WebArena Environments

Self-contained web application environments for evaluating browser-use agents. Each app is a standalone HTML/CSS/JS application served by a small Python HTTP server.

Based on WebArena-Infinity.

Quick Start

# Install dependencies
bash setup.sh

# Run a single app
cd apps/gmail && python3 server.py --port 8000

# Run all apps with a hub page
python3 serve_all.py --port 9000 --demo

Open http://localhost:8000 (single app) or http://localhost:9000 (hub) in your browser.

Available Environments

App	Description
elation-clinical-records	EHR clinical records management
elation-patient-communication	EHR patient messaging
elation-prescriptions	EHR prescription management
figma-slides	Slide deck editor (Figma-style)
figma-text-and-typography	Text/typography editor (Figma-style)
gitlab-plan-and-track	Project planning and issue tracking
gmail	Email client
gmail-accounts-and-contacts	Gmail account and contacts management
google-sheets	Spreadsheet editor with formulas, charts, and multi-sheet workbooks
handshake-career-exploration	Career exploration platform
linear-account-settings	Project management account settings
paypal-my-wallet	Digital wallet management
superhuman-general	Email client (Superhuman-style)
xero-invoicing	Invoice management

Running Agent Evaluations

# Run a single task
uv run python evaluation/run_eval_parallel.py \
    --model gpt \
    --task-id task_e1 \
    --workers 1 \
    --web-app apps/gmail

# Run all easy tasks with visible browser
uv run python evaluation/run_eval_parallel.py \
    --model gpt \
    --difficulty easy \
    --workers 1 \
    --web-app apps/google-sheets \
    --headed

Supported Models

Flag	Model	API Key
gpt	GPT-4o	OPENAI_API_KEY
gemini-flash	Gemini Flash 3	GOOGLE_API_KEY
gemini-pro	Gemini Pro 3	GOOGLE_API_KEY
claude	Claude Sonnet 4.6	ANTHROPIC_API_KEY

Test Mode

Add --test-mode when launching a server to get an in-browser test panel for manually running and verifying tasks:

cd apps/google-sheets && python3 server.py --port 8000 --test-mode