@cerebriline/core 4.100.243 → 4.100.245

package/dist/services/retrieval/code-index.d.ts

@@ -0,0 +1,115 @@
+/**
+ * Search a workspace's code by meaning.
+ *
+ * `search_codebase` is a regex: it finds a name the model already knows.
+ * `ask_lsp` resolves a symbol the model already has. Neither answers "where
+ * does this app check a user's password" when no word of the question is in
+ * the code. An embedding model does, and the Library already has everything
+ * that needs: chunking, a keyword index, vectors, the merge of the two and a
+ * reranker.
+ *
+ * So the code index is a second Library, kept apart from the user's books
+ * under `<data dir>/code-index`, with one collection per workspace and one
+ * document per source file. Nothing here knows about embeddings beyond what
+ * it hands on.
+ *
+ * It is opt-in per workspace: indexing sends every source file to the
+ * embedder, which on a large repository is real work for a GPU.
+ */
+import { type LibrarySettings } from "@cerebriline/shared";
+import type { RetrievalEndpoint } from "./embedding-client";
+import { type EmbedPendingResult, Library } from "./library";
+import type { RetrieveResult } from "./retrieve";
+export declare function resolveCodeIndexDirectory(): string;
+/** A source file larger than this is a data file or a bundle, not code to search. */
+export declare const CODE_INDEX_MAX_FILE_BYTES: number;
+/** Files indexed per workspace, at most. Past it the rest is left out and said so. */
+export declare const CODE_INDEX_MAX_FILES = 20000;
+export declare function isIndexableCodePath(relativePath: string): boolean;
+export interface CodeIndexOptions {
+    /** @default <data dir>/code-index */
+    directory?: string;
+    /** Where LanceDB is installed. @default the Library's */
+    runtimeDirectory?: string;
+    fetch?: typeof fetch;
+}
+export interface CodeSyncOptions {
+    settings?: LibrarySettings;
+    /** The workspace's files, relative with forward slashes. @default ripgrep's list, which honours .gitignore */
+    listFiles?: (root: string) => Promise<Iterable<string>>;
+    onProgress?: (done: number, total: number) => void;
+    signal?: AbortSignal;
+}
+export interface CodeSyncResult {
+    /** Files in the index after the run. */
+    files: number;
+    added: number;
+    updated: number;
+    removed: number;
+    /** Files left out: too large, binary, generated, or unreadable. */
+    skipped: number;
+    /** Set when the workspace has more files than are indexed. */
+    truncated?: number;
+}
+export interface CodeEmbedOptions {
+    embedding: RetrievalEndpoint;
+    settings?: LibrarySettings;
+    /** Download LanceDB when it is not installed. */
+    install?: boolean;
+    onProgress?: (documentsDone: number, documents: number) => void;
+    signal?: AbortSignal;
+}
+export interface CodeSearchOptions {
+    settings?: LibrarySettings;
+    embedding?: RetrievalEndpoint;
+    reranker?: RetrievalEndpoint;
+    /** Passages returned. @default 8 */
+    limit?: number;
+    signal?: AbortSignal;
+}
+export interface CodeHit {
+    /** Relative to the workspace, forward slashes. */
+    path: string;
+    /** 1-based, inclusive. Absent when the file has changed since it was indexed. */
+    startLine?: number;
+    endLine?: number;
+    text: string;
+    rerankScore?: number;
+    similarity?: number;
+}
+export interface CodeSearchResult {
+    hits: CodeHit[];
+    mode: RetrieveResult["mode"];
+    reranked: boolean;
+    notes: string[];
+}
+export interface CodeIndexStatus {
+    indexed: boolean;
+    files: number;
+    chunks: number;
+}
+/** One name for a workspace however its path was spelled. */
+export declare function codeCollectionName(root: string): string;
+export declare class CodeIndex {
+    readonly library: Library;
+    constructor(options?: CodeIndexOptions);
+    private collection;
+    status(root: string): CodeIndexStatus;
+    /**
+     * Bring the index in line with the files on disk: new and changed files
+     * are read and chunked, files that are gone are dropped. A file whose size
+     * and modification time are what they were is not read again.
+     *
+     * Searchable by keyword when this returns; `embed` adds the vectors.
+     */
+    sync(root: string, options?: CodeSyncOptions): Promise<CodeSyncResult>;
+    /** Give vectors to every indexed file that has none for this embedding model. */
+    embed(root: string, options: CodeEmbedOptions): Promise<EmbedPendingResult>;
+    search(root: string, query: string, options?: CodeSearchOptions): Promise<CodeSearchResult>;
+    /** Forget a workspace: its files, chunks and vectors. */
+    remove(root: string): Promise<boolean>;
+    close(): Promise<void>;
+}
+/** The code index of the data folder, opened once. */
+export declare function sharedCodeIndex(): CodeIndex;
+export declare function closeSharedCodeIndex(): Promise<void>;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"name": "@cerebriline/core",
 	"description": "Cerebriline Core SDK for Node Runtime",
-	"version": "4.100.243",
+	"version": "4.100.245",
 	"repository": {
 		"type": "git",
 		"url": "git+https://github.com/mann1x/cline.git",
@@ -69,9 +69,9 @@
 		"ws": "^8.20.0",
 		"yaml": "^2.8.2",
 		"zod": "^4.3.6",
-		"@cerebriline/agents": "4.100.243",
-		"@cerebriline/llms": "4.100.243",
-		"@cerebriline/shared": "4.100.243"
+		"@cerebriline/agents": "4.100.245",
+		"@cerebriline/llms": "4.100.245",
+		"@cerebriline/shared": "4.100.245"
 	},
 	"peerDependencies": {
 		"posthog-node": "^5.8.0"