🪩 VibeCoding Dataset Project

Collecting the vibes of coding — one log at a time.

---

📢 Call for Volunteers

We’re building an open dataset to capture real-world coding interactions between developers and AI coding assistants — and we need your help!

This dataset will help researchers and developers better understand how humans and code models interact across different tools, and improve the future of AI-assisted software development.

---

🎯 Project Overview

The VibeCoding Dataset aims to collect anonymized client ↔ server message logs from popular AI coding tools and interfaces. These logs will form the basis of an open dataset hosted on Hugging Face and GitHub.

• Hugging Face: https://huggingface.co/datasets/QuixiAI/VibeCoding
• GitHub: https://github.com/QuixiAI/vibecoding

---

🧰 Tools We’re Targeting

We’re collecting interaction logs from the following coding assistants and CLIs:

• Claude Code
• OpenAI Codex
• Gemini CLI
• Open-Code
• Cline
• Roo Code
• Continue.dev
• Cursor
• Windsurf
• Goose
• OpenHands
• Aider
• Factory Droid CLI
• charmbracelet/crush

If you regularly use any of these — you’re exactly who we need!

---

🧩 What You’ll Do

1. Set up a logging proxy Use a lightweight tool like LiteLLM or Dolphin Logger to capture your coding assistant’s request/response data.

2. Record your sessions As you use your AI coding tool normally, your proxy will record the message logs between your client and the model API.

3. Anonymize and submit Before submission, make sure your logs contain no private or sensitive information. See our Data Cleaning Guide (coming soon).

4. Contribute your data Submit your anonymized logs via:

   • Pull request to this repo, or
   • Upload through the Hugging Face dataset page

---

🤝 Contributing

Here’s how to get started capturing logs safely and easily.

Option 1 — Using LiteLLM

LiteLLM is a drop-in proxy for OpenAI-compatible APIs.

1️⃣ Install LiteLLM

pip install litellm

2️⃣ Run a local proxy

litellm --port 4000 --log --log_file logs/vibecoding.jsonl

3️⃣ Configure your tool

Change your AI coding tool or CLI to point to:

OPENAI_API_BASE=http://localhost:4000

Keep your normal API key set as usual.

4️⃣ Use your tool normally

LiteLLM will log all incoming/outgoing messages in logs/vibecoding.jsonl.

---

Option 2 — Using Dolphin Logger

Dolphin Logger provides an intercepting proxy that records JSON message streams.

1️⃣ Install

git clone https://github.com/yoheinakajima/dolphin-logger.git
cd dolphin-logger
npm install

2️⃣ Start the logger

npm start

By default, this runs on http://localhost:3000.

3️⃣ Set your proxy

Point your coding assistant’s API endpoint or environment variable to:

HTTP_PROXY=http://localhost:3000

4️⃣ Collect logs

Your logs will appear in the logs/ directory as JSON files.

---

🧼 Clean Your Logs

Before submission, please remove or redact:

• Any personal identifiers (e.g., email, usernames)
• Proprietary or confidential code
• Project names or unique file paths

You can anonymize text manually or use our upcoming sanitize_logs.py script.

---

📨 Submit Your Data

When your logs are ready:

1. Fork this repository
2. Create a folder under submissions/<your_handle>/
3. Add your cleaned .json or .jsonl logs
4. Open a pull request

Alternatively, you can upload them directly to our Hugging Face dataset.

---

🫶 Credits & Acknowledgements

All volunteers who contribute cleaned and usable logs will be credited by name or handle in:

• The dataset release notes
• The model card
• The GitHub contributors section

We appreciate your help in making open-source AI more transparent and human-centered!

---

💬 Get Involved

Join the discussion in our dedicated channel: 👉 #vibecoding-dataset-project

Ask questions, share your setup, or get help with proxy configuration.

---

⚖️ License & Ethics

This project follows the principles of open, ethical data collection:

• No private or proprietary data
• No identifying information
• Only voluntary, informed contributions

Dataset licensed under Apache 2.0.

---

🚀 Quick Links

• 🧠 Dataset: Hugging Face – QuixiAI/VibeCoding
• 💻 Code & Instructions: GitHub – QuixiAI/vibecoding
• 💬 Discussion: #vibecoding-dataset-project

---

Help us capture the rhythm of coding — one conversation at a time.