Claude Parser v2.1.0 🚀

Git-like disaster recovery for Claude Code conversations

Claude Parser treats every Claude API call as a git commit, enabling powerful recovery and analysis capabilities when things go wrong.

🎉 What's New in v2.1.0

New Features

• 🔍 Export Domain - Export conversations to different formats for indexing
• 📚 LlamaIndex Export - export_for_llamaindex() for semantic search
• 🛠️ Fixed Discovery - discover_claude_files() now properly returns file paths
• 📦 Complete API - All filtering functions now properly exported

📋 v2.0.0 Major Changes

• 🎯 Complete API Redesign - Clean, intuitive Python API with 30+ functions
• 📚 15 Domain Architecture - Organized into focused, composable modules
• 🔧 CG Commands - Full Git-like CLI for disaster recovery
• 🪝 CH Hook System - Composable hooks for Claude Code integrations
• 📊 DuckDB Backend - Efficient JSONL querying without intermediate storage
• 🏗️ LNCA Compliant - Every file <80 LOC, 100% framework delegation
• 📖 Full Documentation - Complete API reference, user guides, and examples

Breaking Changes from v1

• New import structure: from claude_parser import load_session (not from claude_parser.main)
• Removed god objects - now uses focused domain modules
• All functions return plain dicts, not custom objects
• Pydantic schema normalization handles all JSONL variations

🚀 Quick Start

Installation

pip install claude-parser

Basic Usage

from claude_parser import load_latest_session, analyze_session

# Load your most recent Claude session
session = load_latest_session()
print(f"Found session with {len(session['messages'])} messages")

# Analyze the session
analysis = analyze_session(session)
print(f"Total tokens: {analysis['total_tokens']}")
print(f"Estimated cost: ${analysis['estimated_cost']:.2f}")

Disaster Recovery with CG Commands

# Oh no! Claude deleted important files!
python -m claude_parser.cli.cg find "important_file.py"

# Found it! Now restore it
python -m claude_parser.cli.cg checkout /path/to/important_file.py

# See what happened
python -m claude_parser.cli.cg reflog

# Jump back to before the disaster
python -m claude_parser.cli.cg reset abc123

🎯 Core Features

Git-like Commands (CG)

• cg status - Current session state
• cg log - Conversation history
• cg find <pattern> - Search across all sessions
• cg blame <file> - Who last modified a file
• cg checkout <file> - Restore deleted files
• cg reflog - Operation history
• cg show <uuid> - Operation details
• cg reset <uuid> - Time travel to any point

Composable Hooks (CH)

# Run hooks with custom executors
python -m claude_parser.cli.ch run --executor my_executor

# Or set default executor
export CLAUDE_HOOK_EXECUTOR=my_executor

Python SDK

from claude_parser import (
    # Session Management
    load_session, load_latest_session, discover_all_sessions,

    # Analytics
    analyze_session, analyze_tool_usage, analyze_token_usage,
    calculate_session_cost, calculate_context_window,

    # File Operations
    restore_file_content, generate_file_diff, compare_files,

    # Navigation
    find_message_by_uuid, get_message_sequence, get_timeline_summary,

    # Discovery
    discover_claude_files, discover_current_project_files,

    # Filtering (NEW in v2!)
    filter_messages_by_type, filter_messages_by_tool,
    search_messages_by_content, exclude_tool_operations,

    # Export (NEW in v2.1!)
    export_for_llamaindex  # Export conversations for semantic search
)

📊 Real-World Use Cases

1. Disaster Recovery

from claude_parser import load_latest_session, restore_file_content

session = load_latest_session()
content = restore_file_content("/deleted/important.py", session)
with open("recovered.py", "w") as f:
    f.write(content)

2. Cost Analysis

from claude_parser import load_latest_session, calculate_session_cost

session = load_latest_session()
cost = calculate_session_cost(
    input_tokens=100000,
    output_tokens=50000,
    model="claude-3-5-sonnet-20241022"
)
print(f"This session cost: ${cost['total_cost']:.2f}")

3. Message Filtering

from claude_parser import load_latest_session, MessageType
from claude_parser.filtering import filter_messages_by_type

session = load_latest_session()
user_messages = filter_messages_by_type(session['messages'], MessageType.USER)
print(f"You sent {len(user_messages)} messages")

4. Real-time Monitoring

from claude_parser.watch import watch

def on_assistant(message):
    print(f"Claude says: {message['content'][:100]}...")

watch("~/.claude/projects/current/session.jsonl", on_assistant=on_assistant)

5. Export for Semantic Search (NEW in v2.1!)

from claude_parser import export_for_llamaindex

# Export conversations to LlamaIndex format
docs = export_for_llamaindex("session.jsonl")
# Returns: [{"text": "message", "metadata": {...}}, ...]

# Use with semantic search services
for doc in docs:
    print(f"Text: {doc['text'][:50]}...")
    print(f"Speaker: {doc['metadata']['speaker']}")

🏗️ Architecture

Clean Domain Organization (19 modules)

claude_parser/
├── analytics/     # Session and tool analysis
├── api/           # API utilities
├── cli/           # CG and CH commands
├── core/          # Core utilities
├── discovery/     # File and project discovery
├── export/        # Export formats (NEW in v2.1!)
├── extensions/    # Extension system
├── filtering/     # Message filtering
├── hooks/         # Hook system and API
├── loaders/       # Session loading
├── messages/      # Message utilities
├── models/        # Data models
├── navigation/    # Timeline and UUID navigation
├── operations/    # File operations
├── queries/       # DuckDB SQL queries
├── session/       # Session management
├── storage/       # DuckDB engine
├── tokens/        # Token counting and billing
└── watch/         # Real-time monitoring

LNCA Principles

• <80 LOC per file - Optimized for LLM comprehension
• 100% Framework Delegation - No custom loops or error handling
• Single Source of Truth - One file per feature
• Pydantic Everything - Schema normalization for all JSONL variations

📖 Documentation

Full documentation available at: https://alicoding.github.io/claude-parser/

• Getting Started Guide
• CLI Commands Reference
• API Reference
• Architecture Overview

🔄 Migration from v1

Old v1 Code

# v1 - Complex imports, god objects
from claude_parser.main import ClaudeParser
parser = ClaudeParser()
parser.load_and_analyze_everything()  # God object doing too much

New v2 Code

# v2 - Clean, focused functions
from claude_parser import load_latest_session, analyze_session
session = load_latest_session()
analysis = analyze_session(session)  # One function, one purpose

Key Differences

1. No more god objects - Each function does one thing well
2. Plain dicts everywhere - No custom classes to learn
3. Explicit imports - Import only what you need
4. Better error handling - Framework delegation (Pydantic/Typer)
5. More features - Filtering, watching, complete hook system

🚢 Deployment

PyPI Release

# Build and upload to PyPI
python -m build
twine upload dist/*

Documentation

Documentation auto-deploys to GitHub Pages on every push to main.

🗺️ Export Format Roadmap

Currently Available (v2.1)

• ✅ LlamaIndex - export_for_llamaindex() - For semantic search indexing

Planned Export Formats

• 🔜 Mem0 - Long-term memory for AI agents
• 🔜 ChromaDB - Vector database format
• 🔜 Pinecone - Cloud vector database
• 🔜 Markdown - Human-readable conversation logs
• 🔜 JSON-LD - Structured data with context
• 🔜 OpenAI Messages - Direct OpenAI API format
• 🔜 Anthropic Messages - Direct Anthropic API format
• 🔜 LangChain Documents - LangChain document format
• 🔜 Haystack Documents - Haystack NLP framework

Export Domain Architecture

claude_parser/export/
├── __init__.py        # Export registry
├── llamaindex.py      # LlamaIndex format (DONE)
├── mem0.py           # Mem0 format (TODO)
├── chroma.py         # ChromaDB format (TODO)
├── markdown.py       # Markdown format (TODO)
└── ...               # More formats

🚀 Roadmap

v3.0.0 - UI-Ready API (Coming Soon)

Complete redesign for zero-boilerplate UI development.

Core/Feature Layer Separation

• ✅ Audit complete - identified all internal vs public functions
• ✅ Framework delegation mapped (humanize, babel, arrow, rich)
• 🔜 claude_parser.core - Low-level utilities for advanced users
• 🔜 claude_parser - High-level UI-ready functions

Display-Ready Functions

# Coming in v3.0.0
from claude_parser import (
    get_session_summary,         # "436 messages, 3 hours, $12.45"
    get_formatted_messages,       # Markdown-formatted conversation
    get_token_breakdown,          # "45,678 tokens ($12.45)"
    get_file_changes_display,     # Formatted diff with colors
    export_as_html,              # Complete HTML report
)

# One-liner, zero parsing needed:
print(get_session_summary(session))  # That's it!

Planned Features

• Session Display: Pre-formatted messages with timestamps, roles, emojis
• Analytics Dashboard: Human-readable metrics (not raw numbers)
• File Operations: Formatted diffs, file lists with status icons
• Export Formats: HTML, Markdown, JSON, PDF - all display-ready
• Smart Defaults: "No messages found" instead of empty arrays
• Number Formatting: "$12.45" not 0.01245, "45,678" not 45678
• Time Formatting: "2:34 PM" not timestamps, "3 hours ago" not seconds

Framework Delegation

All formatting delegated to specialized libraries:

• humanize - Number and size formatting
• babel - Currency formatting
• arrow - Time and date formatting
• rich - Terminal colors and tables
• emoji - Status indicators (✅ ❌ ⚠️)
• tabulate - Markdown/HTML tables
• jinja2 - HTML report generation

v2.2.0 - Bug Fixes (Next Release)

• ✅ Fixed token counting to match UI (v2.1.1)
• ✅ Fixed None message field handling
• 🔜 Additional message extraction improvements

🤝 Contributing

We welcome contributions! Please ensure:

• Files stay under 80 lines of code
• Use framework delegation (no custom loops)
• Add tests for new features
• Update documentation

📜 License

MIT License - See LICENSE file for details.

🙏 Acknowledgments

• Built for the Claude Code community
• Inspired by git's powerful recovery capabilities
• Designed with LNCA principles for LLM-native development

📊 Stats

• 19 specialized domains
• 35+ public functions
• <80 lines per file
• 100% framework delegation
• 0 custom error handling

---

Ready to never lose code again? Install v2.1.0 and experience the power of Git-like recovery for Claude Code!

pip install claude-parser==2.1.0

Documentation | GitHub | PyPI