Actionbook provides a browser action engine designed to significantly improve the efficiency and reliability of AI agents interacting with websites. It addresses challenges like slow execution, high token costs, brittle selectors, and LLM hallucinations by offering pre-computed action manuals and DOM structures. This enables AI agents to operate websites instantly and with greater accuracy, benefiting developers building AI-powered automation and researchers working with web-based AI agents.

How It Works

Actionbook's core innovation lies in its "Action manuals," which are up-to-date, versioned collections of relevant DOM selectors and semantic JSON definitions for specific websites. Instead of requiring AI agents to parse entire HTML pages, Actionbook provides these concise, pre-computed manuals directly into the LLM's context. This approach drastically reduces the time agents spend exploring web pages and minimizes token consumption by sending only necessary element information, leading to faster, more resilient, and cost-effective web automation.

Quick Start & Requirements

• Primary Install: npm install -g @actionbookdev/cli
• Prerequisites: Node.js >= 18 (20+ recommended), pnpm >= 10, PostgreSQL database (local or hosted), and a compatible system browser (Chrome, Brave, Edge, Arc, Chromium).
• Setup: CLI quick start is achievable in under 2 minutes. Development setup involves dependency installation, environment variable configuration, and database migrations.
• Links: Website, Docs, Examples.

Highlighted Details

• Achieves up to 10x faster execution for AI agents.
• Offers up to 100x token savings by providing concise, semantic JSON definitions instead of full HTML.
• Ensures resilient automation through maintained and versioned action manuals that adapt to website changes.
• Provides universal compatibility with any LLM (OpenAI, Anthropic, Gemini) and AI operator framework.

Maintenance & Community

The project is developed as a monorepo using pnpm workspaces and Turborepo. Community engagement is fostered through Discord and updates are shared on X. Contributions can be made by suggesting websites for indexing or joining the waitlist for the private beta.

Licensing & Compatibility

The specific license details are available in the LICENSE file. The system is designed for broad compatibility with various LLMs and AI frameworks, facilitating commercial use and integration into closed-source projects.

Limitations & Caveats

Actionbook is currently in a private beta phase, indicated by a waitlist for interested contributors and users. The development setup requires specific Node.js, pnpm, and PostgreSQL configurations.