Introducing Raccoon AI

We're launching Raccoon AI, a platform built by developers, for developers, to tackle the core challenges of web automation agents today.

There is no marketing stuff ahead, just a small technical dive into how we're making it more reliable, efficient, and developer-centric. 🦝

The Current Landscape

• Traditional Scraping: Dies with the slightest website update. Completely inadequate for dynamic content and easily detectable by any antibot system.
• Headless Browsers and Stuff like Stagehand: Powerful, but you're micromanaging every interaction. Scaling is a major headache (browser instances, generalising and supporting a spectrum of websites, captchas etc.).
• "Browser and Computer Use Agents": These are awesome and could work for building fun projects but are often thin layers over LLMs which make them super dependent upon the capabilities of the LLM in use and you still need to take care of the setting up the infrastructure and all the insane costs that come along with them.

We're addressing these problems with a fundamentally different architecture, and with a clear path to becoming a reliable infra for all things web automation.

Raccoon AI: Made for Today’s Web

Raccoon AI runs on three pillars that play nice together:

1. LAM API: Large Action Model System (It's not a foundational AI model, yet)

Tell it what you want in plain English, it then translates that intent into concrete actions within a browser.

• Key Tech:
  • Elemetry: Predictive DOM Interaction Caching: This is a core innovation. Elemetry caches information about DOM elements (XPath, text, actions, URL, and a semantic role) and the context of user interactions. It uses this cached data to predict future actions, without needing to re-process the entire page.
    • Data Structures: Elemetry uses Node objects (representing DOM elements), Timeline objects (representing sequences of actions), and TimelineNode objects (linking Nodes and context within a Timeline).
    • Prediction Process: When you provide a query and an action_history (a list of previously interacted Node IDs), Elemetry finds matching Timelines, ranks them based on how well their context matches your query, and predicts the next Node (and therefore, the next action).
    • Benefits: Massive performance improvements (reduced DOM processing), increased robustness (can often identify elements even if the XPath changes), and the ability to learn from past interactions.
  • Xtractor: Semantic DOM Simplification: Inspired by projects like Tarsier, Xtractor preprocesses the DOM, extracting only the semantically relevant information and converting it into a simplified, text-based representation. This:
    • Reduces the complexity the LLMs have to deal with.
    • Improves performance.
    • Makes the system more robust to minor website changes (as it focuses on meaning, not just structure).
  • /lam/run Endpoint: Your entry point. Provide a query (natural language), an optional data extraction schema, and a raccoon_passcode (for authenticated actions). Checkout here.

2. Fleet API: Cloud Browsers You Control

Low-level access to our custom browser cloud. We built it from the ground up for scale and dev freedom.

• Key Features:
  • Our Own Stack: No third-party nonsense, full control over the browser setup.
  • Stealth Mode: Avoids CAPTCHAs (solvers incoming), blocks ads, beats anti-bot tricks.
  • Residential Proxies: Pick your location and identity, or provide your own.
  • Websocket/CDP: Plug into Playwright or Puppeteer via /sessions/create (gives a websocket_url).
  • Livestreaming: Watch and debug live with a livestream_url.
  • Custom Extensions: Upload your own browser mods/extensions.

3. Tail SDK: Credential-Free User Auth

Act for your users without touching their passwords.

• Key Concepts:
  • Raccoon Passcodes: Unique IDs for users and their accounts.
  • Session-Based Auth: No creds stored.
  • Easy Linking: OAuth-style flow for any app.
  • User Power: They control what’s linked.

How We Stack Up

Browser-Use

• High-level LLM-to-browser glue.
• Great for quick prototypes and simple prompts.
• Weak on scale and limited by LLMs.
• Raccoon Edge: Elemetry caches DOM smarts, Xtractor simplifies chaos, built for practical applications and play.

OpenAI’s Operator & Google’s Mariner

• Cool LLM experiments.
• Focused more on the consumer side
• Limited developer tooling and API access
• Generally bad sometimes and super costly.
• Raccoon Edge: Full APIs, real-world focus, dev-first from day one.

API-First All the Way

Everything’s an API. Integrate it, tweak it, own it, and if you want more, just ask for it.

Hands-On with Raccoon AI

If you've read this far, I know you want to get to the code:

1. Quickstart: See it rip with the LAM Quickstart.
2. Platform: Grab your key at Platform.
3. Book a Call: Let’s chat about what you’re building. Pick a time.

At Raccoon AI, we're giving you better tools to build amazing things by handling the complex challenges of web automation.