AI Frameworks · browser-use

browser-harness

Browser Harness is a Python framework that connects large language models directly to Chrome via CDP (Chrome DevTools Protocol), enabling autonomous browser automation. The system improves itself during execution by having the agent write missing helper code as needed.

Talk to DEV.co Discuss Implementation

Source: GitHub — github.com/browser-use/browser-harness

15.8k

GitHub stars

1.5k

Forks

Python

Primary language

MIT

License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

Field	Value
Repository	browser-use/browser-harness
Owner	browser-use
Primary language	Python
License	MIT — OSI-approved
Stars	15.8k
Forks	1.5k
Open issues	170
Latest release	v0.1.4 (2026-07-01)
Last updated	2026-07-07
Source	https://github.com/browser-use/browser-harness

What browser-harness is

A ~1k-line CDP harness (MIT-licensed) that exposes Chrome's remote debugging protocol to LLM agents over WebSocket. Agents generate and edit Python helpers and domain-specific skills at runtime; the harness supports local Chrome attachment or Browser Use Cloud deployment.

Quickstart

Get the browser-harness source

Clone the repository and explore it locally.

terminalbash

1git clone https://github.com/browser-use/browser-harness.git2cd browser-harness3# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Autonomous web task completion

Use when you need an LLM to execute complex, multi-step web workflows (form filling, data extraction, account management) without pre-scripted automation code. The self-healing model adapts to UI changes.

Rapid prototyping of browser agents

Ideal for building proof-of-concept agents that learn domain workflows (e.g., LinkedIn outreach, Amazon ordering) as they execute, with reusable skills captured for future runs.

Cloud-based browser automation at scale

Leverage Browser Use Cloud's free tier (3 concurrent browsers, proxies, captcha solving) for headless deployment without managing infrastructure or dealing with anti-bot measures.

Implementation considerations

Agent must have read/write access to `${XDG_CONFIG_HOME:-~/.config}/browser-harness/` for helper code and domain skills; plan config management and versioning accordingly.
Requires Python 3.12 and Chrome/Chromium with `--remote-debugging-port` enabled or Browser Use Cloud API key; local setup involves manual CDT endpoint exposure.
Agent-generated skills are stored in plaintext Python files; implement code review, sandboxing, or validation if running untrusted agent code.
CDP protocol version must match Chrome version; drift between client and browser can break commands. Test with your deployed Chrome version.
LLM call latency per action cycle is opaque in the README; budget for API rate limits and costs if operating at scale (e.g., 3 concurrent browsers).

When to avoid it — and what to weigh

You need deterministic, production-grade automation — Runtime code generation and self-healing logic introduce non-determinism. High-SLA workflows requiring predictable behavior and audit trails should use mature test automation tools.
Your site heavily restricts or detects automation — While Browser Use Cloud offers stealth options, this is a browser-based agent, not a bypass tool. Sites with strict bot detection or legal ToS prohibitions on scraping may require alternative approaches.
You have no LLM API budget or cannot rely on external inference — The harness itself is free/OSS, but operation requires calling an LLM (e.g., Claude, GPT-4) continuously. Offline or cost-sensitive scenarios need local model integration (not described in data).
You need stable, backward-compatible APIs for production integrations — Project is ~2 months old (created 2026-04-17), latest release v0.1.4 (2026-07-01). Early-stage versioning suggests breaking changes are possible; production lock-in carries risk.

License & commercial use

MIT License (permissive OSI-approved). Allows commercial use, modification, and distribution with attribution and liability waiver.

MIT is a permissive, commercially compatible license. However, ensure you own or have rights to the Chrome/Chromium and LLM API credentials; the harness itself imposes no restrictions on commercial deployment. Verify Browser Use Cloud terms for commercial free-tier usage.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

Signal	Assessment
Maintenance	Active
Documentation	Adequate
License clarity	Clear
Deployment complexity	Moderate
DEV.co fit	Good
Assessment confidence	High

Security considerations

Runtime code generation: agent writes Python helpers into user config; no sandboxing mentioned. CDP protocol exposes full browser control; restrict CDT endpoint to trusted networks. Skill files in plaintext; implement access controls if shared across teams. Browser Use Cloud adds third-party credential handling; verify their data residency/encryption policies independently. No security audit or vulnerability disclosure process mentioned in data.

Alternatives to consider

Playwright + custom Python scripting

More mature, deterministic, and widely used. Requires manual orchestration but provides explicit control and extensive documentation. No runtime code generation overhead.

Anthropic's Computer Use (Claude)

Built-in vision-based agent automation via Claude API. Simpler setup for Claude users but less flexible than direct CDP access; pricing per action unclear.

Selenium with BDD frameworks (Cucumber, Robot)

Enterprise-grade, battle-tested automation with audit trails and CI/CD integration. Steeper learning curve; requires pre-written tests but production-ready reliability.

Software development agency

Build on browser-harness with DEV.co software developers

Evaluate Browser Harness for rapid browser automation prototyping. Paste the setup prompt into Claude Code, connect to your browser, and let the agent learn. For production workloads, assess LLM cost, determinism requirements, and anti-bot resilience.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

browser-use BrowserOS harness

Related on DEV.co

Explore the category and the services that help you build with it.

AI Frameworks AI Application Development Web App Development Custom Software Development Talk to DEV.co

browser-harness FAQ

Do I need to write code for the agent to work?

No. The agent generates helper code and domain skills at runtime. You provide the setup prompt and the agent learns as it executes. However, skill review and code governance are recommended for production use.

Can I run this offline or without an LLM API?

No. The harness is a CDP client; the agent is external (Claude, GPT-4, etc.). You must have an LLM API key and internet connectivity to call it. Local model integration is not documented.

What happens if the website changes?

The agent re-learns and updates the domain skill on the fly. Skills are stored as Python code; you can review, fork, or contribute them back to the repo.

Is this suitable for production automation?

Depends on your tolerance for non-determinism and LLM failure modes. Browser Use Cloud provides stealth and captcha solving for tricky sites, but the agent may hallucinate or retry repeatedly. Use for high-uptime, fault-tolerant tasks only; critical workflows need deterministic scripts.

Custom software development services

From first prototype to production, DEV.co delivers software development services around tools like browser-harness. Our software development agency staffs experienced software developers and web developers for custom software development, web development, integrations, and ongoing support across ai frameworks and beyond.

Start building autonomous web agents today.

Talk to DEV.co Schedule a Consultation