Variable TRAGarConst

Type Declaration

• create: (config: CreateConfig) => Promise<TRAGarInstance>

• embedders: {
      custom(
          embed: (batch: string[]) => Promise<Float32Array<ArrayBufferLike>[]>,
          dim: number,
          modelId: string,
      ): CustomEmbedderConfig;
      transformers(modelId?: string, dim?: number): TransformersEmbedderConfig;
  }

  • custom: function

    • custom(
          embed: (batch: string[]) => Promise<Float32Array<ArrayBufferLike>[]>,
          dim: number,
          modelId: string,
      ): CustomEmbedderConfig

      Custom embedder — wraps a caller-supplied embed function. Use for tests, examples, and offline development.

      Parameters

      • embed: (batch: string[]) => Promise<Float32Array<ArrayBufferLike>[]>
      • dim: number
      • modelId: string

      Returns CustomEmbedderConfig

  • transformers: function

    • transformers(modelId?: string, dim?: number): TransformersEmbedderConfig

      Transformers.js embedder — lazy-loads the model on first ingest/query call. Defaults to Xenova/all-MiniLM-L6-v2 (English, ~23 MB, dim 384). Throws TRAGarError("EmbedderLoadFailed") if the module or model fails to load.

      Parameters

      • modelId: string = DEFAULT_MODEL
      • dim: number = DEFAULT_DIM

      Returns TransformersEmbedderConfig

• stores: {
      file(dirPath: string): FileStoreConfig;
      memory(): MemoryStoreConfig;
      opfs(
          opts?: {
              _backend?: FileBackend;
              _fallbackBackend?: FileBackend;
              _simulateOpfsFailure?: boolean;
          },
      ): OpfsStoreConfig;
  }

  • file: function

    • file(dirPath: string): FileStoreConfig

      File store — persists corpus data in a local filesystem directory. Node.js environments only. Throws InvalidConfig immediately when called in a browser or service worker.

      Files are written to {dirPath}/tragar/{namespace}/ and are byte-identical to stores.opfs() output, so they can be served as static assets and opened by a browser TRAGar instance using stores.opfs().

      Parameters

      • dirPath: string

      Returns FileStoreConfig

  • memory: function

    • memory(): MemoryStoreConfig

      In-memory store — no persistence, ideal for tests and the tracer bullet.

      Returns MemoryStoreConfig

  • opfs: function

    • opfs(
          opts?: {
              _backend?: FileBackend;
              _fallbackBackend?: FileBackend;
              _simulateOpfsFailure?: boolean;
          },
      ): OpfsStoreConfig

      OPFS store — persists corpus data in the browser's Origin Private File System. Falls back to IndexedDB when OPFS is unavailable (Safari < 16, locked-down WebViews) and emits onWarn("StoreFallback") when it does.

      Parameters

      • Optionalopts: {
            _backend?: FileBackend;
            _fallbackBackend?: FileBackend;
            _simulateOpfsFailure?: boolean;
        }

      Returns OpfsStoreConfig