Agentboard: Browser Playground for Open-Source AI Agents

Recently launched Agentboard is an easy-to-use playground for open-source AI agents in browser. You can use it to upload your files and analyze your data, extract audio from YouTube videos, and much more.

"I built this because I noticed that many AI agents required a bit of local setup to get working correctly, and for the less technically minded it was basically impossible to set up," says Aamir Rasheed, the creator of Agentboard. "I wanted to make something that was as easy to use as ChatGPT."

Behind the scenes, it runs Open Interpreter on E2B sandboxes, allowing every user to have their own Agentboard session with file storage and internet access. It’s available (for free) on desktop or mobile.

What you can achieve with Agentboard

You can start a conversation via the chat interface or try the following examples:

Find the current local time in Seattle
Convert an example .jpg image to .png, then download it
Extract the audio from a YouTube video

One of the coolest features is the ability to upload your PDF file and let the AI agent analyze the data you provide or plot data visualizations based on your input. This is possible because Open Interpreter is able to install all the necessary packages and execute code for all the steps for the given task. These steps are executed in the secure, cloud-based sandbox.

Behind the scenes

Agentboard is built with:

NextJS + Vercel AI SDK for frontend
Google Cloud for server infrastructure
E2B as a code execution environment for the LLM-generated output.

“E2B is absolutely essential to Agentboard,” shares Aamir. “The API makes it super easy to spin up and manage microVMs. It would have been very difficult to do that on my own.”

In the first version of Agentboard, you can try the popular Open Interpreter framework. The way Agentboard is built allows for adding more AI agents in the future without being limited by the current tech-stack. The processes are executed in the E2B sandbox, which is agnostic towards the choice of LLM and works especially well with AI code interpreters and coding AI agents.

Aamir shared his thought process on building the underlying architecture which is running locally and makes Agentboard super fast, compared to alternatives he has tried before.

"Here's the new architecture I'm implementing now. There are only two hops to the OpenAI API, which reduces latency. The chat happens through the server, and the code execution is delegated to E2B." describes Aamir. "I am using the E2B integration into Open Interpreter."

Meet the author

Aamir Rasheed's background is in Computer Science which he studied at Stanford University. After that, he worked in the field of autonomous vehicles and then left to explore the space of AI agents and build something on his own. He built Agentboard as a way to understand open-source AI agents.

"While Agentboard is still in its early version, I wanted to use this project to learn how to deploy and manage AI agents in the cloud. There are so many cool things AI agents can do (research, code writing, VA replacements, etc) but they're all built on deploying and using them online." shares Aamir.

Follow Aamir on X (Twitter), read his blog, or connect with him via his personal page to stay updated.

Want to bring attention to your project?

Write us at hello@e2b.dev or make a pull request to the E2B cookbook. We have a strong preference for accepting projects such as LLM-powered code interpreters or coding AI agents. We are ready to reward and promote more good quality projects, given that your app uses E2B sandboxes.

Check out other open-source community examples like