vibedev

cargo install vibedev

Analyze your AI coding assistant usage patterns, costs, and productivity.

Features

• Multi-Tool Support - Claude Code, Cursor, Cline, Kiro, Copilot, Windsurf, and 15+ more
• Interactive TUI - Real-time visualization of your AI usage (like dust for logs)
• Embedded LLM - Offline AI chat to analyze your data (no API keys needed)
• Deep Insights - Usage patterns, cost estimates, productivity metrics
• Dataset Export - Create sanitized training datasets from your conversations
• Backup & Restore - Compress and archive your AI learning data

Quick Start

# Install
cargo install --path .

# Discover AI tool logs on your system
vibedev discover

# Interactive TUI dashboard
vibedev tui

# Generate analysis report
vibedev analyze --format html --output report.html

Commands

| Command | Description | |---------|-------------| | discover | Scan system for AI tool logs | | analyze | Generate analysis reports (text/json/html/markdown) | | tui | Interactive terminal dashboard | | chat | Chat with your data using embedded LLM | | daemon | Keep LLM loaded for fast queries | | models | Manage offline LLM models | | claude | Manage Claude Code provider configuration | | backup | Create compressed backup archive | | prepare | Export sanitized training dataset | | insights | Comprehensive HTML dashboard | | compare | Compare usage across tools |

Claude Provider Management

Switch between different API providers for Claude Code:

# List all supported providers
vibedev claude list

# Show current configuration
vibedev claude show

# Set a provider (saves configuration)
vibedev claude set z.ai --api-key sk-xxx
vibedev claude set openrouter --api-key sk-or-xxx
vibedev claude set chatgpt --api-key sk-xxx
vibedev claude set litellm --api-key sk-xxx

# Apply configuration to Claude Code (updates config files)
vibedev claude set openrouter --api-key sk-or-xxx --apply

# Custom provider with custom endpoint
vibedev claude set custom --endpoint https://api.example.com/v1 --model gpt-4 --api-key xxx --apply

Supported Providers:

• z.ai - High-performance Claude API proxy
• openrouter - Unified LLM API with multiple models
• chatgpt - OpenAI GPT-4 and other OpenAI models
• litellm - Local proxy for multiple providers
• custom - Your own custom endpoint

Embedded LLM (Offline)

Analyze your data with a fully offline LLM - no API keys or internet required:

# Download a model (one-time)
vibedev models download qwen-coder-1.5b

# Chat with your data
vibedev chat --query "What are my top productivity patterns?"

# Keep model loaded for fast queries
vibedev daemon start
vibedev chat --query "How can I reduce costs?"
vibedev daemon stop

Available Models:

• qwen-coder-0.5b - Fast, 1GB
• qwen-coder-1.5b - Balanced, 3GB (recommended)
• qwen-coder-3b - High quality, 6GB
• deepseek-coder-1.3b - Efficient, 2.6GB

GPU Acceleration:

cargo build --release --features cuda   # NVIDIA
cargo build --release --features metal  # Apple Silicon

Interactive TUI

vibedev tui

Real-time visualization inspired by dust:

• Storage breakdown by tool
• Session counts and token estimates
• Cost analysis
• Keyboard navigation (↑↓ to navigate, Enter to expand, q to quit)

Analysis Report

vibedev analyze --output report.md

Sample output:

# AI Coding Tools - Analysis Report

## Global Metrics
- Total Storage: 1.01 GB
- Total Sessions: 1,020
- Estimated Tokens: 21,245,300
- Peak Usage Hour: 14:00 UTC
- Most Used Tool: Claude Code

## Cost Estimate
- Monthly Cost: $165.71
- Optimization Potential: $49.71 (30%)

## Recommendations
1. Backup debug logs (345 MB compressible)
2. Clean file-history older than 30 days
3. Consider token caching for repeated queries

Supported Tools

AI Coding Assistants:

• Claude Code, Cursor, Windsurf, Cline, Continue.dev
• GitHub Copilot, Sourcegraph Cody, Amazon Q
• Kiro, Roo Code, Kilo, Aider

Code Completion:

• Tabnine, Supermaven, CodeWhisperer, CodeGPT, Bito AI

Logs detected from:

• ~/.claude/ - Claude Code
• ~/.cursor/ - Cursor
• ~/.config/Code/User/globalStorage/ - VSCode extensions
• ~/.config/Kiro/ - Kiro
• And 80+ more locations

Dataset Export

Create sanitized datasets for fine-tuning:

vibedev prepare --output ~/datasets

Sanitization removes:

• API keys (OpenAI, Anthropic, GitHub, AWS, etc.)
• Passwords and credentials
• Email addresses and phone numbers
• Personal file paths
• IP addresses

See PREPARE_GUIDE.md for details.

Build from Source

git clone https://github.com/openSVM/vibedev.git
cd vibedev
cargo build --release
./target/release/vibedev --help

Requirements:

• Rust 1.82+
• ~8MB binary size

Configuration

vibedev works out of the box with no configuration. Optional settings:

# Filter to specific tool
vibedev analyze --tool claude

# Time range
vibedev analyze --days 30

# Output format
vibedev analyze --format json --output data.json

Contributing

Contributions welcome! Please read the code of conduct and submit PRs.

# Run tests
cargo test

# Run with debug logging
vibedev --verbose analyze

License

MIT License - see LICENSE for details.

Built with Rust. Inspired by the AI coding revolution.