Robot Framework Agent

Robot Framework Agent is an open-source experiment to bring intent-level, LLM-augmented automation directly into Robot Framework suites. The goal is simple: describe what should happen in natural language and let the agent turn that intent into Appium/Selenium actions, visual checks, and future autonomous behaviors. No hype—just a practical attempt to give Robot Framework a “brain” testers can actually use.

Project Status & Vision

This repository is an ongoing exploration toward building a practical agentic toolkit for software testers.

Current focus

• Agent.Do — interpret natural-language actions
• Agent.Check — interpret natural-language assertions

Both keywords are in active development. Work is underway to make them stable, predictable, and usable on real production application.

Near-term goals

• add mid-level capabilities such as locating visual/semantic elements
• extend Agent.ReportBug so legacy suites (without agentic capabilities) can file actionable reports automatically
• expose internal locator/vision data for debugging
• experiment with autonomous behaviours in controlled, test-safe ways

None of this is final; the library is evolving through experiments, refactoring, and real-world usage. Contributions, critiques, and field tests are welcome.

Presented at RoboCon 2026 (Helsinki)

This project will be showcased at RoboCon 2026 during the talk “What if Robot Framework Had a Brain?”
👉 https://www.robocon.io/agenda/helsinki#what-if-robot-framework-have-a-brain

The session will cover the architecture behind the agent, early field results, and how intent-based keywords can help teams work at a higher abstraction level without abandoning Robot Framework’s strengths.

Support & Sponsorship

If you find this work useful, inspiring, or simply want to help move agentic testing forward, you can support the project by:

• sponsoring dedicated development time
• funding compute for VLM/LLM experimentation
• contributing code, test cases, or research notes

This is not a finished product; it’s an open laboratory for building smarter testing tools. Any support—small or large—helps the project grow and stay independent.

Overview

Robot Framework Agent exposes high-level keywords to describe actions and checks in natural language; the agent then translates them into Appium (real device/emulator) and other UI interactions.

Usage (Robot Framework)

In your .robot file:

*** Settings ***
Library    Agent    llm_client=openai    llm_model=gpt-4o-mini

*** Test Cases ***
Agent Example
    Agent.Do      accept cookies
    Agent.Check   the screen correctly shows the Homepage

Notes

• Provided keywords: Agent.Do <instruction>, Agent.Check <instruction>.
• Works with real mobile devices and BrowserStack; see more details below.

Running Tests on BrowserStack

Prerequisites

1. BrowserStack Account

   • Sign up at browserstack.com if you don't have an account
   • Get your username and access key from BrowserStack dashboard

2. Environment Setup

   # Install required dependencies
   pip install -r requirements.txt
   pip install browserstack-sdk

3. Configuration

   • Create/update browserstack.yml in your project root
   • Configure your credentials:

     userName: YOUR_USERNAME
     accessKey: YOUR_ACCESS_KEY

   • Update app path with your app's BrowserStack URL:

     app: bs://YOUR_APP_ID

Running Tests

1. Single Test Execution

   browserstack-sdk robot tests/atest/your_test.robot

2. Running Test Suites

   browserstack-sdk robot tests/atest/

3. Parallel Execution

   browserstack-sdk robot --variable parallel_execution:true tests/atest/

Device Configuration

• Edit browserstack.yml to specify target devices:

  platforms:
    - deviceName: Samsung Galaxy S21
      platformVersion: 12.0
      platformName: Android
    # Add more devices as needed

Development Tips

1. Debug Mode

   • Set debug: true in browserstack.yml
   • Enable detailed logs: consoleLogs: info

2. Local Testing

   • For testing with local apps/servers:

     browserstackLocal: true

3. Build Identification

   • Set unique build names for tracking:

     buildName: "Dev Build"
     buildIdentifier: ${BUILD_NUMBER}

Viewing Results

1. Real-time monitoring: BrowserStack dashboard
2. Test reports: Available in log/ directory after execution
3. Session videos: Automatically recorded and available in BrowserStack dashboard