_orm_1.eq)(db_1.codeChunk.serviceName, name));
+            await tx.delete(db_1.codeSymbol).where((0, drizzle_orm_1.eq)(db_1.codeSymbol.project, name));
+            await tx.delete(db_1.codeCall).where((0, drizzle_orm_1.eq)(db_1.codeCall.project, name));
+            await tx.delete(db_1.codeDependency).where((0, drizzle_orm_1.eq)(db_1.codeDependency.project, name));
+        });
+    }
     /** Guard against SQL injection — the dimension is interpolated into DDL. */
     assertDim(dimensions) {
         if (!Number.isInteger(dimensions) || dimensions <= 0 || dimensions > 2000) {

package/dist/src/services/EmbedderFactory.d.ts

@@ -4,6 +4,6 @@ import type { ILogger } from '../interfaces/ILogger';
 import { type EmbedderDeps } from './OpenAICompatibleEmbedder';
 /**
  * Construct the `IEmbedder` implementation selected by `config.embedding.provider`.
- * The default (`ollama`) keeps existing installs working unchanged.
+ * The default (`infinity`) is the recommended local backend.
  */
 export declare function createEmbedder(config: Config, logger?: ILogger, deps?: EmbedderDeps): IEmbedder;

package/dist/src/services/EmbedderFactory.js

@@ -2,13 +2,12 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.createEmbedder = createEmbedder;
 const LoggerService_1 = require("./LoggerService");
-const OllamaEmbedder_1 = require("./OllamaEmbedder");
 const OpenAICompatibleEmbedder_1 = require("./OpenAICompatibleEmbedder");
 const InfinityEmbedder_1 = require("./InfinityEmbedder");
 const VoyageEmbedder_1 = require("./VoyageEmbedder");
 /**
  * Construct the `IEmbedder` implementation selected by `config.embedding.provider`.
- * The default (`ollama`) keeps existing installs working unchanged.
+ * The default (`infinity`) is the recommended local backend.
  */
 function createEmbedder(config, logger = LoggerService_1.LoggerService.silent(), deps = {}) {
     switch (config.embedding.provider) {
@@ -18,12 +17,10 @@ function createEmbedder(config, logger = LoggerService_1.LoggerService.silent(),
             return new InfinityEmbedder_1.InfinityEmbedder(config, logger, deps);
         case 'voyage':
             return new VoyageEmbedder_1.VoyageEmbedder(config, logger, deps);
-        case 'ollama':
-            return new OllamaEmbedder_1.OllamaEmbedder(config, logger, deps);
         default: {
             // The Zod enum should already reject this; guard anyway for a clear message.
             const p = config.embedding.provider;
-            throw new Error(`Unknown embedding provider: "${p}". Expected one of: ollama, openai-compatible, voyage, infinity.`);
+            throw new Error(`Unknown embedding provider: "${p}". Expected one of: openai-compatible, voyage, infinity.`);
         }
     }
 }

package/dist/src/services/IndexerService.d.ts

@@ -51,6 +51,10 @@ export declare class IndexerService {
     scanAndIndex(dir: string, serviceName: string, force?: boolean, signal?: AbortSignal, preCollectedFiles?: string[], opts?: {
         rootTsConfigPath?: string;
     }): Promise<IndexStats>;
+    pruneProjects(keep: string[]): Promise<{
+        removed: string[];
+        failed: string[];
+    }>;
     private static inferRole;
     private getEnrichedPrompt;
     private static hashContent;