Usage

Start the development server

After completing the install steps, start both services with a single command:

# macOS / Linux
npm run dev

# Windows (required due to asyncio event loop differences)
npm run dev:win

This starts:

• Next.js frontend on http://localhost:3001
• FastAPI backend on http://localhost:8000

Open http://localhost:3001 in your browser.

Running an agent task

1. Select an AI provider (Claude, GPT-4, Gemini, Fireworks, Ollama, etc.) from the dropdown.

2. Type a natural-language instruction, for example:

   "Go to news.ycombinator.com and tell me the top 5 headlines."

3. Click Run. The agent takes control of a Steel cloud browser, navigating and interacting with the page in real time. A live browser-stream panel shows what the agent sees.

4. The agent returns a result and a full action trace when it finishes.

Architecture at a glance

Browser (localhost:3001)  -->  Next.js  -->  FastAPI (localhost:8000)
                                                   |
                                           Langchain agent
                                                   |
                                        Steel cloud browser session
                                                   |
                                          Target website

The FastAPI backend manages the Steel session lifecycle and streams agent steps back to the frontend via Server-Sent Events.

Supported AI providers (as of initial release)

Provider	Models
Anthropic	Claude 3 Opus / Sonnet / Haiku
OpenAI	GPT-4o, GPT-4 Turbo
Google	Gemini 1.5 Pro / Flash
Fireworks	Llama, Mixtral and others
Ollama	Any locally running model