WebVoyager

WebVoyager: Building an End-to-End Web Agent with Large Multimodal Models
WebVoyager is a new vision-powered web-browsing agent that uses browser screenshots and “Set-of-mark” prompting to conduct research, analyze images, and perform other tasks.
In this video, we will show you how to build WebVoyager using LangGraph, an open-source framework for building stateful, multi-actor AI applications.
Links:
Python Code: github.com/lan...
WebVoyager Paper: arxiv.org/abs/...
Set-of-mark Paper: arxiv.org/abs/...
Developing AI applications is easier with LangSmith. Create a free account at smith.langchai....
New to LangGraph? Check out the intro video: • LangGraph: Intro

[@ricand5498]

My left ear enjoyed this video very much

[@Jakolo121]

LOL I thought my headphones were broken

[@dineshkumarkinjangi8994]

@@Jakolo121i kept mine for charging 😂

[@LangChain]

Sorry about that... not sure why!

[@AtulR]

On mac: system settings > accessibility > audio > play stereo audio as mono. Just remember to switch it back to off after this video

[@anonymous6666]

I greatly appreciate the thorough, simple and easy to understand explanations, especially surrounding LangGraph

[@darkmatter9583]

please lets make a crowfunding to give him money for a better microphone, his videos are really good, he deserves it, thanks for the amazing contribution to the community

[@andrushka324]

That is so cool that you guy make video about different use cases. Please, improve sound quality and describe topics more detailed.🙂

[@ajinkya81194]

Is there a way to do this using other LMMs such Gemini pro vision or Llava 1.6 ?

[@mayanklohani19]

Can it used to define any url and do kind of functionality testing? Tried changing the url but didn't worked.

[@mr.daniish]

Creative and clean! the sound could be improved though. Still great value

[@free_thinker4958]

I read the example code when I came here I was understanding a little bit the code but once I take a look at its langgraph video here I feel so confused because the pace of the video is so fast

[@oluwatoniemmanuel6348]

Finally someone can relate

[@antwierasmus]

How you run this as a python script and not in jupyter notebook? I am getting an error "Event loop is closed", perhaps related to asyncio

[@muhammadsalman845]

Did you got it solved? If so can you help?

[@mayanklohani19]

can we use llava model here from ollama?

[@aifarmerokay]

We want agent with local
Open source Llm with memory implementation, 😊

[@gitmaxd]

This is great, ty!

[@KushJuvekar-j3f]

Is anyone else getting prompt must be 'str' error with this code?

[@aiexplainai2]

very interesting idea!

[@zenofthepup1530]

prompt error on the hub

[@build.aiagents]

Phenomenal

[@metamarketing3402]

This is very Cool.😃

[@VivekGautam-o8v]

These are good , But looking for JavaScript support

[@DanielGonzalez-wr7fz]

I would like to implement a "Learning Mode" for this WebVoyager Agent. In order to teach this agent an action by recording a manual navigation through the browser and then save it as a "Tool" or a "Succesion of steps".
Could you please give me some references or some clues of how can I acchieve this ?

[@piyushsinha5545]

If you got the solution, please do share. working on something similar

[@ayeshaimran]

perhaps use RAG for this purpose... so every set of action can be added to a vector database along with its result and before taking any steps the agent can do a quick vector search to see if that action has been done before and the successful series of steps taken

[@avisimkin1719]

Did anyone try this with a local model? (Llava for example)

[@2107mann]

Awesome

[@TristanvanDoorn]

Nice, but it seems to have some glitches that need to be ironed out. Nevertheless, great work!

[@cuties4698]

Awesome project, but he is only speaking to my right ear.

[@Cynosureepr]

You have your headphones on backward.