# IronBee Express

> A goal-driven browser agent that carries out a one-sentence task in a real browser, then reviews the page, API responses, and backend logs to find root causes.

IronBee Express is a goal-driven browser agent from IronBee. You describe a task in one sentence, such as logging in, adding a product to the cart and opening the cart, and it carries the task out in a real browser. After the run it reviews the final page, the API responses and, when IronBee is connected, backend traces and logs, so you do not write assertions.

## What It Is

IronBee Express is a TypeScript command-line tool with a local web UI. Each step is one decision by TypeSafe Jev over the page's indexed controls, followed by one call to IronBee DevTools that performs the action and returns the next snapshot. Model output never becomes a selector, coordinate or script, so an action can only target a control, option or value that was offered.

A run ends with a PASSED or FAILED verdict. For example, if a page says an order was placed while the API response shows a payment failure, the run fails and the review points to the response or backend log that explains why.

## How a Run Works

Jev chooses operations such as click, type text, select, press enter, hover, scroll, tab switching, wait, ask-user, done and blocked. The README states that DONE is only a claim: it is accepted only when the page and API responses show the goal completed.

After the run, anomalies are collected mechanically: 4xx/5xx responses, failed requests, console errors, failed or slow spans and warning-level logs. Similar anomalies are grouped, and Jev rates each for the goal as none, minor, major or critical.

## Scenarios, Secrets and Handover

A passing run can be saved as a scenario and replayed without engine decisions. If the page changes, the engine takes over from the step that no longer matches and the recording is updated. Secrets such as passwords are typed by reference and are not shown to models. When a social login, CAPTCHA or SMS code needs a person, the run pauses with a "Your turn" prompt and continues from the page you leave it on.

## Models and IronBee Platform

By default there is no LLM in the loop. Optionally, an Anthropic, OpenAI or OpenRouter model, or the Claude Code or Codex CLI, can write free text, take over when the engine is stuck and explain failed runs. Connecting IronBee reports each run as a session with video and reads back the distributed trace from the browser through backend services. The README notes the review is a model's judgement and is therefore probabilistic.

## Setup Path

It requires Node.js 22+, Google Chrome and a TypeSafe API key. Clone the repository, run npm install and npm run build, then start the web UI or use the ibexpress CLI. The project is released under the Elastic License 2.0, and the README says a waitlist is open for IronBee Express on the IronBee platform.

## Features
- Goal-driven browser agent driven by one-sentence tasks
- Step decisions by TypeSafe Jev over indexed page controls
- Post-run review of page, API responses, console, spans and logs
- Root-cause explanation of failed runs
- Record once and replay saved scenarios without engine decisions
- Self-healing recordings when the page changes
- Secrets typed by reference and masked from models
- Human handover for CAPTCHA, SSO or SMS codes
- Local web UI and ibexpress CLI
- Optional LLM via Anthropic, OpenAI, OpenRouter, Claude Code or Codex CLI
- Saved browser profiles
- Native dialog, new tab and optional iframe handling

## Integrations
IronBee DevTools, TypeSafe Jev, IronBee platform, Anthropic, OpenAI, OpenRouter, Claude Code CLI, Codex CLI, Google Chrome

## Platforms
IOS, WEB, API, CLI

## Pricing
Free

## Links
- Website: https://ironbee.ai
- Documentation: https://docs.ironbee.ai/
- Repository: https://github.com/ironbee-ai/ironbee-express
- EveryDev.ai: https://www.everydev.ai/tools/ironbee-express