LaVague Weekly Update
LaVague Large Action Model framework for building Web Agents

LaVague Weekly Update

Welcome to our weekly updates, where we announce the latest news on our open-source framework for Web Agents & share what's coming next and how you can get involved!

Laura Yie

Welcome to our weekly dev updates!

Here’s what we’ve been up to over the past couple of weeks:

🆕 Recent Developments

  • 🤖 Launching our first implementation of Web Agents: You can get started with our Web Agents here.
  • 🚀 Refactoring the code base: We've moved to a multi-package namespace structure with a core bundle of packages that are downloaded when you run pip install lavague, plus additional optional packages!
  • 🗺️ Launching our World Model module: The World Model is a key element of our Web Agent, responsible for turning global objectives into sub-instructions based on the state of the webpage.
  • 👀 Integrating vision: The World Model now considers a screenshot of the webpage's current state when generating instructions.
  • 📑 Adding documentation: We have updated and added several new pages to the docs including Use cases and Architecture pages.
  • 🔑 Re-focusing on key features: To focus on building the best possible framework for Web agents, we have temporarily dropped or stopped maintaining the CLI module, Gradio module (pending a Web Agent relaunch) & VSCode extension (no longer maintained but still available here).

🛤️ Coming Soon

  • Improved Action Engine performance
  • Eval module for evaluating Action Engine performance
  • A Memory module implementation for our agents
  • A relaunched Gradio demo module for Web Agents

📢 Coming Soon

  • Contribute knowledge files: Knowledge files are text files containing several examples that are included in the prompt to our World Model to teach it how we want it to reason and break down a user objective into thoughts. By providing a file with several examples tailored for a specific webpage, we can boost agent performance. You can add to our knowledge hub by forking our GitHub repo and creating a PR with your files added to our examples/knoweldge folder!
Objective: Ask the AI model 'Command R plus' 'What is love'
- I am on the Hugging Face website.
- Hugging Face is a company that hosts AI models, and allows users to interact with models on them through the chat.
- Therefore, to answer the objective of asking the AI model 'Command R Plus' 'What is love', we need first to find the model page.
- Given the current screenshot, the fastest way to find the model page seems to be to use the search bar.
Instruction: Type 'Command R plus' on the search bar with placeholder "Search ..." and click on the first result

Feel free to join discussions on our Discord and check out our GitHub repo!

See you next time!