LaVague Weekly Update #6

Welcome to our sixth weekly dev update.

Here's what we've been up to:

• 🔒 Safeguards have been added to avoid arbitrary code execution. We now only allow set relevant actions to be performed.
• 🌐 A LaVague Chrome extension POC has been built - keep following our blog for the full release to come.
• 🔍 End-to-end debugging guide added to docs.

🏆 Community PRs merged:

• aaronted009 added some missing arguments to our ActionEngine.from_context method.
• shawon-majid added compatibility to pass chrome_options to our drivers.
• jashuajoy added an optional argument to our Evaluators for custom metrics display in comparison graphs.

📢 Help wanted:

• Test out the Chrome Extension POC and give us your feedback. Instructions to get started with the extension can be found here. Note, full features and UI/UX are not implemented yet.
• Integrate a local video recording option to the agent.run() method for easier debugging.
• Update the agent to return output of Python Engine directly where there is an output, rather than a regenerated output reproduced by World Model.

For a full list of our help wanted GH issues, follow this link.

Coming soon:

• 🌐 LaVague Chrome extension full release.
• 🏆 LLM & retriever evaluation leaderboard for performance with LaVague to be launched.
• 🐞 Improvements to Retriever to deal with some issues found in identifying interactable elements.
• 📊 Creation of alternative evaluation based on set actions performed on a small set of popular websites.

Thanks for following our progress and please feel free to reach out to us with any questions or feedback on our Discord!