### NeuroAPI Response Example (JSON)

Source: https://neuroapi.host/docs/overview/index

Illustrates a typical JSON response from NeuroAPI, showing the structure for a chat completion, including the message content, usage statistics, and model identifier.

```json
{
  "id": "chatcmpl-xxxxxxxxxxxxxx",
  "choices": [
    {
      "finish_reason": "stop",
      "message": {
        "role": "assistant",
        "content": "Hello!"
      }
    }
  ],
  "usage": {
    "prompt_tokens": 10,
    "completion_tokens": 2,
    "total_tokens": 12
  },
  "model": "gpt-4o"
}
```

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

### Finish Reason Normalization

Source: https://neuroapi.host/docs/overview/index

Explains the normalized values for the `finish_reason` field across different models and providers.

```APIDOC
## Finish Reason

NeuroAPI normalizes the `finish_reason` of each model to one of the following values to provide a consistent experience across different providers:

- `tool_calls`: The model stopped because it intended to call a tool.
- `stop`: The model stopped because it reached a natural stopping point.
- `length`: The model stopped because it reached the maximum token limit.
- `content_filter`: The model stopped due to a content filter.
- `function_call`: The model stopped because it intended to call a function (used for legacy function calling).

Note: Some models or providers might have additional, specific finish reasons. The original `finish_reason` string returned by the model is accessible via the `native_finish_reason` property if needed.
```

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

### Token Counting

Source: https://neuroapi.host/docs/overview/index

Details how tokens are counted for requests, covering both text and image data.

```APIDOC
## Token Counting

The system counts tokens before processing a request to estimate costs and enforce limits. This count includes tokens from message content and any tools used.

### Text Data Tokenization

- A tokenizer compatible with OpenAI models is used for text content.
- The token count includes tokens from message roles (e.g., 'user', 'assistant') and the content of each message.
- Tokens from function descriptions provided in the `tools` parameter are also included in the count.

### Image Data Tokenization

- The token cost for processing images depends on their size and the `detail` parameter within the `image_url` object.
- **`detail: 'low'`**: The image is processed at a lower resolution, incurring a fixed and lower token cost.
- **`detail: 'high'`**: The image is processed at a higher resolution. It is scaled and then divided into 512x512 pixel tiles. The final token cost is determined by the number of these tiles.
```

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

### NeuroAPI Response Schema (TypeScript)

Source: https://neuroapi.host/docs/overview/index

Defines the structure of the response from NeuroAPI, normalizing it to match the OpenAI Chat API. Includes types for non-streaming and streaming choices, usage data, and error information.

```typescript
type Response = {
  id: string;
  choices: (NonStreamingChoice | StreamingChoice)[];
  created: number;
  model: string;
  object: 'chat.completion' | 'chat.completion.chunk';
  system_fingerprint?: string;
  usage?: ResponseUsage;
};

type ResponseUsage = {
  prompt_tokens: number;
  completion_tokens: number;
  total_tokens: number;
};

type NonStreamingChoice = {
  finish_reason: string | null;
  message: {
    content: string | null;
    role: string;
    tool_calls?: ToolCall[];
  };
  error?: ErrorResponse;
};

type StreamingChoice = {
  finish_reason: string | null;
  delta: {
    content: string | null;
    role?: string;
    tool_calls?: ToolCall[];
  };
  error?: ErrorResponse;
};

type ErrorResponse = {
  code: number;
  message: string;
  metadata?: Record<string, unknown>;
};

type ToolCall = {
  id: string;
  type: 'function';
  function: {
    name: string;
    arguments: string;
  };
};
```

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

### Chat Completion Response Schema

Source: https://neuroapi.host/docs/overview/index

Defines the structure of the response received from the chat completion endpoint, including details for both streaming and non-streaming outputs.

```APIDOC
## Chat Completion Response Schema

This schema describes the structure of the response from the chat completion endpoint. It is designed to be compatible with the OpenAI Chat API.

### Response Structure

- **id** (string): Unique chat identifier.
- **choices** (Array<NonStreamingChoice | StreamingChoice>): An array containing the generated choices. The structure differs based on the `stream` parameter.
- **created** (number): Unix timestamp (in seconds) of chat creation.
- **model** (string): The model used for generation (e.g., "gpt-4o", "claude-3-sonnet").
- **object** (string): Type of the object, either 'chat.completion' or 'chat.completion.chunk'.
- **system_fingerprint** (string, optional): A unique identifier for the internal backend configuration used.
- **usage** (ResponseUsage, optional): Token usage data for the request. Sent in the last message when streaming.

### Response Usage

- **prompt_tokens** (number): Number of tokens in the prompt.
- **completion_tokens** (number): Number of tokens in the generated response.
- **total_tokens** (number): Total number of tokens in the request and response.

### Non-Streaming Choice

- **finish_reason** (string | null): The reason the model stopped generating tokens ('stop', 'length', 'tool_calls', 'content_filter').
- **message** (object):
    - **content** (string | null): The generated message content.
    - **role** (string): Author's role ('assistant').
    - **tool_calls** (Array<ToolCall>, optional): Tools called by the model.
- **error** (ErrorResponse, optional): Error information, if any occurred.

### Streaming Choice

- **finish_reason** (string | null): The reason the model stopped generating tokens. Appears only in the last chunk.
- **delta** (object):
    - **content** (string | null): Part of the message content.
    - **role** (string, optional): Author's role (usually in the first fragment).
    - **tool_calls** (Array<ToolCall>, optional): Tools called by the model (appear in parts).
- **error** (ErrorResponse, optional): Error information, if any occurred.

### Error Response

- **code** (number): Error code.
- **message** (string): Error message.
- **metadata** (Record<string, unknown>, optional): Additional metadata about the error.

### Tool Call

- **id** (string): Tool call ID.
- **type** (string): Tool type ('function').
- **function** (object):
    - **name** (string): Function name.
    - **arguments** (string): Function arguments as a JSON string.

### Response Example

```json
{
  "id": "chatcmpl-xxxxxxxxxxxxxx",
  "choices": [
    {
      "finish_reason": "stop",
      "message": {
        "role": "assistant",
        "content": "Hello!"
      }
    }
  ],
  "usage": {
    "prompt_tokens": 10,
    "completion_tokens": 2,
    "total_tokens": 12
  },
  "model": "gpt-4o"
}
```
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.