BetterWright/betterwright

A persistent, policy-guarded Playwright browser for AI agents — network policy, encrypted credential vault, proof screenshots, and CAPTCHA solving.

What it solves

BetterWright provides AI agents with a persistent, controllable browser environment. It solves the problem of agents needing to start a new browser session every time they perform a task, allowing them to maintain state, navigate pages, and interact with the web across multiple calls.

How it works

It uses a managed browser runtime (BetterChromium) based on Playwright JavaScript. The system operates as a background daemon that keeps tabs and page state alive between CLI calls. It can be integrated into existing agents via a CLI, an MCP (Model Context Protocol) server, or a JavaScript/TypeScript SDK. It also includes a built-in agent that can execute natural-language tasks using configured local or API-based models.

Who it’s for

Developers building AI agents that require web navigation, data extraction, and interaction with websites, as well as users who want to run a browser-controlling agent locally via a CLI.

Highlights

  • Persistent Sessions: Maintains tabs and cookies across calls using named sessions and profiles.
  • Flexible Integration: Supports CLI, MCP server, and SDK for easy addition of browser access to existing agents.
  • Cerebras and Local AI Support: Built-in harness supports local models (via GPU acceleration) and Cerebras API.
  • Human Handoff: Allows a person to take over the browser session to handle steps like MFA (Multi-Factor Authentication).
  • Compact Observations: Provides optimized page snapshots, screenshots, and diffs to reduce the amount of data the agent observes.
  • Credential Management: Includes a trusted vault for encrypted credential filling and redaction.

Related

  • Project
    tumourlove/monolith

    An MCP-based plugin for Unreal Engine that exposes over 1,400 editor actions to AI assistants using a namespace-dispatch pattern to minimize context window usage.

  • Project
    sudoingX/qwen38-mtp

    A community‑driven guide that shows how to enable llama.cpp’s draft‑MTP speculative decoding for the Qwen‑3.8‑27B GGUF model, delivering 33‑145 % faster token generation on consumer GPUs. It provides the exact launch flags, seven tuning rules, and a large table of measured speed‑ups across many GPU/CPU configurations, plus a `probe.py` script for users to verify the boost on their own hardware.

  • Project
    JuliaGPU/OpenCL.jl

    A Julia interface for the OpenCL parallel computation API that enables high-performance parallel programming on GPUs, FPGAs, DSPs, and multicore CPUs.

  • Project
    NeuralEnsemble/PyNN

    A simulator-independent Python language for building neuronal network models that can run on multiple simulators and neuromorphic hardware without modification.