Excited to introduce Index - a new SOTA open-source browser agent powered by Anthropic Claude 3.7 thinking. It reached 92% on webvoyager. Use it as: - a Python package - API via Laminar - or a self-hostable Chat UI We built it by inventing a new type of observability🧵⤵️

Use Index via chat UI for free here Star our repo here Learn how to use API here

Debugging browser agents is very tricky, because it's really hard to make sense of errors only by looking at the traces. Simply put, traditional observability is not enough for browser agents.

We've added THE missing part - the ability to see what the browser agent "sees". Essentially, we record the whole browser session and perfectly sync it with agent traces. Now, you can see the whole trajectory of an agent, and if you spot that something is wrong, you know which step you have to inspect.

Then, we spent a lot of time and money on cracking the evals. Thankfully, @lmnrai is the best platform to run massive evals in parallel.

tldr: great observability and evals is all you need to build SOTA agents right now

🧠 Unified Search. Smarter Meetings. Effortless CRM. MightyBot is your AI agent platform for seamless workflows—record meetings, automate CRM updates, and find answers across apps in seconds. 🌟 Focus on what matters. We'll handle the grind.

@lmnrai You can fully self-host the chat UI, instructions here

@AnthropicAI @lmnrai Claude thinking is crazy; we’ve found really good success with it calling @Stagehanddev tools as well — are you using computer use?

@AnthropicAI @lmnrai @Stagehanddev No, raw 3.7 with extended thinking and bag of tricks. Check it out here Also, when are you gonna try out Laminar observability? don’t miss out on the best browser observability out there :)

prompt used: go to summarize first 3 companies in the W25 batch and make new spreadsheet in google sheets.

@AnthropicAI @lmnrai looking great, congrats on this one!