Skip to content

group_chat_format.md

Latest commit

 

History

History
289 lines (226 loc) · 6.95 KB

group_chat_format.md

File metadata and controls

289 lines (226 loc) · 6.95 KB

Group Chat Format Specification

Overview

This is an open-source group chat data format specification for standardizing the storage and exchange of group chat conversation data.

Format Definition

For the complete format definition, please refer to group_chat_format.py

Core Features

1. Separated Metadata and Message List

{
  "version": "1.0.0",
  "conversation_meta": { ... },
  "conversation_list": [ ... ]
}
  • version: Format version number (follows semantic versioning)
  • conversation_meta: Conversation metadata
  • conversation_list: Message list

2. Scene Types and Scene Descriptions

Two core conversation scenes are supported:

  • assistant: Human-AI assistant conversation scene, one-on-one dialogue, AI acts as a personal assistant
  • group_chat: Work group chat scene, multi-person group chat, team collaboration

Assistant Scene Example:

"conversation_meta": {
  "scene": "assistant",
  "scene_desc": {
    "description": "User and AI assistant conversation about Beijing tourism, health management, and sports rehabilitation"
  },
  ...
}

Group Chat Scene Example:

"conversation_meta": {
  "scene": "group_chat",
  "scene_desc": {},
  ...
}
  • scene: Scene type identifier (assistant or group_chat)
  • scene_desc: Scene description information
    • Assistant scene: Contains description field, describing the conversation scene
    • Group chat scene: Usually an empty object, representing a multi-person collaboration scene

3. User Details

All user details are centrally stored in conversation_meta.user_details:

"user_details": {
  "user_101": {
    "full_name": "Alex",
    "role": "user",
    "custom_role": "Tech Lead",
    "department": "Technology",
    "email": "alex@example.com"
  },
  "robot_001": {
    "full_name": "AI Assistant",
    "role": "assistant"
  }
}

Field Descriptions:

  • full_name: User's display name (optional)
  • role: User type role (user for human, assistant for AI) (optional)
  • custom_role: User's job/position role (e.g., Tech Lead, Product Manager) (optional)
  • department: Department (optional)
  • email: Email address (optional)
  • extra: Additional extended information (optional)

4. Message Structure

Each message uses user ID as sender, with an optional sender_name for readability:

{
  "message_id": "msg_001",
  "create_time": "2025-02-01T10:00:00+00:00",
  "sender": "user_103",
  "sender_name": "Chen",
  "role": "user",
  "type": "text",
  "content": "Message content",
  "refer_list": []
}

4.1 Message Sender Role

The optional role field identifies the source of the message:

  • user: Message from a human user
  • assistant: Message from an AI assistant

This is compatible with OpenAI/mem0/memos message format.

Human Message Example:

{
  "message_id": "msg_001",
  "sender": "user_101",
  "sender_name": "Alex",
  "role": "user",
  "type": "text",
  "content": "Can you help me summarize this document?"
}

AI Response Example:

{
  "message_id": "msg_002",
  "sender": "robot_001",
  "sender_name": "AI Assistant",
  "role": "assistant",
  "type": "text",
  "content": "Here's the summary of the document..."
}

5. Timezone-Aware Timestamps

  • Uses ISO 8601 format
  • Recommended to include timezone information (e.g., +00:00)
  • If a message doesn't have timezone information, it can be obtained from conversation_meta.default_timezone

6. Message Types

Multiple message types are supported:

  • text: Text message
  • image: Image message
  • file: File message
  • audio: Audio message
  • video: Video message
  • link: Link message
  • system: System message

7. Message References

Flexible reference methods are supported, each element in refer_list can be:

Method 1: String Reference (message_id only)

"refer_list": ["msg_002", "msg_005"]

Method 2: Object Reference (only message_id is required, other fields are optional)

Minimal form:

"refer_list": [
  {
    "message_id": "msg_002"
  }
]

Including partial fields (e.g., content for preview):

"refer_list": [
  {
    "message_id": "msg_002",
    "content": "Good morning. Let's align on the goal first. Is it an MVP for internal testing or direct customer pilot?"
  }
]

Including complete information:

"refer_list": [
  {
    "message_id": "msg_002",
    "create_time": "2025-02-01T10:01:00+00:00",
    "sender": "user_102",
    "sender_name": "Betty",
    "type": "text",
    "content": "Good morning. Let's align on the goal first. Is it an MVP for internal testing or direct customer pilot?",
    "refer_list": []
  }
]

Method 3: Mixed Usage

"refer_list": [
  "msg_001",
  {
    "message_id": "msg_002",
    "content": "Partial content..."
  }
]

Usage Recommendations:

  • String reference: Most concise, suitable for scenarios with clear reference relationships
  • Minimal object reference: Only message_id, suitable for scenarios requiring unified format but no additional information
  • With preview content: Includes message_id + content, suitable for quick preview scenarios
  • Complete object reference: Includes all fields, suitable for export, archiving, or use independent of original data

8. Extension Fields

Use extra field to store additional information:

"extra": {
  "file_name": "UI_draft_v1.pdf",
  "file_size": 2048576,
  "file_type": "application/pdf"
}

Example Files

Usage

Python

from data.group_chat_format import GroupChatFormat, validate_group_chat_format
import json

# Read group chat data
with open('group_chat_example.json', 'r', encoding='utf-8') as f:
    data = json.load(f)

# Validate format
is_valid = validate_group_chat_format(data)
print(f"Format validation: {'Passed' if is_valid else 'Failed'}")

# Access user information
user_details = data['conversation_meta']['user_details']
print(f"User list: {list(user_details.keys())}")

# Iterate through messages
for msg in data['conversation_list']:
    sender_info = user_details[msg['sender']]
    print(f"{sender_info['full_name']}: {msg['content']}")

Creating New Group Chat Data

from data.group_chat_format import create_example_group_chat
import json

# Create example data
chat_data = create_example_group_chat()

# Save to file
with open('my_chat.json', 'w', encoding='utf-8') as f:
    json.dump(chat_data, f, ensure_ascii=False, indent=2)

Version History

  • 1.0.0 (2025-02-01)
    • Initial version
    • Support for basic message types
    • Support for user details
    • Support for message references
    • Support for timezone-aware timestamps

Contributing

Issues and Pull Requests are welcome to improve this format specification.

License

Open source license to be determined