Deterministic, Self Healing Workflows (RPA 2.0)

⚙️ Workflow Use is the easiest way to create and execute deterministic workflows with variables which fallback to Browser Use if a step fails. You just show the recorder the workflow, we automatically generate the workflow.

❗ This project is in very early development so we don't recommend using this in production. Lots of things will change and we don't have a release schedule yet. Originally, the project was born out of customer demand to make Browser Use more reliable and deterministic.

🚀 NEW: Generation Mode

Automatically generate workflows from natural language! Describe your task, we run browser-use once, then create a reusable semantic workflow stored in a database.

Quick Commands

# Generate workflow from task description
python cli.py generate-workflow "Find GitHub stars for browser-use repo"

# List all workflows
python cli.py list-workflows

# Filter by generation mode
python cli.py list-workflows --generation-mode browser_use

# Run stored workflow
python cli.py run-stored-workflow <workflow-id> --prompt "Find stars for playwright repo"

# View workflow details
python cli.py workflow-info <workflow-id>

# Delete workflow
python cli.py delete-workflow <workflow-id>

How It Works

1. Describe: Give a task in natural language
2. Execute: Browser-use completes the task once
3. Generate: Execution history → semantic workflow with parameters
4. Store: Save to database with metadata
5. Reuse: Run the workflow with different inputs, no AI needed

Advanced Options

# Custom models for generation
python cli.py generate-workflow "Your task" \
  --agent-model "gpt-4.1-mini" \
  --extraction-model "gpt-4.1-mini" \
  --workflow-model "gpt-4o"

# Use Browser-Use Cloud browser
python cli.py generate-workflow "Your task" --use-cloud

# Save to custom location
python cli.py generate-workflow "Your task" --output-file ./my-workflow.json

# Skip database storage
python cli.py generate-workflow "Your task" --no-save-to-storage

Storage

Workflows stored at workflows/storage/:

• metadata.json - Searchable index of all workflows
• workflows/<id>.workflow.json - Individual workflow files

Programmatic Usage

from workflow_use.healing.service import HealingService
from workflow_use.storage.service import WorkflowStorageService
from browser_use.llm import ChatOpenAI

healing_service = HealingService(llm=ChatOpenAI(model='gpt-4o'))
storage_service = WorkflowStorageService()

# Generate workflow
workflow = await healing_service.generate_workflow_from_prompt(
    prompt="Fill contact form on example.com",
    agent_llm=ChatOpenAI(model='gpt-4.1-mini'),
    extraction_llm=ChatOpenAI(model='gpt-4.1-mini'),
    use_cloud=True  # Optional: use Browser-Use Cloud
)

# Save to storage
metadata = storage_service.save_workflow(
    workflow=workflow,
    generation_mode='browser_use',
    original_task="Fill contact form on example.com"
)

# Retrieve and execute
loaded_workflow = storage_service.get_workflow(metadata.id)

Quick start

git clone https://github.com/browser-use/workflow-use

Build the extension

cd extension && npm install && npm run build

Setup workflow environment

cd .. && cd workflows
uv sync
source .venv/bin/activate # for mac / linux
playwright install chromium
cp .env.example .env # add your OPENAI_API_KEY to the .env file

Run workflow as tool

python cli.py run-as-tool examples/example.workflow.json --prompt "fill the form with example data"

Run workflow with predefined variables

python cli.py run-workflow examples/example.workflow.json

Record your own workflow

python cli.py create-workflow

See all commands

python cli.py --help

Usage from python

Running the workflow files is as simple as:

from workflow_use import Workflow

workflow = Workflow.load_from_file("example.workflow.json")
result = asyncio.run(workflow.run_as_tool("I want to search for 'workflow use'"))

Cloud Browser Support

Run workflows in Browser-Use Cloud with semantic abstraction (no AI): (NOTE: Set BROWSER_USE_API_KEY environment variable)

from workflow_use import Workflow

workflow = Workflow.load_from_file("workflow.json", llm, use_cloud=True)
result = await workflow.run_with_no_ai()  # No LLM calls, uses semantic mapping

Examples:

• examples/cloud_browser_demo.py - Load recorded workflow and run on cloud

Launch the GUI

The Workflow UI provides a visual interface for managing, viewing, and executing workflows.

Option 1: Using the CLI command (Recommended)

The easiest way to start the GUI is with the built-in CLI command:

cd workflows
python cli.py launch-gui

This command will:

• Start the backend server (FastAPI)
• Start the frontend development server
• Automatically open http://localhost:5173 in your browser
• Capture logs to the ./tmp/logs directory

Press Ctrl+C to stop both servers when you're done.

Option 2: Start servers separately

Alternatively, you can start the servers individually:

Start the backend server

cd workflows
uvicorn backend.api:app --reload

Start the frontend development server

cd ui
npm install
npm run dev

Once both servers are running, you can access the Workflow GUI at http://localhost:5173 in your browser. The UI allows you to:

• Visualize workflows as interactive graphs
• Execute workflows with custom input parameters
• Monitor workflow execution logs in real-time
• Edit workflow metadata and details

Demos

Workflow Use filling out form instantly

workflow-use-demo.mp4

Gregor's explanation

workflow-use-gregor-demo.mp4

Features

• 🔁 Record Once, Reuse Forever: Record browser interactions once and replay them indefinitely.
• ⏳ Show, don't prompt: No need to spend hours prompting Browser Use to do the same thing over and over again.
• ⚙️ Structured & Executable Workflows: Converts recordings into deterministic, fast, and reliable workflows which automatically extract variables from forms.
• 🪄 Human-like Interaction Understanding: Intelligently filters noise from recordings to create meaningful workflows.
• 🔒 Enterprise-Ready Foundation: Built for future scalability with features like self-healing and workflow diffs.

Vision and roadmap

Show computer what it needs to do once, and it will do it over and over again without any human intervention.

Workflows

• Nice way to use the .json files inside python code
• Improve LLM fallback when step fails (currently really bad)
• Self healing, if it fails automatically agent kicks in and updates the workflow file
• Better support for LLM steps
• Take output from previous steps and use it as input for next steps
• Expose workflows as MCP tools
• Use Browser Use to automatically create workflows from websites

Developer experience

• Improve CLI
• Improve extension
• Step editor

Agent

• Allow Browser Use to use the workflows as MCP tools
• Use workflows as website caching layer