Ai Data Extraction

AI Coding Assistant Training Data Extraction Toolkit

Complete toolkit to extract ALL chat, agent, and code context data from AI coding assistants for machine learning training.

🎯 What This Does

Automatically discovers and extracts **complete conversation history** including:

• ✅ User messages & AI responses
• ✅ Code context (file paths, line numbers, snippets)
• ✅ Code diffs and suggested edits
• ✅ Multi-file contexts
• ✅ Tool use and execution results
• ✅ Timestamps and metadata

📦 Included Scripts

1. `extract_claude_code.py`

Extracts from Claude Code / Claude Desktop

• Searches: ~/.claude, ~/.claude-code, ~/.claude-local, ~/.claude-m2, ~/.claude-zai
• Formats: JSONL session files
• Includes: Messages, tool use, file contexts, diffs

2. `extract_codex.py`

Extracts from Codex (if installed)

• Searches: ~/.codex, ~/.codex-local
• Formats: Rollout JSONL files
• Includes: User/agent messages, tool results, diffs

3. `extract_cursor.py`

Extracts from Cursor (Chat + Composer + Agent) - ALL VERSIONS

• Searches: ~/Library/Application Support/Cursor (macOS) or equivalent
• Formats: SQLite databases (state.vscdb, cursorDiskKV)
• Handles:
  • Old Chat mode (workspace storage)
  • Composer inline storage (v1.x - messages in composerData array)
  • Composer separate storage (v1.x-v2.0 transition - messages in bubbleId keys)
  • Latest Composer/Agent (v2.0+)
• Includes:
  • Code context, selections, diffs
  • Suggested edits and code blocks
  • Tool results and execution outputs

4. `extract_trae.py`

Extracts from Trae

• Searches: ~/.trae, ~/Library/Application Support/Trae
• Formats: JSONL and SQLite databases
• Includes: Chat, agent data, tool use, diffs

5. `extract_windsurf.py`

Extracts from Windsurf

• Searches: ~/Library/Application Support/Windsurf or equivalent
• Formats: SQLite databases (VSCode-like format)
• Includes: Chat, agent/flow conversation