Overshoot SDK

https://docs.overshoot.ai
Admin
The Overshoot SDK enables users to run AI models on live video streams in real-time by providing a...
Tokens:3,323
Snippets:26
Trust Score:2.6
Update:8 months ago
Show doc for...
Tokens:
Raw
### RealtimeVision Initialization with Processing Parameters

Source: https://docs.overshoot.ai/getting-started/configuration

Demonstrates how to initialize the RealtimeVision client with advanced processing parameters to control video analysis behavior.

```APIDOC
## RealtimeVision Initialization with Processing Parameters

### Description
Initializes the RealtimeVision client, including optional processing parameters that fine-tune how the video stream is analyzed.

### Method
Constructor

### Endpoint
N/A (Client-side initialization)

### Parameters
#### Path Parameters
None

#### Query Parameters
None

#### Request Body
- **`apiUrl`** (string) - Required - The base URL for the Overshoot AI API.
- **`apiKey`** (string) - Required - Your secret API key for authentication.
- **`prompt`** (string) - Required - The initial instruction or question for the AI regarding the video content.
- **`processing`** (object) - Optional - An object containing parameters to control video processing.
  - **`clip_length_seconds`** (number) - Optional - The duration of video clips analyzed by the AI (default: 1 second).
  - **`delay_seconds`** (number) - Optional - The interval between receiving new results (default: 1 second).
  - **`fps`** (number) - Optional - The maximum frames per second to capture from the video source (default: 30).
  - **`sampling_ratio`** (number) - Optional - The fraction of frames to send to the AI for analysis (default: 0.1, i.e., 10%).

### Request Example
```javascript
const vision = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Read any visible text',
  processing: {
    clip_length_seconds: 1,
    delay_seconds: 1,
    fps: 30,
    sampling_ratio: 0.1
  }
});
```

### Response
N/A (This is a client-side initialization)

#### Success Response (200)
N/A

#### Response Example
N/A
```

--------------------------------

### Install Overshoot AI SDK using npm

Source: https://docs.overshoot.ai/getting-started

Installs the Overshoot AI SDK package using npm. This is the first step to integrate Overshoot AI into your project.

```bash
npm install @overshoot/sdk
```

--------------------------------

### RealtimeVision Initialization

Source: https://docs.overshoot.ai/getting-started/configuration

This section covers the initial configuration of the RealtimeVision client, including API endpoint, authentication key, the primary prompt for the AI, and the video source.

```APIDOC
## RealtimeVision Initialization

### Description
Initializes the RealtimeVision client with essential configuration parameters.

### Method
Constructor

### Endpoint
N/A (Client-side initialization)

### Parameters
#### Path Parameters
None

#### Query Parameters
None

#### Request Body
- **`apiUrl`** (string) - Required - The base URL for the Overshoot AI API.
- **`apiKey`** (string) - Required - Your secret API key for authentication.
- **`prompt`** (string) - Required - The initial instruction or question for the AI regarding the video content.
- **`source`** (object) - Required - Specifies the video input source.
  - **`type`** (string) - Required - The type of source, e.g., 'camera' or 'video'.
  - **`cameraFacing`** (string) - Optional - If `type` is 'camera', specifies which camera to use (e.g., 'environment', 'user').

### Request Example
```javascript
const vision = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Read any visible text',
  source: { type: 'camera', cameraFacing: 'environment' }
});
```

### Response
N/A (This is a client-side initialization)

#### Success Response (200)
N/A

#### Response Example
N/A
```

--------------------------------

### Basic Real-time Vision Processing with Overshoot AI SDK

Source: https://docs.overshoot.ai/getting-started

Demonstrates how to initialize and use the RealtimeVision class for real-time camera processing. It requires an API key, API URL, and a prompt. Results are handled via a callback function.

```javascript
import { RealtimeVision } from '@overshoot/sdk'
 
const vision = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Read any visible text',
  onResult: (result) => {
    console.log(result.result)
  }
})
 
await vision.start()   // starts the camera and begins processing
await vision.stop()    // stops everything
```

--------------------------------

### Update Prompt

Source: https://docs.overshoot.ai/getting-started/configuration

Allows dynamically changing the AI's prompt during an active video stream without needing to reinitialize the client.

```APIDOC
## Update Prompt

### Description
Updates the AI's prompt dynamically while the stream is active. This allows for changing the focus of analysis without restarting the stream.

### Method
`updatePrompt`

### Endpoint
N/A (Client-side method)

### Parameters
#### Path Parameters
None

#### Query Parameters
None

#### Request Body
- **`newPrompt`** (string) - Required - The new prompt to be used for subsequent analysis.

### Request Example
```javascript
// Change what the AI looks for mid-stream
vision.updatePrompt('Count the number of people');
```

### Response
N/A (This is a client-side method)

#### Success Response (200)
N/A

#### Response Example
N/A
```

--------------------------------

### Video File Analysis with Overshoot AI SDK

Source: https://docs.overshoot.ai/getting-started

Shows how to use the RealtimeVision class to process a video file. Similar to real-time processing, it requires configuration and provides results continuously as the video plays.

```javascript
const vision = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Describe what you see',
  source: { type: 'video', file: videoFile },
  onResult: (result) => {
    console.log(result.result)
  }
})
```

--------------------------------

### Configure Camera and Video File Input with Overshoot AI SDK (JavaScript)

Source: https://docs.overshoot.ai/getting-started/video-input-source

Demonstrates how to initialize the RealtimeVision SDK to use either the device's camera or a video file as the input source. The camera is the default, with options to specify camera facing. Video file input requires a file object. Both examples include API URL, API key, and a prompt.

```javascript
// Use the camera. This is the default if you don't specify it. 
const vision = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Read any visible text',
  source: { type: 'camera', cameraFacing: 'user'}
})

// Use a video file instead (e.g. from an <input type="file">)
const visionFromFile = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Describe what you see',
  source: { type: 'video', file: videoFile }
})
```

--------------------------------

### Initialize Realtime Vision Stream with Basic Configuration

Source: https://docs.overshoot.ai/getting-started/configuration

Initializes the RealtimeVision client with essential configuration parameters including the API endpoint, your secret API key, the initial prompt for the AI, and the video source. The source can be set to use the camera with a specified facing direction.

```javascript
const vision = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Read any visible text',
  source: { type: 'camera', cameraFacing: 'environment' }
})
```

--------------------------------

### Update AI Prompt Mid-Stream

Source: https://docs.overshoot.ai/getting-started/configuration

Allows for dynamic modification of the AI's prompt while the RealtimeVision stream is active. This enables changing the AI's task or questions without needing to restart the stream, making the interaction more flexible.

```javascript
// Change what the AI looks for mid-stream
vision.updatePrompt('Count the number of people')
```

--------------------------------

### Initialize Realtime Vision Stream with Advanced Processing Parameters

Source: https://docs.overshoot.ai/getting-started/configuration

Initializes the RealtimeVision client with both basic configuration and advanced processing parameters. These parameters control the video analysis window, result frequency, frame capture rate, and the proportion of frames sent for analysis, allowing for fine-tuning of performance and cost.

```javascript
const vision = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Read any visible text',
  processing: {
    clip_length_seconds: 1,
    delay_seconds: 1,
    fps: 30,
    sampling_ratio: 0.1
  }
})
```

--------------------------------

### Initialize RealtimeVision SDK (JavaScript)

Source: https://docs.overshoot.ai/getting-started/models

Initializes the `RealtimeVision` SDK for interacting with Overshoot AI's vision models. Requires an API URL, API key, a prompt for the model, and optionally a specific model name. If no model is specified, the default model is used.

```javascript
const vision = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Read any visible text',
  model: 'Qwen/Qwen3-VL-30B-A3B-Instruct'
})
```

--------------------------------

### Overshoot AI onResult Callback (JavaScript)

Source: https://docs.overshoot.ai/getting-started/output

Handles the AI's analysis results by logging the response, inference latency, and total latency. This callback is invoked each time the AI finishes analyzing a video window.

```javascript
onResult: (result) => {
  console.log(result.result)                    // the AI's response
  console.log(result.inference_latency_ms)      // model time (ms)
  console.log(result.total_latency_ms)          // end-to-end time (ms)
}
```

--------------------------------

### Overshoot AI Structured JSON Output (JavaScript)

Source: https://docs.overshoot.ai/getting-started/output

Configures Overshoot AI to return JSON output by specifying an outputSchema. This allows for structured data retrieval, such as counting objects in a frame. The result.result is then parsed as JSON.

```javascript
const vision = new RealtimeVision({
  apiUrl: 'https://cluster1.overshoot.ai/api/v0.2',
  apiKey: 'your-api-key',
  prompt: 'Count the people in the frame',
  outputSchema: {
    type: 'object',
    properties: {
      count: { type: 'number' }
    }
  },
  onResult: (result) => {
    const data = JSON.parse(result.result)
    console.log('People count:', data.count)
  }
})
```
Tokens:
Raw