> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nullboard.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# API Reference

> Detailed documentation for the Ink class and types.

## `Ink` Class

### `constructor(options?: InkOptions)`

Creates a new instance of the Ink recognition engine.

**`InkOptions` Properties:**

* `auto` (boolean): Whether to automatically train the model on load.
* `modelUrl` (string): Path to a pre-trained TensorFlow\.js model JSON file (disables auto-training if provided).
* `trainingData` (Record): Custom data to train on.
* `onTrainProgress` (function): Callback for monitoring training status.

### `loadModel(): Promise<void>`

Initializes the model. If `auto` is true, this triggers the training process using bundled data.

### `isReady(): boolean`

Returns `true` if the model is trained/loaded and ready to recognize input.

### `startStroke(): Stroke`

Creates a brand new stroke and returns it. Future points should be added to this object.

### `undo(): Stroke | undefined`

Removes the most recently added stroke from the internal history. Useful for building "Undo" UI buttons.

### `clear(): void`

Resets the internal state, removing all strokes.

### `recognize(): Promise<InkResult>`

asynchronously processes all current strokes and returns the recognition result.

## `InkResult` Interface

The object returned by `recognize()` contains:

* **`expression`** (`string`): The normalized mathematical expression (e.g., "1 + 2").
* **`result`** (`number | null`): The evaluated result of the expression.
* **`valid`** (`boolean`): `true` if the input was successfully parsed as valid math.
* **`characters`** (`RecognitionResult[]`): Array containing confidence scores and bounding boxes for each recognized symbol.