deepseek-api

DeepSeek API

DeepSeek API is a powerful Node.js library designed to interact with DeepSeek's chat system. With this library, you can create chat sessions, send messages, and handle chat streams efficiently.

Installation

To install the DeepSeek API package, use the following command:

npm install deepseek-api

Core Features

• Send Messages: Interact with DeepSeek chats by sending messages.
• Chat Streams: Stream responses in real-time.
• Chat Management: Create and manage chat sessions.
• Structured Data: Well-formatted response objects for easy integration.

Important Note

DeepSeek uses Cloudflare for changing IPs, which might cause issues when using this API server-side with tokens obtained from your browser. We are actively working on a workaround solution to address this.

API Reference

Importing the Library

import { sendMessage, createNewChat, chats } from 'deepseek-api';

Functions

createNewChat(token: string, id?: string): Promise<string | { error: string }>

Creates a new chat session and stores its details in the chats map.

• Parameters:

  • token (string): The authentication token for DeepSeek.
  • id (string, optional): A custom chat ID. If not provided, the original ID from DeepSeek will be used.

• Returns:

  • A promise that resolves to the chat ID or an error object.

• Example:

const chatID = await createNewChat('your-token-here');
console.log(`Chat created with ID: ${chatID}`);

sendMessage(text: string, chat: { id: string; token: string; parent_id?: string }, callback: Function): Promise<any>

Sends a message to the specified chat session and streams the response.

• Parameters:

  • text (string): The message to send.
  • chat (object): The chat details, including:
    • id (string): The chat session ID.
    • token (string): The authentication token.
    • parent_id (string, optional): The ID of the message to reply to (defaults to the last message in the chat).
  • callback (Function): A callback function to process each chunk of the response.

• Returns:

  • A promise that resolves to the final response data or an error object.

• Example:

await sendMessage('Hello, DeepSeek!', { 
  id: 'chat-id-here', 
  token: 'your-token-here' 
}, (chunk) => {
  console.log('Stream chunk:', chunk);
});

chats

A Map object that stores all chat details, indexed by their IDs. Each chat entry includes:

• id: The chat session ID.
• token: The authentication token.
• cookies: Any associated cookies (currently null).

Utility Functions

requestChatStream(payload: object, text: string): Promise<any>

Initiates a chat stream request.

streamResponse(response: any, callback: Function): Promise<any>

Processes a streaming response, invoking the callback for each chunk.

Authentication

To authenticate with DeepSeek:

• Log in to DeepSeek Chat.

• Retrieve the token using the browser's console:

  localStorage.getItem('userToken');
  

  The token will look like this:

  {"value":"your-token-here","__version":"0"}
  

• Use the value field from the token object as your authentication token.

Usage Example

Here is a complete example of creating a chat session and sending a message:

import { createNewChat, sendMessage } from 'deepseek-api';

(async () => {
  const token = 'your-token-here';

  // Step 1: Create a new chat session
  const chatID = await createNewChat(token);

  if (typeof chatID === 'string') {
    console.log(`Chat session created: ${chatID}`);

    // Step 2: Send a message
    await sendMessage('Hello, DeepSeek!', {
      id: chatID,
      token: token
    }, (chunk) => {
      console.log('Response chunk:', chunk);
    });
  } else {
    console.error('Error creating chat session:', chatID.error);
  }
})();

Error Handling

• Ensure that the chat ID is valid and registered in the chats map before sending a message.
• Check that the token is correctly retrieved and passed to the functions.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Future Development

This package currently retrieves raw JSON responses directly from the DeepSeek API. We are actively working on enhancing the library to parse this data and provide a user-friendly interface that mimics a real chat experience. Stay tuned for updates!

Contributions

Contributions, issues, and feature requests are welcome! Feel free to check the issues page.

Response Format

The library now provides structured response objects for different types of messages:

Message Response

{
    type: 'message',
    content: 'The message content',
    id: 'message-id',
    role: 'assistant',
    metadata: {
        model: 'model-name',
        usage: { /* token usage stats */ },
        finish_reason: 'stop'
    },
    timestamp: '2024-01-01T00:00:00.000Z'
}

Other Response Types

// Done Response
{
    type: 'done',
    timestamp: '2024-01-01T00:00:00.000Z'
}

// Error Response
{
    type: 'error',
    error: 'error_type',
    details: 'Error details',
    timestamp: '2024-01-01T00:00:00.000Z'
}

// Thinking State
{
    type: 'thinking',
    status: 'thinking_status',
    timestamp: '2024-01-01T00:00:00.000Z'
}

// Search Results
{
    type: 'search',
    results: [/* search results */],
    timestamp: '2024-01-01T00:00:00.000Z'
}

Enhanced Usage Example

Here's how to use the enhanced parsing features:

import { createNewChat, sendMessage } from 'deepseek-api';

(async () => {
    const token = 'your-token-here';
    const chatID = await createNewChat(token);

    if (typeof chatID === 'string') {
        await sendMessage('Write a JavaScript function', {
            id: chatID,
            token: token
        }, (response) => {
            switch(response.type) {
                case 'message':
                    console.log('Content:', response.content);
                    break;

                case 'thinking':
                    console.log('AI is thinking...');
                    break;

                case 'error':
                    console.error('Error:', response.error);
                    break;

                case 'done':
                    console.log('Response complete');
                    break;
            }
        });
    }
})();

Parsing Features

Code Block Parsing

• Automatically extracts code blocks from messages
• Identifies programming languages
• Separates code from regular text

Metadata

• Model information
• Token usage statistics
• Completion status
• Timestamps for all events

Response Types

• message: Regular message content
• thinking: AI processing status
• search: Search results when enabled
• error: Error information
• done: Completion indicator