⚡ Dipole

A small demo of an AI agent that deploys a web project to Netlify or Vercel from a chat prompt.

Tell the agent where your project is, and it analyzes the project type, picks a deployment target, runs the deploy, and streams the logs back — with a mock mode so you can watch the whole flow without touching a real provider.

What it demonstrates

• A Node CLI agent (agent/) that analyzes a project, decides between Netlify and Vercel, deploys via the provider CLI or API, and streams onLog chunks.
• A Streamlit chat UI (demo/) that drives the CLI over subprocess: plan → deploy → diagnose → tail logs, with structured JSON between the two layers.
• A mock mode (FAST_DEPLOY_MOCK=success|fail|rate_limit) for demoing the agent loop with no real deployment or API key.
• A validation harness (validation/) with fixture apps (static, Vite/React, Next.js) and JSON-only output checks.

How to run

Prerequisites: Python 3.10+, Node.js 18+ (the Streamlit app shells out to node agent/cli/index.js), and an OpenAI API key.

git clone https://github.com/zelinewang/dipole.git
cd dipole

# Python UI + deps
python3 -m venv .venv && source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r demo/requirements.txt

# Configure (required)
export OPENAI_API_KEY=sk-...          # agent reasoning
export FAST_DEPLOY_MOCK=success       # optional: dry-run without a real deploy

streamlit run demo/streamlit_app.py   # opens http://localhost:8501

The Node CLI runs on Node built-ins (dotenv is loaded only if present), so there is no npm install step. For real deploys, install the provider CLI (netlify-cli or vercel) and set NETLIFY_AUTH_TOKEN / VERCEL_TOKEN — either in the environment, a root .env, or the in-app Secrets panel.

Validate locally

Run all four credential-free component validations with one command:

bash validation/run_validation.sh

The script stops at the first failure and covers fixture detection, analyzer behavior, JSON-only CLI output, and mocked Netlify/Vercel deploy flows. It does not perform a live provider deployment.

Typical flow

1. "My project path is /absolute/path/to/project" → the agent runs plan.
2. "Use Netlify CLI" (or Vercel) → updates the deployment preference.
3. "Deploy now" → live logs stream in; on success a URL appears and previews in an iframe.
4. "Diagnose record <id>" / "Tail logs for <id>" → inspect a past run.

What this is not

• Not a hosted service. The landing page is static documentation; install and launch the working Streamlit UI on your own machine.
• Not production-hardened. This is a ~7-commit demo of the agent loop, not a maintained tool.
• Not a replacement for the Netlify/Vercel CLIs — it orchestrates them.
• Detection-limited. It handles the common front-end project types it can recognize (static, React/Vite, Next.js, and similar); anything unusual may need manual setup.

Tech

Node.js agent (standard library + optional dotenv) · Streamlit + LangChain chat UI · OpenAI models · Netlify / Vercel deploy CLIs.

License

MIT.

Acknowledgments

Built with Streamlit, LangChain, and OpenAI; deploys via Netlify and Vercel.