xkuang/e2b

Vercel for AI agents. We're a platform for developers that helps them with building, deploying, and monitoring AI agents.

Open-source platform for building & deplying AI agents - "Vercel for AI agents"

Website | Discord | Twitter

e2b or etob (english2bits) allows you to create & deploy virtual software developers. These virtual developers are powered by specialized AI agents that build software based on your instructions and can use tools.

Agents operate in our own secure sandboxed cloud environments that's powered by Firecracker.

e2b currently supports building only REST servers in Node.js. Specifically using the Express framework. We'll support more use-cases with time.

🚀 Get started

We're working on the cloud-hosted version. In the meantime, the fastest way try out e2b is to run it locally via Docker.

🐳 Start e2b with Docker

You will need:

• OpenAI API key (support for more and custom models coming soon)
• Docker
• Node.js 16+
• Free ports 3000 (Next.js app), 54321 (Supabase API Gateway), 54322 (Supabase Database)

To start e2b run:

npm start

Then open page on http://localhost:3000.

npm start starts local Supabase in the background - to stop it you have to run npm run stop.

💻 Development setup

For developing with hot reloading and contributing to the project you may want to run the app locally without Docker (npm start command).

Follow these steps to set it up.

Features

🛠 Bring your own X

While e2b will offer the "batteries-included" solution, our goal is to let users:

• BYOM - Bring Your Own Model
• BYOP - Bring Your Own Prompt
• BYOT - Bring Your Own Tools

🤖 Supported models and model hosting providers

• OpenAI
  • GPT-4
  • GTP-3.5
• Anthropic
  • Claude v1.3
  • Claude Instant v1
• Replicate
• Hugging Face
  • Inference API
  • Inference Endpoints
• Banana
• Azure OpenAI

Model or model hosting provider you like isn't supported?

👉 Please open the "New model request" issue 👈

👉 Or open a PR and start contributing 👈

👀 Early demos

• AI Agent using coding tools
• Build your custom "Just-In-Time" UI
• Agent coded a full Stripe customer checkout by following a technical spec provided by user

ℹ️ Community & Support

• Discord - live discussion and support
• GitHub issues - for reporting bugs
• Twitter - to stay up to date

🤝 Contributing

We welcome any contributions! If you want to contribute to the project check out the contibution guide and join our Discord.

📆 Short-term Roadmap

1. ✅ Make sure people can run e2b locally without issues and the DX is smooth.
2. 🚧 Add support for more models.
   • ✅ OpenAI
   • ✅ Anthropic
   • Model hosting providers:
     • ✅ Replicate
     • ✅ Hugging Face Inference API
     • ✅ Hugging Face Inference Endpoints
     • ✅ Banana
     • Paperspace
3. 🚧 Improve agent's reliability and quality of output.
4. 🚧 Improve the feedback loop (UX/UI) between the developer and agents.
5. 🚧 Improve agent's understanding of the context based on the instructions.
6. 🚧 Deployable agents.
7. Support more tools and 3rd party integrations that agents can use.
8. ✅ Let users edit prompts.
9. Let users customize tools and build custom workflows for the agent.
10. Release cloud version.