Rahulkishanm / clickolas-cage

a chrome extension that performs web browsing actions autonomously to complete a given goal/task (using LLM as brain).

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

discord: https://discord.gg/wD3Ufgmj

view short explanation more video

clickolas-cage

a chrome extension that performs web browsing actions autonomously to complete a given goal/task (using LLM as a brain). )

Installing

  1. Check if your Node.js version is >= 14.
  2. Run npm install to install the dependencies.
  3. replace your GEMINI api key in .env.template and rename the file to .env

Developing

run these commands to run everything locally:

$ cd clickolas-cage

$ npm run dev

and in a seperate terminal:

$ npx @portkey-ai/gateway

This runs the chrome extension locally.

To link the running extension to chrome:

  1. set your Chrome browser 'Developer mode' on
  2. click 'Load unpacked', and select clickolas-cage/build folder

The main source files are:

  • src/popup/popup.jsx The popup window that shows when you press the extension icon
  • src/background/background.js Persists and facilitates planning
  • src/contentScript/contentScript.js Executes on web pages, executes actions and scrapes elements
  • src/utils.js Helper functions
  • src/llm-utils.js LLM helper functions

Current Flow

  1. User clicks extension icon, enters and submits goal prompt
  2. Popup.jsx sends a message to background.js with request type 'new_goal'
  3. background.js calls promptToFirstStep() in llm-utils.js
  4. LLM returns starting URL which feeds into navURL()
  5. navURL will open a new tab with the starting URL
  6. getNextStep() is called which will get contentScript to extract clickable elements and send it to LLM to generate next step

Packing

After the development of your extension run the command

$ npm run build

Now, the content of build folder will be the extension ready to be submitted to the Chrome Web Store. Just take a look at the official guide to more infos about publishing.


About

a chrome extension that performs web browsing actions autonomously to complete a given goal/task (using LLM as brain).

License:MIT License


Languages

Language:JavaScript 85.9%Language:CSS 10.9%Language:HTML 3.2%