Anything API (Notte): Turn any website into an API, the "Full-Stack" solution for Browser Agents

2026-03-05 | ProductHunt | Official Site | GitHub

Interface Breakdown: Notte's Studio interface—a visual orchestration platform for browser agents. On the left is the task description input; on the right is the real-time browser preview and operation log. Developers can "watch the agent work" and then convert it to production code with one click.

30-Second Quick Judgment

What does this app do?: You tell it "help me do something on a certain website," and it sends an AI agent to open a browser and do it. Once finished, it packages the entire process into a callable API. Simply put, for any website without an API, Notte "builds" one for you.

Is it worth watching?: Yes. YC S25 incubated, $2.5M pre-seed, 1.8k GitHub stars. It solves the "what to do when a site has no API" problem that has plagued developers for over a decade. The hybrid architecture (Script + AI) is much more reliable than pure AI agents, and the pricing is friendly (100 free hours). If you frequently need to scrape data or automate tasks across various websites, this tool will save you massive amounts of time writing and maintaining crawlers.

Three Questions for Me

Is it relevant to me?

Target users: Primarily developers and technical product teams. Specifically:

• People who need to scrape data from websites without APIs.
• People tired of maintaining a pile of Selenium/Playwright scripts that break with every site update.
• People who want AI agents to "work" on web pages (filling forms, placing orders, monitoring prices).

Am I the target?: If you spend more than 2 hours a week writing or maintaining browser automation scripts, you are. If you only occasionally copy-paste data manually, you might not need it yet.

When would I use it?:

• Competitor price monitoring—automatically scrape price changes every 6 hours.
• Job info aggregation—batch extract job listings from platforms like LinkedIn.
• Legacy system integration—old systems with only a web UI and no API; use Notte as a bridge.
• You don't need this if—the site you're using already has a mature REST API.

Is it useful to me?

ROI Judgment: If you spend 10+ hours a month maintaining browser automation scripts, Notte can save you 80% of that time. The free tier is enough for an indie dev to experiment for months. It's worth a try.

Is it "Wow"?

The "Aha!" moments:

• Agent Mode → Script Conversion: Use natural language to let the agent explore and see how it operates. Once satisfied, convert it to a deterministic script with one click. The workflow is incredibly smooth.
• Session Profiles: Log in once, and all subsequent runs skip authentication automatically. A lifesaver for automation scenarios requiring login states.
• Hybrid Architecture: Use scripts for deterministic parts and AI for the uncertainties. It's both reliable and flexible—far better than the "Schrödinger's success rate" of pure AI agents.

The "Wow" moment:

"Switched from Playwright + LLM chaining to Notte for web agents. Ran 500+ e-commerce price checks without a single CAPTCHA block or ban." -- @ai_engineer42 (Twitter)

Real User Reviews:

Positive: "Notte browser agent handled a nasty multi-tab auth flow I couldn't get working with other frameworks. Success rate jumped from 40% to 95%." -- @dev_ml_guy (Twitter) Positive: "Quick demo of bootstrapping a Notte agent for LinkedIn job scraping...>92% reliability on dynamic sites." -- @lucasgiordano (Twitter, CTO's own post, take with a grain of salt) Neutral: "wow I'm testing this right now, got really high hopes for it" -- Slashdot community user

For Indie Hackers

Tech Stack

• Language/SDK: Python (notte-sdk on PyPI), TypeScript/Node.js support.
• Browser Engine: Compatible with Playwright, Puppeteer, Selenium via CDP connection.
• AI/Models: Model-agnostic; supports OpenAI, Anthropic, Gemini with no token markup.
• Infrastructure: Edge network deployment; compute runs co-located with the browser for zero latency.
• Security: SOC 2 Type II certified, encrypted credential storage.

Core Implementation

Notte's core innovation is the Perception Layer. Traditional browser automation interacts directly with the DOM, which is messy for LLMs to read. Notte adds a layer in between: it converts the webpage into a structured, natural-language "map." For the LLM, it's like reading a manual instead of source code.

This significantly reduces the LLM's cognitive load, making reasoning more accurate while consuming fewer tokens. Notte claims this design outperforms other browser agent solutions on all benchmarks.

The API provides three core endpoints: /observe (get page state and available actions), /step (execute an action on the page), and /scrape (structured data extraction). It also supports Pydantic models for structured output formats.

Open Source Status

• Is it open source?: Core is open source, 1.8k GitHub stars, 166 forks.
• License: SSPL-1.0 (Note: Not MIT/Apache; has restrictions on commercial use, similar to MongoDB's approach).
• Similar Open Source Projects: Browser Use (Pure AI solution, MIT license), Steel (Pure infrastructure solution).
• Difficulty of building it yourself: High. Building the Perception Layer + Hybrid Architecture + Stealth Browser + Credential Management would take an estimated 6-8 person-months. However, you can get running using the open-source version with your own LLM key.

Business Model

• Monetization: Usage-based billing (browser hours + proxy traffic) + Subscriptions (rate limits).
• Pricing: 100 free hours → starts at $20/month → $0.05/h pay-as-you-go.
• Open-core + Paid Cloud Service: Classic model (advanced features like stealth, personas, and vault are in the cloud version).

Big Tech Risk

Medium-High. Google has added Auto Browse to Chrome, ChatGPT has Atlas agent mode, and Perplexity has Comet. However, these are consumer-facing "browser assistants," whereas Notte positions itself as developer infrastructure. In the short term, giants are unlikely to build "turn websites into APIs" developer tools. Real competition comes from Browserbase (recently raised a large round) and Browser Use (active open-source community).

For Product Managers

Pain Point Analysis

• What problem does it solve?: Most websites lack public APIs. Programmatically interacting with them requires writing browser automation scripts, which are painful to write, expensive to maintain, and easily blocked by anti-bot measures.
• How painful is it?: High-frequency, core pain point. Any team doing data aggregation, competitor monitoring, or system integration faces this. Traditional solutions require rewriting scripts every time a site changes, where maintenance costs can swallow development costs.

User Persona

• Developers/Engineering Teams: Needing bulk data collection or automated operations.
• Growth Hackers: Needing to monitor competitors or scrape user feedback.
• Enterprise IT: Needing to interface with legacy systems that lack APIs.

Feature Breakdown

Competitive Differentiation

Takeaways

1. "Agent first, Script later" progressive experience: Let users see results with natural language first, then convert to reliable code. This lowers the barrier to entry while ensuring production reliability.
2. Perception Layer abstraction: Don't let the LLM read HTML directly; add a "translation" layer. This concept can be applied to any scenario where LLMs need to understand complex inputs.
3. Hybrid Architecture thinking: Don't strive for "pure AI solves everything." Let AI and scripts each do what they are best at. This is a valuable reference for any AI product.

For Tech Bloggers

Founder Story

• Andrea Pinto (CEO): MIT AI/ML researcher, Fulbright scholar, ETH Zurich + EPFL background, formerly at IBM Research and Amazon. Even started a physical clothing business at age 16-18. A classic "academic heavyweight enters entrepreneurship" path.
• Lucas Giordano (CTO): Master's in Communication Systems from EPFL, MLO lab alum, 4 years of experience at the intersection of ERP and AI.
• The two met at EPFL in 2019 and have collaborated for 6 years. Their core insight: The internet was built for humans, not AI agents; a "Perception Layer" is needed to bridge the gap.
• Switzerland + San Francisco combo: R&D in Switzerland, marketing in SF. YC S25 cohort, a lean team of 5.

Controversy/Discussion Angles

• The SSPL License Choice: Using SSPL instead of MIT (like MongoDB) shows they are protecting against "cloud provider freeloading." However, this might limit community contributions and commercial users. A great topic for open-source strategy discussions.
• The Ethics of "Turning Any Site into an API": This is essentially an advanced crawler platform. What about website TOS? How are legal gray areas handled?
• AI Agent Security Risks: Letting AI operate a browser opens the door to prompt injection on malicious pages, potentially hijacking the session for dangerous actions. Malwarebytes has already issued warnings on this.
• "Hybrid Architecture" vs. "Pure AI Agent": Will AI capabilities eventually make hybrid architectures redundant, or will production environments always require deterministic scripts?

Hype Data

• PH Ranking: 293 votes (above average).
• GitHub: 1.8k stars (good growth for a project founded in 2025).
• Twitter Engagement: The CTO's demo tweet got 312 likes—active, but not yet viral.
• Search Trends: Appearing in multiple "Best AI Browser Agents 2026" lists (Firecrawl, KDnuggets, Zapier, etc.).

Content Suggestions

• Angle: "Why do we still need crawlers in 2026? -- Notte's Hybrid Architecture Approach."
• Trend Jacking: Ride the "AI Agent" and "MCP" (Notte already has an MCP Server integration) waves.

For Early Adopters

Pricing Analysis

Hidden costs: If using your own LLM key, model fees are extra (though Notte claims no markup). Proxy traffic at $10/GB can add up if scraping large amounts of data.

Onboarding Guide

• Setup Time: 15-30 minutes.
• Learning Curve: Low (for developers with Python basics).
• Steps:
  1. pip install notte-sdk
  2. Set API key (get it from notte.cc).
  3. Write three lines of code: Initialize client → Describe task → Call run.
  4. Or use the Anything API Web UI to describe the task without writing code.

from notte import NotteClient
client = NotteClient(api_key="...")
result = client.agents.run(task="Monitor price of RTX 5090 on amazon.com every 6h", schedule="6h")

Gotchas and Gripes

1. Inherent Fragility of Browser Automation: Site redesigns, dynamic loading, and A/B testing can still break automation. Notte's hybrid architecture mitigates this but doesn't solve it fundamentally.
2. SSPL License: If you want to embed Notte's core code into your own SaaS product, read the terms carefully. It's not as free as MIT/Apache.
3. Startup Stability: Founded in 2025, long-term stability is yet to be proven. Have a Plan B if your business-critical workflows depend on it.

Security and Privacy

• SOC 2 Type II Certified: Enterprise-grade security compliance.
• Credential Management: Encrypted storage, fetched at runtime, never hardcoded.
• Data Storage: Browser session data is processed in the cloud.
• Risk Point: Your website passwords and cookies pass through Notte's servers; you must trust the third party.

Alternatives

For Investors

Market Analysis

• Sector Size: The Agentic Browser market was $4.5B in 2024, projected to reach $76.8B by 2034.
• Growth Rate: Extremely high CAGR (~33%), one of the hottest AI infra sectors.
• Drivers: Breakthroughs in LLM reasoning (GPT-4o, Claude 4, Gemini); enterprise demand for automation; the overall "agentic AI" narrative heating up.

Competitive Landscape

Timing Analysis

• Why now?: In 2026, three conditions converged—LLMs can understand web pages, browser infra is mature, and enterprise acceptance of AI agents is rising.
• Tech Maturity: The Perception Layer methodology is proven (86.2% success rate), but still short of "100% reliability." Hybrid architecture is a pragmatic transition.
• Market Readiness: The developer community is already active (1.8k GitHub stars); enterprise demand is clear but has longer procurement cycles.

Team Background

• CEO Andrea Pinto: MIT AI researcher, ETH Zurich/EPFL background, IBM Research/Amazon experience.
• CTO Lucas Giordano: EPFL MLO lab, 4 years of ERP+AI experience.
• Team Size: 5 people, based in San Francisco.
• 6-year collaboration with solid technical foundations.

Funding Status

• Raised: $2.5M pre-seed + YC S25 ($500K).
• Investors: Kima Ventures, Drysdale Ventures, 2100 Ventures, Digital Currency Group, Eden Ventures.
• Total Funding: ~$3.42M.
• Valuation: Undisclosed.

Conclusion

In a nutshell: Notte is currently the most complete full-stack solution in the "turn websites into APIs" space. The hybrid architecture is a smart, pragmatic move. However, as a 5-person team founded in 2025, their success against rivals like Browserbase and Browser Use will depend on execution speed.

Resource Links

2026-03-05 | Trend-Tracker v7.3