browser-use/browser-harness-js

Self-healing browser harness that enables LLMs to complete any task

What it solves

It removes the restrictive "helper" layers typically found in browser automation tools, which often hide complex parameters and limit what an AI agent can actually do. Instead of providing simplified functions like click(), it provides a direct, typed bridge to the Chrome DevTools Protocol (CDP).

How it works

The project uses a codegen process that reads the official Chrome protocol JSON files to generate typed JavaScript wrappers for every single CDP method. It establishes a persistent WebSocket connection to Chrome, allowing an LLM to call any of the 652 available methods (across 56 domains) directly using the syntax session.Domain.method(params). It includes a tiny CLI and a Bun-based HTTP server to manage the session and forward snippets.

Who it’s for

Developers building AI agents that need full, unconstrained control over a web browser without the limitations of pre-defined automation libraries.

Highlights

  • Zero Wrapping: Exposes the full Chrome DevTools Protocol without simplifying or hiding parameters.
  • Fully Typed: Every CDP method is generated as a typed wrapper, providing autocomplete and documentation based on the official protocol.
  • ’No Version Drift’: The SDK is regenerated from upstream protocol JSON, ensuring new Chrome features are available immediately.
  • Direct LLM Access: Designed for agents to write the CDP calls themselves rather than relying on a fixed set of helper functions.

Related

  • Project
    voidcraft-dev/memory-forge-rs

    A local desktop app for browsing and editing the conversation history of AI coding assistants to fix hallucinations and manage sessions across multiple platforms.

  • Project
    snowflakedb/ArcticInference

    An open-source vLLM plugin that optimizes LLM and embedding inference to increase throughput and reduce latency using advanced parallelism and speculative decoding.

  • Project
    dropbox/gemlite

    A collection of Triton kernels for efficient low-bit matrix multiplication that accelerates LLM prefill and decoding performance.

  • Project
    lucidrains/rectified-flow-pytorch

    A PyTorch implementation of Rectified Flow and its variants, enabling more efficient generative modeling by straightening the paths between noise and data.