Surf.new

Open-source playground for AI agents that browse and automate web tasks like a human.

4.7 (6)

İnceleyen Daniel Nikulshyn·Güncellendi Mayıs 2026

Self-Hostable Web Scraping Research Open Source AI Agents Playground Developer Tools Browser Automation

Genel Bakış

Surf.new is an experimental platform from Steel that lets users test and interact with AI agents capable of navigating websites, filling forms, and completing multi-step web tasks autonomously. It serves as a sandbox where developers and curious users can see how modern browsing agents behave in real-world scenarios. The tool connects language models to a controllable browser environment, allowing agents to click, scroll, type, and reason about page content. Because the project is open source, developers can inspect the underlying framework, customize agent behavior, and integrate similar capabilities into their own products. It is best suited for prototyping, research, and exploring the practical limits of agentic web automation rather than as a polished end-user productivity app.

Temel özellikler

AI agents that control a real browser
Natural language task instructions
Multi-step web navigation and form filling
Open-source and self-hostable
Integrates with various LLM providers
Built on the Steel browser infrastructure

Kullanım senaryoları

Prototype browsing agents

Developers can quickly test how AI agents navigate websites, click elements, and fill forms before integrating similar capabilities into their own applications.

Research web automation behavior

Researchers can observe how different LLMs reason about page content and handle multi-step web tasks in a controlled sandbox environment.

Self-host a custom agent stack

Teams can fork the open-source codebase, customize agent frameworks, and deploy a private browsing-agent infrastructure built on Steel's browser.

Compare LLM providers on web tasks

Users can swap in different LLM providers within the playground to benchmark how each model performs on real browsing and automation scenarios.

Artılar ve eksiler

Artılar

Free and open-source codebase
Live playground to test browsing agents
Supports multiple agent frameworks and models
Useful for developers researching web automation

Eksiler

Experimental and prone to errors on complex sites
Limited polish compared to commercial tools
Requires technical know-how for self-hosting
Performance depends on the underlying LLM

İncelemeler

4.7

6 puandan ortalama.

İnceleme bırakmak için giriş yap.

Joanna Kowalski

Use it every day

Honestly didn't expect to like it this much. AI agents that control a real browser is exactly what I needed, and supports multiple agent frameworks and models. I do wish experimental and prone to errors on complex sites, but I reach for it almost every day now and it just clicks.

Pierre Dubois

Years in this space

I've evaluated a lot of these over the years. What stands out here is natural language task instructions — handled better than most — and supports multiple agent frameworks and models. Worth the time if this is your use case.

Robert Ainsworth

Compared a few options

Evaluated this against two competitors. Where it wins: natural language task instructions and free and open-source codebase. Where it lags: requires technical know-how for self-hosting. On balance the feature set — especially built on the Steel browser infrastructure — justifies the 5 stars for our use case.

Olga Ivanova

Use it every day

Honestly didn't expect to like it this much. Multi-step web navigation and form filling is exactly what I needed, and useful for developers researching web automation. I do wish limited polish compared to commercial tools, but I reach for it almost every day now and it just clicks.

Daniel Schmidt

Use it every day

Honestly didn't expect to like it this much. Integrates with various LLM providers is exactly what I needed, and supports multiple agent frameworks and models. but I reach for it almost every day now and it just clicks.

Ingrid Bauer

Compared a few options

Evaluated this against two competitors. Where it wins: natural language task instructions and live playground to test browsing agents. Where it lags: experimental and prone to errors on complex sites. On balance the feature set — especially multi-step web navigation and form filling — justifies the 4 stars for our use case.