* Build follow-up prompt after tool execution (IMPLEMENTED in C++ ToolCallingBridge) * * Creates continuation prompt with tool result for next L

* Convert VLM generation options to JSON string for C++ bridge. */

* Check if any state machine can handle this task

* Check if a model is compatible with the current device * Compares model RAM/storage requirements against device capabilities * @param modelId

* Check model status and load appropriate chat model. */

Clean up the STT resources.

Clean up the TTS resources.

* Clear a directory: delete all contents, then recreate the empty directory. */

* Configure the logging system. */

Helper to check if string contains substring (case-insensitive)

* Convert base64 PCM float32 audio to WAV file * TTS output is base64-encoded float32 PCM samples

Get current Unix timestamp

* Computes decode throughput in tokens/second. * Returns 0.0 if decode time is 0 or output_tokens is 0. */

* Delete a file or directory */

Delete from all storage backends (local filesystem, OPFS, memory cache).

* Check if a directory exists. */

* Emit download completed event via C++. */

* Emit download failed event via C++. */

Ensure the LLM component is created.

* Execute a suspending function with circuit breaker protection */

* Execute a platform LLM request. * * @param request The LLM request * @param callback Callback for streaming tokens (optio

* Extract all interactive elements from the page

* Extract relevant attributes for agent context

* Find label text for a form element

* @brief Find the first occurrence of a character in a string starting from a position * * @param str Input string * @param ch Character to find *

* Find the nested directory (single visible subdirectory) in an extracted archive. * Ported from Swift's ExtractionService.findNestedDirectory(). *

Create from a file path */

Create from raw RGB pixel data */

* Match an NPU chip from a SoC model string (e.g. "SM8750"). * Returns null if the SoC is not a supported NPU chipset. */

* Generate text (non-streaming) * @param prompt The prompt text * @param optionsJson JSON string with generation options: * - max_tokens: M

* Generate a unique ID for generation tracking. */

* Get all AI service types (components that process models). */

* Get all models — tries native first, falls back to local cache

* Get the API key for authentication. * Exposed for CppBridgeDevice to use in production mode. */

* Get available RAM in bytes

* Get backend info as JSON string

* Extract base model ID (remove extension)

* Get the base URL for device registration. * Exposed for CppBridgeDevice to use in production mode. */

* Get number of CPU cores

* Get the stored context (for platform bridge operations) */

* Get the application context (alias for applicationContext for compatibility) */

* Get meaningful text from an element

* Get the byte size of a stored file without reading it into memory. * * @param key - Model identifier or nested path * @returns File size in

* Get info for a specific model * @param modelId Model identifier * @returns JSON with model info

* Get the string representation of an HTTP method. */

* Get the string representation of an HTTP method. */

* Get the download progress as a percentage (0-100). */

* Retrieve a stored string value * @param key Unique identifier for the value * @return Stored string value or null if not found * @thr

* Get total RAM in bytes

* Get a valid access token, automatically refreshing if needed. * This is the preferred way to get the token for requests. * * @return

Get the base directory for model storage

* Handle RESULTS state - select a product

* Handle SEARCHING state - type query and submit

* Check if device has GPU

* Check if a model exists in OPFS. * * @param key - Model identifier or nested path

* Check if device has NPU (Neural Processing Unit)

* Check if C++ tool calling functions are available in the WASM module. * * On WebGPU builds, all WASM imports are wrapped with WebAssembly.Suspendi