agentrhq/webcmd

The Browser Infra that learns and cuts token spend by upto 90%

What it solves

Webcmd provides a self-learning browser infrastructure for AI agents to prevent them from having to rediscover the same websites and navigational paths every time they run a task. This reduces token expenditure by up to 90% and increases the reliability and speed of browser automation by turning variable browser interactions into deterministic commands.

How it works

Webcmd implements a layered approach to browser interaction:

  • Layer 0 (Live Browser Control): Allows agents to use webcmd browser to inspect, click, and extract data from unfamiliar sites using a sandboxed Playwright-style environment.
  • Layer 1 (Sitemap Memory): Captures agent-facing sitemaps of observed pages, workflows, and APIs to avoid repeating exploration.
  • Layer 2 (CLI Authoring): Creates reusable adapters for specific sites with structured output, shifting the agent's token spend from navigation to the actual task.
  • Layer 3 (Extended CLIs): Develops tailored commands for deterministic workflows that run instantly.

Users can manage browser state via Profiles (cookie jars) and Sessions (independent browser windows), allowing parallel agents to operate securely.

Who it’s for

Developers building AI agents that require frequent, reliable, and cost-effective web browsing and data extraction from diverse sites (social media, research databases, shopping platforms, and AI tools).

Highlights

  • Token Efficiency: Significantly reduces cost and variance by collapsing browser interactions into CLI commands.
  • Layered Learning: Progresses from raw browser control to deterministic CLI adapters.
  • Session Management: Supports profiles and isolated sessions for authenticated browsing.
  • Plugin Ecosystem: Includes a catalog of community-developed site adapters (e.g., for Hacker News, PyPI, Skyscanner).
  • High Performance: Outperforms other tools on the BU Bench V1 benchmark in terms of accuracy, cost per task, and agent turns.

Related

  • Project
    webbrain-one/webbrain

    An open-source AI browser agent extension that automates web tasks, extracts data, and runs multi-step workflows using local or cloud LLMs.

  • Project
    only-cli/oc

    A CLI tool that converts websites into a compact, numbered interface for AI agents, drastically reducing token usage and bypassing bot blockers during web browsing.

  • Project
    browser-use/workflow-use

    A tool for creating deterministic, self-healing browser automation workflows by recording interactions or generating them from natural language, reducing reliance on constant LLM prompting.

  • Project
    platonai/Browser4

    Browser4 is an AI‑native browser automation platform. It offers a Rust‑based backend and a human‑friendly CLI (`browser4‑cli`) that lets humans or LLM‑driven agents open real Chromium sessions, interact with pages (click, fill, wait, etc.), and extract data via deterministic X‑SQL/CSS selectors or LLM‑assisted commands. It supports zero‑token, ML‑based spreadsheet generation (WebMiner), high‑throughput swarm crawling, and an extensible plugin/skill system, making it suitable for large‑scale web automation and autonomous agents.

  • Project
    jackwener/OpenCLI

    A tool that converts websites and browser sessions into a CLI, enabling humans and AI agents to automate web interactions using logged-in Chrome profiles.