Skip to content

Latest commit

 

History

History
118 lines (83 loc) · 6.31 KB

config-feature-browse.md

File metadata and controls

118 lines (83 loc) · 6.31 KB
Raw

Browse Functionality in big-AGI 🌐

Allows users to load web pages across various components of big-AGI. This feature is supported by Puppeteer-based browsing services, which are the most common way to render web pages in a headless environment.

Once configured, the Browsing service provides the following functionality:

  • ✅ Paste a URL: Simply paste/drag a URL into the chat, and big-AGI will load and attach the page (very effective)
  • ✅ Use /browse: Type /browse [URL] in the chat to command big-AGI to load the specified web page
  • ✅ ReAct: ReAct will automatically use the loadURL() function whenever a URL is encountered

It does not yet support the following functionality:

  • ✖️ Auto-browsing by LLMs: if an LLM encounters a URL, it will NOT load the page and will likely respond that it cannot browse the web - No technical limitation, just haven't gotten to implement this yet outside of /react yet

First of all, you need to procure a Puppteer web browsing service endpoint. big-AGI supports services like:

Service Working Type Location Special Features
BrightData Scraping Browser Yes Proprietary Cloud Advanced scraping tools, global IP pool
Cloudflare Browser Rendering ? Proprietary Cloud Integrated CDN, optimized browser rendering
⬇️ Browserless 2.0 Okay OpenSource Local (Docker) Parallelism, debug viewer, advanced APIs
⬇️ Your Chrome Browser (ALPHA) Alpha Proprietary Local (Chrome) Personal, experimental use (ALPHA!)
other Puppeteer-based WSS Services ? Varied Cloud/Local Service-specific features

Configuration

  1. Procure an Endpoint

    • Ensure that your browsing service is running (remote or local) and has a WebSocket endpoint available
    • Write down the address: wss://${auth}@{some host}:{port}, or ws:// for local services on your machine

  2. Configure big-AGI

    • navigate to Preferences > Tools > Browse
    • Enter the 'wss://...' connection string provided by your browsing service

  3. Enable Features: Choose which browse-related features you want to enable:

    • Attach URLs: Automatically load and attach a page when pasting a URL into the composer
    • /browse Command: Use the /browse command in the chat to load a web page
    • ReAct: Enable the loadURL() function in ReAct for advanced interactions

🌐 Browserless 2.0

Browserless 2.0 is a Docker-based service that provides a headless browsing experience compatible with big-AGI. An open-source solution that simplifies web automation tasks, in a scalable manner.

Launch Browserless with:

docker run -p 9222:3000 browserless/chrome:latest

Now you can use the following connection string in big-AGI: ws://127.0.0.1:9222. You can also browse to http://127.0.0.1:9222 to see the Browserless debug viewer and configure some options.

The chat agent won't be able to access the web sites if the browserless container does not have direct Internet access. You can resolve the issue by defining internet proxy for the running container. You can then use the evironment file in the a `docker-compose.yaml

 browserless:
    image: browserless/chrome:latest
    env_file:
      - .env
    ports:
      - "9222:3000"  # Map host's port 9222 to container's port 3000
    environment:
      - MAX_CONCURRENT_SESSIONS=10

You can then add the proxy lines to your .env file.

https_proxy=http://PROXY-IP:PROXY-PORT
http_proxy=http://PROXY-IP:PROXY-PORT

This is how you can define it in a one liner docker docker run --env https_proxy=http://PROXY-IP:PROXY-PORT --env http_proxy=http://PROXY-IP:PROXY-PORT -p 9222:3000 browserless/chrome:latest

Note: if you are using docker-compose, please see the docker/docker-compose-browserless.yaml file for an example on how to run big-AGI and Browserless simultaneously in a single application.

🌐 Your own Chrome browser

EXPERIMENTAL - UNTESTED - You can use your own Chrome browser as a browsing service, by configuring it to expose a WebSocket endpoint.

  • close all the Chrome instances (on Windows, check the Task Manager if still running)
  • start Chrome with the following command line options (on Windows, you can edit the shortcut properties):
    • --remote-debugging-port=9222
  • go to http://localhost:9222/json/version and copy the webSocketDebuggerUrl value
    • it should be something like: ws://localhost:9222/...
  • paste the value into the Endpoint configuration (see point 2 in the configuration)

Server-Side Configuration

You can set the Puppeteer WebSocket endpoint (PUPPETEER_WSS_ENDPOINT) in the deployment before running it. This is useful for self-hosted instances or when you want to pre-configure the endpoint for all users, and will allow your to skip points 2 and 3 above.

Always deploy your own user authentication, authorization and security solution. For this feature, the tRPC route that provides browsing service, shall be secured with a user authentication and authorization solution, to prevent unauthorized access to the browsing service.

Support

If you encounter any issues or have questions about configuring the browse functionality, join our community on Discord for support and discussions.

Official Discord

Enjoy the enhanced browsing experience within big-AGI and explore the web without ever leaving your chat!

Last updated on Feb 27, 2024 (edit on GitHub)