# AI Platform Export Format Research
This document analyzes export formats from different AI platforms to design universal import support.
## Current Claude Memory Format
```json
{
"id": "conv_20250613_120000_1234",
"title": "Example Conversation",
"content": "Human: Hello\n\nClaude: Hi there!",
"date": "2025-06-13T12:00:00",
"topics": ["greeting", "conversation"],
"created_at": "2025-06-13T12:00:00"
}
```
## Platform Format Analysis
### 1. ChatGPT/OpenAI Format
- **Export Method**: Settings > Data Export > Export Data
- **Format**: JSON with message arrays
- **Key Fields**: conversations, messages, role (user/assistant)
- **Metadata**: creation_time, update_time, conversation_id
**Typical Structure:**
```json
{
"conversations": [
{
"id": "conversation-uuid",
"title": "Conversation Title",
"create_time": "2025-01-01T12:00:00",
"update_time": "2025-01-01T12:30:00",
"messages": [
{
"id": "message-uuid",
"role": "user",
"content": "Hello",
"create_time": "2025-01-01T12:00:00"
},
{
"id": "message-uuid",
"role": "assistant",
"content": "Hi there!",
"create_time": "2025-01-01T12:00:30"
}
]
}
]
}
```
### 2. Cursor AI Format
- **Export Method**: Currently limited export options
- **Format**: Session-based with code context
- **Key Fields**: session_id, files, commands, ai_responses
- **Metadata**: workspace_path, timestamp, model_used
**Typical Structure:**
```json
{
"session_id": "session-uuid",
"workspace": "/path/to/project",
"timestamp": "2025-01-01T12:00:00",
"model": "claude-3.5-sonnet",
"interactions": [
{
"type": "user_input",
"content": "Fix this bug",
"files": ["src/main.py"],
"timestamp": "2025-01-01T12:00:00"
},
{
"type": "ai_response",
"content": "Here's the fix...",
"changes": [{"file": "src/main.py", "diff": "..."}],
"timestamp": "2025-01-01T12:00:30"
}
]
}
```
### 3. Claude Web Interface Format
- **Export Method**: Manual copy/paste or browser automation
- **Format**: Markdown-style conversation
- **Key Fields**: Role indicators, timestamps
- **Metadata**: Limited - mainly timestamp and conversation flow
**Typical Structure:**
```markdown
# Conversation with Claude
**Human**: Hello there!
**Claude**: Hi! How can I help you today?
**Human**: I need help with Python
**Claude**: I'd be happy to help with Python! What specifically are you working on?
```
### 4. Other Platforms (Preliminary)
#### Windsurf AI
- **Status**: Emerging platform
- **Format**: TBD - likely similar to Cursor with enhanced code context
#### Perplexity AI
- **Export**: Limited
- **Format**: Search-focused with sources
#### Claude Desktop (MCP)
- **Format**: Same as web but with enhanced metadata
- **Advantage**: Direct API access for real-time sync
## Universal Format Requirements
Based on analysis, our universal format needs:
1. **Platform Identification**: Source platform and model
2. **Flexible Message Structure**: Support both conversation and session-based formats
3. **Rich Metadata**: Timestamps, file context, model info
4. **Content Preservation**: Maintain original formatting and context
5. **Extensibility**: Allow platform-specific fields
## Proposed Universal Internal Format
```json
{
"id": "universal-conversation-id",
"platform": "chatgpt|cursor|claude|generic",
"model": "gpt-4|claude-3.5-sonnet|cursor-ai",
"title": "Generated or Extracted Title",
"content": "Unified conversation content",
"messages": [
{
"id": "message-id",
"role": "user|assistant|system",
"content": "Message content",
"timestamp": "ISO8601",
"metadata": {
"files": ["optional file context"],
"changes": ["optional code changes"],
"platform_specific": {}
}
}
],
"date": "ISO8601 - conversation start",
"last_updated": "ISO8601 - last message",
"topics": ["extracted", "topics"],
"session_context": {
"workspace": "optional workspace path",
"project_type": "optional project context"
},
"import_metadata": {
"imported_at": "ISO8601",
"source_format": "original format",
"import_version": "1.0"
}
}
```
## Implementation Strategy
1. **Phase 1**: Format detection and ChatGPT import
2. **Phase 2**: Cursor and code-focused imports
3. **Phase 3**: Generic markdown/text import
4. **Phase 4**: Real-time platform integrations
*Last Updated: June 13, 2025*
