Automate the web without selectors. Capture the APIs. Hand the result to AI.
Agent Bob is a desktop app for RPA developers who are tired of brittle UI selectors and Selenium-style browser clicking. Instead of hunting for buttons on a page, you use the website yourself — and Bob quietly records the traffic and API calls happening behind the scenes.
Logins. Searches. Form saves. Popups. The lot.
Then Bob groups everything into a structured JSON of URLs, request bodies, and responses. That file is ready to hand to a custom AI agent that turns the recorded steps into real automation code.
Classic RPA watches the screen. Agent Bob watches the network.
If a website already talks to APIs when you click around, those calls are often a cleaner, more stable path to automation than clicking pixels.
flowchart LR
A[You use the web app] --> B[Agent Bob records the traffic]
B --> C[Structured JSON of steps and calls]
C --> D[Your custom AI agent]
D --> E[Automation code without UI selectors]
Built for robotics / RPA developers who want to automate websites and web apps when:
- Selectors keep breaking after every UI tweak
- Selenium-style flows feel too fragile
- The real work is already happening through APIs
- You want a clean recording you can feed into an AI coding agent
flowchart TB
Start[Open Agent Bob] --> URL[Enter the website]
URL --> Record[Click Start and browse normally]
Record --> Steps[Mark steps like Login Search Checkout]
Steps --> Stop[Click Stop]
Stop --> JSON[Export structured JSON]
JSON --> AI[Hand it to your custom AI agent]
AI --> Code[Get code that repeats the workflow]
- Launch Agent Bob
- Open the site or web app
- Start recording — a real Chrome window appears
- Do the workflow exactly as a user would
- Name your steps as important moments happen
- Stop and export the structured JSON
- Pass that JSON to your custom AI agent to generate the automation
While you browse, Bob listens across pages, tabs, and popups in the session.
flowchart LR
Browser[Your browsing] --> Traffic[API calls and page traffic]
Traffic --> Bob[Agent Bob]
Bob --> Grouped[Grouped by your steps]
Grouped --> File[JSON with URLs bodies and responses]
Each useful call can include:
- The URL
- The request body
- The response
- The step it belongs to
So your AI agent doesn’t just see “someone clicked around” — it sees the actual conversation the website had with its backend.
| Old way | Agent Bob way |
|---|---|
| Find fragile UI selectors | Capture the real API traffic |
| Rebuild flows in Selenium | Record once by using the site |
| Break when the UI moves | Lean on the calls the app already makes |
| Hand vague notes to AI | Hand structured JSON of URLs, bodies, and responses |
Open the portable build:
dist/AgentBob-1.0.0-portable.exe
No Node.js needed just to run it.
npm install
npm run build:portableInstaller + portable:
npm run build:winnpm startBob prefers real Google Chrome with a saved profile, so logins can stick between sessions and the browser feels less like a bot. If Chrome isn’t available, it can fall back to a bundled browser.
Show the workflow once. Capture the APIs. Let AI write the automation.
Agent Bob helps RPA developers skip UI selectors and Selenium gymnastics — and start from the traffic that already makes the web app work.