brainbank 0.1.3 → 0.1.4

package/src/types.ts

@@ -0,0 +1,509 @@
+/**
+ * BrainBank — Type Definitions
+ * 
+ * All interfaces and types for the semantic knowledge bank.
+ */
+
+
+import type { CollectionItem, CollectionSearchOptions, CollectionAddOptions } from './services/collection.ts';
+
+// Re-export collection types so consumers don't need to import from services/
+export type { CollectionItem, CollectionSearchOptions, CollectionAddOptions };
+
+
+/** Public contract for a KV collection. Plugins depend on this interface, not the concrete class. */
+export interface ICollection {
+    /** Collection name. */
+    readonly name: string;
+    /** Add an item. Returns its ID. */
+    add(content: string, options?: CollectionAddOptions | Record<string, unknown>): Promise<number>;
+    /** Update an item's content (re-embeds). Returns the new ID. */
+    update(id: number, content: string, options?: CollectionAddOptions): Promise<number>;
+    /** Add multiple items. Returns their IDs. */
+    addMany(items: { content: string; metadata?: Record<string, unknown>; tags?: string[]; ttl?: string }[]): Promise<number[]>;
+    /** Search this collection. */
+    search(query: string, options?: CollectionSearchOptions): Promise<CollectionItem[]>;
+    /** List items (newest first). */
+    list(options?: { limit?: number; offset?: number; tags?: string[] }): CollectionItem[];
+    /** Count items in this collection. */
+    count(): number;
+    /** Keep only the N most recent items. */
+    trim(options: { keep: number }): Promise<{ removed: number }>;
+    /** Remove items older than a duration string. */
+    prune(options: { olderThan: string }): Promise<{ removed: number }>;
+    /** Remove a specific item by ID. */
+    remove(id: number): void;
+    /** Clear all items in this collection. */
+    clear(): void;
+}
+
+
+export interface BrainBankConfig {
+    /** Root path of the repository to index. Default: '.' */
+    repoPath?: string;
+    /** SQLite database path. Default: '.brainbank/data/brainbank.db' */
+    dbPath?: string;
+
+    /** Max git commits to index. Default: 500 */
+    gitDepth?: number;
+    /** Max file size in bytes to index. Default: 512_000 (500KB) */
+    maxFileSize?: number;
+    /** Max diff bytes per commit. Default: 8192 */
+    maxDiffBytes?: number;
+    /** HNSW M parameter (connections per node). Default: 16 */
+    hnswM?: number;
+    /** HNSW efConstruction (build-time candidates). Default: 200 */
+    hnswEfConstruction?: number;
+    /** HNSW efSearch (query-time candidates). Default: 50 */
+    hnswEfSearch?: number;
+    /** Embedding dimensions. Default: 384 */
+    embeddingDims?: number;
+    /** Max HNSW elements. Default: 2_000_000 */
+    maxElements?: number;
+    /** Custom embedding provider (default: local WASM model) */
+    embeddingProvider?: EmbeddingProvider;
+    /** Optional LLM noise filter — drops irrelevant results before formatting */
+    pruner?: Pruner;
+    /** Optional LLM context expander — discovers additional relevant chunks after pruning */
+    expander?: Expander;
+    /** Port for optional webhook server (enables push-based watch plugins). */
+    webhookPort?: number;
+    /** Context field defaults from config.json "context" section. */
+    contextFields?: Record<string, unknown>;
+}
+
+export interface ResolvedConfig {
+    repoPath: string;
+    dbPath: string;
+    gitDepth: number;
+    maxFileSize: number;
+    maxDiffBytes: number;
+    hnswM: number;
+    hnswEfConstruction: number;
+    hnswEfSearch: number;
+    embeddingDims: number;
+    maxElements: number;
+    embeddingProvider?: EmbeddingProvider;
+    pruner?: Pruner;
+    expander?: Expander;
+    webhookPort?: number;
+    /** Context field defaults from config.json "context" section. */
+    contextFields?: Record<string, unknown>;
+}
+
+
+export interface EmbeddingProvider {
+    /** Vector dimensions produced by this provider. */
+    readonly dims: number;
+    /** Embed a single text string. */
+    embed(text: string): Promise<Float32Array>;
+    /** Embed multiple texts (batch). */
+    embedBatch(texts: string[]): Promise<Float32Array[]>;
+    /** Release resources. */
+    close(): Promise<void>;
+}
+
+
+
+/** Item passed to the pruner for noise classification. */
+export interface PrunerItem {
+    /** Positional index (used to map back to SearchResult[]) */
+    id: number;
+    /** File path — primary signal for relevance */
+    filePath: string;
+    /** Trimmed content preview */
+    preview: string;
+    /** Chunk metadata (type, name, language, lines, etc.) */
+    metadata: Record<string, unknown>;
+}
+
+export interface Pruner {
+    /**
+     * Filter noise from search results and order by semantic relevance.
+     * @param query - The search query
+     * @param items - Items to evaluate (filePath + metadata + trimmed preview)
+     * @param context - Optional full task context (e.g. issue description, requirements)
+     * @returns Array of item IDs to KEEP, ordered by relevance to the query
+     *          (most relevant first). Everything else is dropped.
+     */
+    prune(query: string, items: PrunerItem[], context?: string): Promise<number[]>;
+    /** Release resources. */
+    close?(): Promise<void>;
+}
+
+/** Lightweight chunk descriptor for the expander manifest. */
+export interface ExpanderManifestItem {
+    /** Chunk ID (code_chunks.id) */
+    id: number;
+    /** File path */
+    filePath: string;
+    /** Chunk name (e.g. function/class name) */
+    name: string;
+    /** Chunk type (function, class, method, etc.) */
+    chunkType: string;
+    /** Line range (e.g. "L45-L89") */
+    lines: string;
+    /** True if this chunk is from an import-graph neighbor of the search results. */
+    priority?: boolean;
+    /** Brief content synopsis (~150 chars) for priority chunks — helps expander LLM understand chunk purpose. */
+    synopsis?: string;
+}
+
+/** Result from the expander: chunk IDs to add + optional free-text note. */
+export interface ExpanderResult {
+    /** Additional chunk IDs to splice into results. */
+    ids: number[];
+    /** Brief observation for the agent (e.g. "file X not found", "module Y is deprecated"). */
+    note?: string;
+}
+
+export interface Expander {
+    /**
+     * Given a task and a manifest of available chunks,
+     * return additional chunk IDs + an optional contextual note.
+     * Runs after pruning. Fail-open: errors return empty result.
+     *
+     * @param query - The search query / task description
+     * @param currentIds - IDs of chunks already in results (from search + prune)
+     * @param manifest - All available chunks (lightweight descriptors)
+     * @param context - Optional full task context (e.g. issue description, requirements)
+     * @returns Chunk IDs to add + optional note
+     */
+    expand(query: string, currentIds: number[], manifest: ExpanderManifestItem[], context?: string): Promise<ExpanderResult>;
+    /** Release resources. */
+    close?(): Promise<void>;
+}
+
+
+export interface SearchHit {
+    id: number;
+    score: number;
+}
+
+export interface VectorIndex {
+    /** Initialize the index. Must be called before add/search. */
+    init(): Promise<this>;
+    /** Add a vector with an integer ID. Idempotent: duplicate IDs are skipped. */
+    add(vector: Float32Array, id: number): void;
+    /** Mark a vector as deleted so it no longer appears in searches. */
+    remove(id: number): void;
+    /** Search for k nearest neighbors. */
+    search(query: Float32Array, k: number): SearchHit[];
+    /** Clear all vectors and reset to empty state. */
+    reinit(): void;
+    /** Number of vectors in the index. */
+    readonly size: number;
+}
+
+
+export interface CodeChunk {
+    /** Auto-incremented DB id (set after insert) */
+    id?: number;
+    /** Relative file path from repo root */
+    filePath: string;
+    /** Chunk type: 'file' | 'function' | 'class' | 'block' */
+    chunkType: string;
+    /** Function/class name (if detected) */
+    name?: string;
+    /** Start line (1-indexed) */
+    startLine: number;
+    /** End line (1-indexed, inclusive) */
+    endLine: number;
+    /** Raw content of the chunk */
+    content: string;
+    /** Language identifier */
+    language: string;
+}
+
+
+export interface GitCommitRecord {
+    id?: number;
+    hash: string;
+    shortHash: string;
+    message: string;
+    author: string;
+    date: string;
+    timestamp: number;
+    filesChanged: string[];
+    diff?: string;
+    additions: number;
+    deletions: number;
+    isMerge: boolean;
+}
+
+
+
+
+
+export type SearchResultType = 'code' | 'commit' | 'document' | 'collection';
+
+// Typed metadata per result type
+
+export interface CodeResultMetadata {
+    /** Database chunk ID (used by call graph annotations). */
+    id?: number;
+    /** File path (may duplicate CodeResult.filePath for metadata-only access). */
+    filePath?: string;
+    /** Adjacent chunk IDs from the same file (used by context expansion). */
+    chunkIds?: number[];
+    chunkType: string;
+    name?: string;
+    startLine: number;
+    endLine: number;
+    language: string;
+    searchType?: string;
+    rrfScore?: number;
+}
+
+export interface CommitResultMetadata {
+    hash: string;
+    shortHash: string;
+    author: string;
+    date: string;
+    files: string[];
+    additions?: number;
+    deletions?: number;
+    diff?: string;
+    searchType?: string;
+    rrfScore?: number;
+}
+
+
+
+export interface DocumentResultMetadata {
+    collection?: string;
+    title?: string;
+    seq?: number;
+    path?: string;
+    searchType?: string;
+    /** Internal chunk ID used by hybrid search to map fused results. */
+    chunkId?: number;
+    rrfScore?: number;
+}
+
+// Discriminated union
+
+export interface CodeResult {
+    type: 'code';
+    score: number;
+    filePath: string;
+    content: string;
+    context?: string;
+    metadata: CodeResultMetadata;
+}
+
+export interface CommitResult {
+    type: 'commit';
+    score: number;
+    filePath?: string;
+    content: string;
+    context?: string;
+    metadata: CommitResultMetadata;
+}
+
+
+
+export interface DocumentResult {
+    type: 'document';
+    score: number;
+    filePath: string;
+    content: string;
+    context?: string;
+    metadata: DocumentResultMetadata;
+}
+
+export interface CollectionResultMetadata {
+    id?: number;
+    collection?: string;
+    rrfScore?: number;
+    [key: string]: unknown;
+}
+
+export interface CollectionResult {
+    type: 'collection';
+    score: number;
+    filePath?: string;
+    content: string;
+    context?: string;
+    metadata: CollectionResultMetadata;
+}
+
+export type SearchResult = CodeResult | CommitResult | DocumentResult | CollectionResult;
+
+
+/** Narrow a SearchResult to CodeResult. */
+export function isCodeResult(r: SearchResult): r is CodeResult {
+    return r.type === 'code';
+}
+/** Narrow a SearchResult to CommitResult. */
+export function isCommitResult(r: SearchResult): r is CommitResult {
+    return r.type === 'commit';
+}
+/** Narrow a SearchResult to DocumentResult. */
+export function isDocumentResult(r: SearchResult): r is DocumentResult {
+    return r.type === 'document';
+}
+
+/** Narrow a SearchResult to CollectionResult. */
+export function isCollectionResult(r: SearchResult): r is CollectionResult {
+    return r.type === 'collection';
+}
+
+
+type MatchHandlers<T> = {
+    code?:       (r: CodeResult)       => T;
+    commit?:     (r: CommitResult)     => T;
+    document?:   (r: DocumentResult)   => T;
+    collection?: (r: CollectionResult) => T;
+    _?:          (r: SearchResult)     => T;
+};
+
+/**
+ * Pattern-match on SearchResult type. Calls the matching handler
+ * or the `_` fallback. Returns undefined if no handler matches.
+ */
+export function matchResult<T>(
+    result: SearchResult,
+    handlers: MatchHandlers<T>,
+): T | undefined {
+    switch (result.type) {
+        case 'code':       return (handlers.code       ?? handlers._)?.(result);
+        case 'commit':     return (handlers.commit     ?? handlers._)?.(result);
+        case 'document':   return (handlers.document   ?? handlers._)?.(result);
+        case 'collection': return (handlers.collection ?? handlers._)?.(result);
+    }
+}
+
+
+export interface ContextOptions {
+    /** Per-source result limits. Built-in: 'code', 'git'. Default: { code: 6, git: 5 } */
+    sources?: Record<string, number>;
+    /** Files the agent is about to modify (improves co-edit suggestions) */
+    affectedFiles?: string[];
+    /** Minimum similarity score threshold. Default: 0.25 */
+    minScore?: number;
+    /** Use MMR for diversity. Default: true */
+    useMMR?: boolean;
+    /** MMR lambda (0 = diversity, 1 = relevance). Default: 0.7 */
+    mmrLambda?: number;
+    /** Filter results to files under these path prefixes (e.g. 'src/services/' or ['src/', 'lib/']). */
+    pathPrefix?: string | string[];
+    /** Exclude results whose filePath starts with any of these prefixes (e.g. ['src/tests/', 'src/mocks/']). */
+    ignorePaths?: string[];
+    /** File paths to exclude from results (e.g. files already returned in a previous query). */
+    excludeFiles?: Set<string>;
+    /** Optional per-request pruner override (e.g. HaikuPruner for LLM noise filtering). */
+    pruner?: Pruner;
+    /** Caller origin for debug logging. */
+    source?: 'cli' | 'mcp' | 'daemon' | 'api';
+    /**
+     * General task context for the pruner — provides background signal.
+     * Pass a GitHub issue body, requirements doc, or conversation context.
+     * The pruner uses this to understand the broader task.
+     */
+    context?: string;
+    /**
+     * Specific pruning instructions — tells the pruner exactly what to focus on.
+     * E.g. "only keep files related to WebSocket JWT refresh".
+     * Both `context` and `pruner` are merged and sent to the Haiku pruner.
+     */
+    prunerContext?: string;
+    /**
+     * Context field overrides. Merged on top of config.json "context" defaults.
+     * Plugin-defined fields like `lines`, `callTree`, `symbols`, `compact`, `imports`.
+     * Example: `{ lines: true, callTree: { depth: 3 }, symbols: true }`
+     */
+    fields?: Record<string, unknown>;
+}
+
+
+export interface DocumentCollection {
+    /** Collection name (e.g. 'notes', 'docs') */
+    name: string;
+    /** Directory path to index */
+    path: string;
+    /** Glob pattern for files (default: all markdown) */
+    pattern?: string;
+    /** Glob patterns to ignore */
+    ignore?: string[];
+    /** Context description for this collection */
+    context?: string;
+}
+
+export interface DocChunk {
+    id?: number;
+    /** Collection name */
+    collection: string;
+    /** Relative file path within the collection */
+    filePath: string;
+    /** Document title (first heading or filename) */
+    title: string;
+    /** Chunk content */
+    content: string;
+    /** Chunk sequence within the document (0, 1, 2...) */
+    seq: number;
+    /** Character position in original document */
+    pos: number;
+    /** Content hash for incremental updates */
+    contentHash: string;
+}
+
+
+/** Plugin-provided stats. Key is the plugin name. */
+export interface IndexStats {
+    [pluginName: string]: Record<string, number | string> | undefined;
+}
+
+
+/** File-level progress (used by indexers). */
+export type ProgressCallback = (file: string, current: number, total: number) => void;
+
+/** Stage-level progress (used by BrainBank.index() orchestrator). */
+export type StageProgressCallback = (stage: string, message: string) => void;
+
+export interface IndexResult {
+    indexed: number;
+    skipped: number;
+    chunks?: number;
+    removed?: number;
+}
+
+
+export interface CoEditSuggestion {
+    file: string;
+    count: number;
+}
+
+
+/** Generalized watch event — works for files, APIs, webhooks. */
+export interface WatchEvent {
+    /** Event type. 'sync' is for batch/poll sources that don't distinguish CRUD. */
+    type: 'create' | 'update' | 'delete' | 'sync';
+    /** Unique ID of the changed item (file path, PR#123, PROJ-456, etc.). */
+    sourceId: string;
+    /** Source descriptor (e.g. 'file', 'github:pr', 'jira:card'). */
+    sourceName: string;
+    /** Optional raw payload to avoid re-fetching. */
+    payload?: unknown;
+}
+
+/** Callback that plugins invoke when they detect a change. */
+export type WatchEventHandler = (event: WatchEvent) => void;
+
+/** Lifecycle handle returned by WatchablePlugin.watch(). */
+export interface WatchHandle {
+    /** Stop watching and release resources. */
+    stop(): Promise<void>;
+    /** Whether the watcher is still active. */
+    readonly active: boolean;
+}
+
+/** Optional hints from plugin to core — debounce, batching, priority. */
+export interface WatchConfig {
+    /** Debounce interval in ms. 0 = process immediately. Default: inherited from WatchOptions. */
+    debounceMs?: number;
+    /** Max events to batch before triggering re-index. Default: unlimited. */
+    batchSize?: number;
+    /** Processing priority. Default: 'realtime'. */
+    priority?: 'realtime' | 'background';
+}

package/dist/chunk-2P3EGY6S.js

@@ -1,37 +0,0 @@
-import {
-  __name
-} from "./chunk-7QVYU63E.js";
-
-// src/embeddings/math.ts
-function cosineSimilarity(a, b) {
-  if (a.length !== b.length) {
-    throw new Error(`Vector dimension mismatch: ${a.length} vs ${b.length}`);
-  }
-  if (a.length === 0) return 0;
-  let dot = 0;
-  for (let i = 0; i < a.length; i++) {
-    dot += a[i] * b[i];
-  }
-  return dot;
-}
-__name(cosineSimilarity, "cosineSimilarity");
-function normalize(vec) {
-  let norm = 0;
-  for (let i = 0; i < vec.length; i++) {
-    norm += vec[i] * vec[i];
-  }
-  norm = Math.sqrt(norm);
-  if (norm === 0) return new Float32Array(vec.length);
-  const result = new Float32Array(vec.length);
-  for (let i = 0; i < vec.length; i++) {
-    result[i] = vec[i] / norm;
-  }
-  return result;
-}
-__name(normalize, "normalize");
-
-export {
-  cosineSimilarity,
-  normalize
-};
-//# sourceMappingURL=chunk-2P3EGY6S.js.map

package/dist/chunk-2P3EGY6S.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/embeddings/math.ts"],"sourcesContent":["/**\n * BrainBank — Math Utilities\n * \n * Pure vector math functions for similarity calculations.\n * No dependencies — works on Float32Array directly.\n */\n\n/**\n * Cosine similarity between two vectors.\n * Assumes vectors are already normalized (unit length).\n * Returns value between -1.0 and 1.0.\n */\nexport function cosineSimilarity(a: Float32Array, b: Float32Array): number {\n    if (a.length !== b.length) {\n        throw new Error(`Vector dimension mismatch: ${a.length} vs ${b.length}`);\n    }\n    if (a.length === 0) return 0;\n\n    let dot = 0;\n    for (let i = 0; i < a.length; i++) {\n        dot += a[i] * b[i];\n    }\n    return dot;\n}\n\n/**\n * Full cosine similarity (normalizes first).\n * Use this when vectors may not be pre-normalized.\n */\nexport function cosineSimilarityFull(a: Float32Array, b: Float32Array): number {\n    if (a.length !== b.length) {\n        throw new Error(`Vector dimension mismatch: ${a.length} vs ${b.length}`);\n    }\n    if (a.length === 0) return 0;\n\n    let dot = 0, normA = 0, normB = 0;\n    for (let i = 0; i < a.length; i++) {\n        dot += a[i] * b[i];\n        normA += a[i] * a[i];\n        normB += b[i] * b[i];\n    }\n    const denom = Math.sqrt(normA) * Math.sqrt(normB);\n    return denom === 0 ? 0 : dot / denom;\n}\n\n/**\n * L2-normalize a vector to unit length.\n * Returns a new Float32Array.\n */\nexport function normalize(vec: Float32Array): Float32Array {\n    let norm = 0;\n    for (let i = 0; i < vec.length; i++) {\n        norm += vec[i] * vec[i];\n    }\n    norm = Math.sqrt(norm);\n    if (norm === 0) return new Float32Array(vec.length);\n\n    const result = new Float32Array(vec.length);\n    for (let i = 0; i < vec.length; i++) {\n        result[i] = vec[i] / norm;\n    }\n    return result;\n}\n\n/**\n * Euclidean distance between two vectors.\n */\nexport function euclideanDistance(a: Float32Array, b: Float32Array): number {\n    if (a.length !== b.length) {\n        throw new Error(`Vector dimension mismatch: ${a.length} vs ${b.length}`);\n    }\n    let sum = 0;\n    for (let i = 0; i < a.length; i++) {\n        const d = a[i] - b[i];\n        sum += d * d;\n    }\n    return Math.sqrt(sum);\n}\n"],"mappings":";;;;;AAYO,SAAS,iBAAiB,GAAiB,GAAyB;AACvE,MAAI,EAAE,WAAW,EAAE,QAAQ;AACvB,UAAM,IAAI,MAAM,8BAA8B,EAAE,MAAM,OAAO,EAAE,MAAM,EAAE;AAAA,EAC3E;AACA,MAAI,EAAE,WAAW,EAAG,QAAO;AAE3B,MAAI,MAAM;AACV,WAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AAC/B,WAAO,EAAE,CAAC,IAAI,EAAE,CAAC;AAAA,EACrB;AACA,SAAO;AACX;AAXgB;AAqCT,SAAS,UAAU,KAAiC;AACvD,MAAI,OAAO;AACX,WAAS,IAAI,GAAG,IAAI,IAAI,QAAQ,KAAK;AACjC,YAAQ,IAAI,CAAC,IAAI,IAAI,CAAC;AAAA,EAC1B;AACA,SAAO,KAAK,KAAK,IAAI;AACrB,MAAI,SAAS,EAAG,QAAO,IAAI,aAAa,IAAI,MAAM;AAElD,QAAM,SAAS,IAAI,aAAa,IAAI,MAAM;AAC1C,WAAS,IAAI,GAAG,IAAI,IAAI,QAAQ,KAAK;AACjC,WAAO,CAAC,IAAI,IAAI,CAAC,IAAI;AAAA,EACzB;AACA,SAAO;AACX;AAbgB;","names":[]}

package/dist/chunk-3GAIDXRW.js

@@ -1,105 +0,0 @@
-import {
-  __name
-} from "./chunk-7QVYU63E.js";
-
-// src/embeddings/openai.ts
-var DEFAULT_MODEL = "text-embedding-3-small";
-var DEFAULT_DIMS = {
-  "text-embedding-3-small": 1536,
-  "text-embedding-3-large": 3072,
-  "text-embedding-ada-002": 1536
-};
-var API_URL = "https://api.openai.com/v1/embeddings";
-var MAX_BATCH = 100;
-var OpenAIEmbedding = class {
-  static {
-    __name(this, "OpenAIEmbedding");
-  }
-  dims;
-  _apiKey;
-  _model;
-  _baseUrl;
-  _requestDims;
-  _retrying = false;
-  constructor(options = {}) {
-    this._apiKey = options.apiKey ?? process.env.OPENAI_API_KEY ?? "";
-    this._model = options.model ?? DEFAULT_MODEL;
-    this._baseUrl = options.baseUrl ?? API_URL;
-    if (options.dims && this._model.startsWith("text-embedding-3")) {
-      this._requestDims = options.dims;
-      this.dims = options.dims;
-    } else {
-      this.dims = options.dims ?? DEFAULT_DIMS[this._model] ?? 1536;
-    }
-  }
-  async embed(text) {
-    const results = await this._request([text]);
-    return results[0];
-  }
-  async embedBatch(texts) {
-    if (texts.length === 0) return [];
-    const results = [];
-    for (let i = 0; i < texts.length; i += MAX_BATCH) {
-      const batch = texts.slice(i, i + MAX_BATCH);
-      const embeddings = await this._request(batch);
-      results.push(...embeddings);
-    }
-    return results;
-  }
-  async close() {
-  }
-  _isTokenLimitError(errText) {
-    return errText.includes("maximum input length") || errText.includes("maximum context length") || errText.includes("too many tokens");
-  }
-  async _request(input) {
-    if (!this._apiKey) {
-      throw new Error("OpenAI API key required. Set OPENAI_API_KEY env var or pass apiKey option.");
-    }
-    const MAX_CHARS = 24e3;
-    const safeInput = input.map((t) => t.length > MAX_CHARS ? t.slice(0, MAX_CHARS) : t);
-    const body = {
-      model: this._model,
-      input: safeInput
-    };
-    if (this._requestDims) {
-      body.dimensions = this._requestDims;
-    }
-    const res = await fetch(this._baseUrl, {
-      method: "POST",
-      headers: {
-        "Content-Type": "application/json",
-        "Authorization": `Bearer ${this._apiKey}`
-      },
-      body: JSON.stringify(body)
-    });
-    if (!res.ok) {
-      const err = await res.text();
-      const isTokenLimit = res.status === 400 && this._isTokenLimitError(err);
-      if (isTokenLimit && safeInput.length > 1) {
-        const results = [];
-        for (const text of safeInput) {
-          const r = await this._request([text.slice(0, 8e3)]);
-          results.push(r[0]);
-        }
-        return results;
-      }
-      if (isTokenLimit && safeInput.length === 1 && !this._retrying) {
-        this._retrying = true;
-        try {
-          return await this._request([safeInput[0].slice(0, 6e3)]);
-        } finally {
-          this._retrying = false;
-        }
-      }
-      throw new Error(`OpenAI embedding API error (${res.status}): ${err}`);
-    }
-    const json = await res.json();
-    const sorted = json.data.sort((a, b) => a.index - b.index);
-    return sorted.map((d) => new Float32Array(d.embedding));
-  }
-};
-
-export {
-  OpenAIEmbedding
-};
-//# sourceMappingURL=chunk-3GAIDXRW.js.map

package/dist/chunk-3GAIDXRW.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/embeddings/openai.ts"],"sourcesContent":["/**\n * BrainBank — OpenAI Embedding Provider\n * \n * Uses OpenAI's embedding API via fetch (no SDK dependency).\n * Supports text-embedding-3-small, text-embedding-3-large, and ada-002.\n * \n * Usage:\n *   const brain = new BrainBank({\n *     embeddingProvider: new OpenAIEmbedding({ model: 'text-embedding-3-small' }),\n *   });\n */\n\nimport type { EmbeddingProvider } from '../types.ts';\n\nconst DEFAULT_MODEL = 'text-embedding-3-small';\nconst DEFAULT_DIMS: Record<string, number> = {\n    'text-embedding-3-small': 1536,\n    'text-embedding-3-large': 3072,\n    'text-embedding-ada-002': 1536,\n};\nconst API_URL = 'https://api.openai.com/v1/embeddings';\nconst MAX_BATCH = 100; // OpenAI limit per request\n\nexport interface OpenAIEmbeddingOptions {\n    /** OpenAI API key. Falls back to OPENAI_API_KEY env var. */\n    apiKey?: string;\n    /** Model name. Default: 'text-embedding-3-small' */\n    model?: string;\n    /** Vector dimensions. If omitted, uses model default. text-embedding-3-* supports custom dims. */\n    dims?: number;\n    /** Base URL override (for Azure, proxies, etc.) */\n    baseUrl?: string;\n}\n\nexport class OpenAIEmbedding implements EmbeddingProvider {\n    readonly dims: number;\n\n    private _apiKey: string;\n    private _model: string;\n    private _baseUrl: string;\n    private _requestDims: number | undefined;\n    private _retrying = false;\n\n    constructor(options: OpenAIEmbeddingOptions = {}) {\n        this._apiKey = options.apiKey ?? process.env.OPENAI_API_KEY ?? '';\n        this._model = options.model ?? DEFAULT_MODEL;\n        this._baseUrl = options.baseUrl ?? API_URL;\n\n        // Custom dims only supported by text-embedding-3-*\n        if (options.dims && this._model.startsWith('text-embedding-3')) {\n            this._requestDims = options.dims;\n            this.dims = options.dims;\n        } else {\n            this.dims = options.dims ?? DEFAULT_DIMS[this._model] ?? 1536;\n        }\n    }\n\n    async embed(text: string): Promise<Float32Array> {\n        const results = await this._request([text]);\n        return results[0];\n    }\n\n    async embedBatch(texts: string[]): Promise<Float32Array[]> {\n        if (texts.length === 0) return [];\n\n        const results: Float32Array[] = [];\n\n        // Split into chunks of MAX_BATCH\n        for (let i = 0; i < texts.length; i += MAX_BATCH) {\n            const batch = texts.slice(i, i + MAX_BATCH);\n            const embeddings = await this._request(batch);\n            results.push(...embeddings);\n        }\n\n        return results;\n    }\n\n    async close(): Promise<void> {\n        // No resources to release\n    }\n\n    private _isTokenLimitError(errText: string): boolean {\n        return errText.includes('maximum input length') ||\n               errText.includes('maximum context length') ||\n               errText.includes('too many tokens');\n    }\n\n    private async _request(input: string[]): Promise<Float32Array[]> {\n        if (!this._apiKey) {\n            throw new Error('OpenAI API key required. Set OPENAI_API_KEY env var or pass apiKey option.');\n        }\n\n        // Truncate texts that would exceed token limit (~4 chars per token, 8192 max)\n        const MAX_CHARS = 24_000;\n        const safeInput = input.map(t => t.length > MAX_CHARS ? t.slice(0, MAX_CHARS) : t);\n\n        const body: Record<string, any> = {\n            model: this._model,\n            input: safeInput,\n        };\n\n        if (this._requestDims) {\n            body.dimensions = this._requestDims;\n        }\n\n        const res = await fetch(this._baseUrl, {\n            method: 'POST',\n            headers: {\n                'Content-Type': 'application/json',\n                'Authorization': `Bearer ${this._apiKey}`,\n            },\n            body: JSON.stringify(body),\n        });\n\n        if (!res.ok) {\n            const err = await res.text();\n            const isTokenLimit = res.status === 400 && this._isTokenLimitError(err);\n\n            // If token limit error in a batch, retry each item individually with more aggressive truncation\n            if (isTokenLimit && safeInput.length > 1) {\n                const results: Float32Array[] = [];\n                for (const text of safeInput) {\n                    const r = await this._request([text.slice(0, 8_000)]);\n                    results.push(r[0]);\n                }\n                return results;\n            }\n            // Last resort: if single item still fails, truncate to ~2k tokens\n            if (isTokenLimit && safeInput.length === 1 && !this._retrying) {\n                this._retrying = true;\n                try {\n                    return await this._request([safeInput[0].slice(0, 6_000)]);\n                } finally {\n                    this._retrying = false;\n                }\n            }\n            throw new Error(`OpenAI embedding API error (${res.status}): ${err}`);\n        }\n\n        const json = await res.json() as {\n            data: Array<{ embedding: number[]; index: number }>;\n        };\n\n        // Sort by index (API may return out of order)\n        const sorted = json.data.sort((a, b) => a.index - b.index);\n\n        return sorted.map(d => new Float32Array(d.embedding));\n    }\n}\n"],"mappings":";;;;;AAcA,IAAM,gBAAgB;AACtB,IAAM,eAAuC;AAAA,EACzC,0BAA0B;AAAA,EAC1B,0BAA0B;AAAA,EAC1B,0BAA0B;AAC9B;AACA,IAAM,UAAU;AAChB,IAAM,YAAY;AAaX,IAAM,kBAAN,MAAmD;AAAA,EAlC1D,OAkC0D;AAAA;AAAA;AAAA,EAC7C;AAAA,EAED;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA,YAAY;AAAA,EAEpB,YAAY,UAAkC,CAAC,GAAG;AAC9C,SAAK,UAAU,QAAQ,UAAU,QAAQ,IAAI,kBAAkB;AAC/D,SAAK,SAAS,QAAQ,SAAS;AAC/B,SAAK,WAAW,QAAQ,WAAW;AAGnC,QAAI,QAAQ,QAAQ,KAAK,OAAO,WAAW,kBAAkB,GAAG;AAC5D,WAAK,eAAe,QAAQ;AAC5B,WAAK,OAAO,QAAQ;AAAA,IACxB,OAAO;AACH,WAAK,OAAO,QAAQ,QAAQ,aAAa,KAAK,MAAM,KAAK;AAAA,IAC7D;AAAA,EACJ;AAAA,EAEA,MAAM,MAAM,MAAqC;AAC7C,UAAM,UAAU,MAAM,KAAK,SAAS,CAAC,IAAI,CAAC;AAC1C,WAAO,QAAQ,CAAC;AAAA,EACpB;AAAA,EAEA,MAAM,WAAW,OAA0C;AACvD,QAAI,MAAM,WAAW,EAAG,QAAO,CAAC;AAEhC,UAAM,UAA0B,CAAC;AAGjC,aAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK,WAAW;AAC9C,YAAM,QAAQ,MAAM,MAAM,GAAG,IAAI,SAAS;AAC1C,YAAM,aAAa,MAAM,KAAK,SAAS,KAAK;AAC5C,cAAQ,KAAK,GAAG,UAAU;AAAA,IAC9B;AAEA,WAAO;AAAA,EACX;AAAA,EAEA,MAAM,QAAuB;AAAA,EAE7B;AAAA,EAEQ,mBAAmB,SAA0B;AACjD,WAAO,QAAQ,SAAS,sBAAsB,KACvC,QAAQ,SAAS,wBAAwB,KACzC,QAAQ,SAAS,iBAAiB;AAAA,EAC7C;AAAA,EAEA,MAAc,SAAS,OAA0C;AAC7D,QAAI,CAAC,KAAK,SAAS;AACf,YAAM,IAAI,MAAM,4EAA4E;AAAA,IAChG;AAGA,UAAM,YAAY;AAClB,UAAM,YAAY,MAAM,IAAI,OAAK,EAAE,SAAS,YAAY,EAAE,MAAM,GAAG,SAAS,IAAI,CAAC;AAEjF,UAAM,OAA4B;AAAA,MAC9B,OAAO,KAAK;AAAA,MACZ,OAAO;AAAA,IACX;AAEA,QAAI,KAAK,cAAc;AACnB,WAAK,aAAa,KAAK;AAAA,IAC3B;AAEA,UAAM,MAAM,MAAM,MAAM,KAAK,UAAU;AAAA,MACnC,QAAQ;AAAA,MACR,SAAS;AAAA,QACL,gBAAgB;AAAA,QAChB,iBAAiB,UAAU,KAAK,OAAO;AAAA,MAC3C;AAAA,MACA,MAAM,KAAK,UAAU,IAAI;AAAA,IAC7B,CAAC;AAED,QAAI,CAAC,IAAI,IAAI;AACT,YAAM,MAAM,MAAM,IAAI,KAAK;AAC3B,YAAM,eAAe,IAAI,WAAW,OAAO,KAAK,mBAAmB,GAAG;AAGtE,UAAI,gBAAgB,UAAU,SAAS,GAAG;AACtC,cAAM,UAA0B,CAAC;AACjC,mBAAW,QAAQ,WAAW;AAC1B,gBAAM,IAAI,MAAM,KAAK,SAAS,CAAC,KAAK,MAAM,GAAG,GAAK,CAAC,CAAC;AACpD,kBAAQ,KAAK,EAAE,CAAC,CAAC;AAAA,QACrB;AACA,eAAO;AAAA,MACX;AAEA,UAAI,gBAAgB,UAAU,WAAW,KAAK,CAAC,KAAK,WAAW;AAC3D,aAAK,YAAY;AACjB,YAAI;AACA,iBAAO,MAAM,KAAK,SAAS,CAAC,UAAU,CAAC,EAAE,MAAM,GAAG,GAAK,CAAC,CAAC;AAAA,QAC7D,UAAE;AACE,eAAK,YAAY;AAAA,QACrB;AAAA,MACJ;AACA,YAAM,IAAI,MAAM,+BAA+B,IAAI,MAAM,MAAM,GAAG,EAAE;AAAA,IACxE;AAEA,UAAM,OAAO,MAAM,IAAI,KAAK;AAK5B,UAAM,SAAS,KAAK,KAAK,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;AAEzD,WAAO,OAAO,IAAI,OAAK,IAAI,aAAa,EAAE,SAAS,CAAC;AAAA,EACxD;AACJ;","names":[]}