For the last couple of months, we've been exploring how to make agentic AI development more approachable for web developers.<p>Here's what we have, so far:<p><pre><code> import { Agent } from "react-agents";
const NoteTaker = () => {
return (
<Agent>
<Prompt>
You are a note-taking assistant that helps organize and summarize information.
</Prompt>
<Memory type="vectorstore" />
</Agent>
)
}
</code></pre>
It's somewhat inspired by [Priompt](<a href="https://github.com/anysphere/priompt">https://github.com/anysphere/priompt</a>): instead of rendering a user interface, our framework renders a prompt + action interface for AI agents.<p>The renderer is backed by a stateful agentic runtime loop, with a library of components you can import to add capabilities like TTS, database and social media integration. We want the experience to feel nice and webby like Vercel, but for AI agents.<p>For those interested in trying it out:<p><pre><code> npm install usdk -g
usdk create
</code></pre>
We're particularly curious about:
- Does this approach feel natural to React developers? Does it help you transition to agentic AI?
- DX roadblocks we should fix for our next release
- Integrations you'd like us to add (We already have Discord, X, and we're working on Twitter Spaces)
- Different environments this can integrate with (we've tested local, electron, browser, and server)<p>The source is open and we're actively working on improvements.<p>Links:
- [1] Homepage: <a href="https://upstreet.ai" rel="nofollow">https://upstreet.ai</a>
- [2] Documentation: <a href="https://docs.upstreet.ai" rel="nofollow">https://docs.upstreet.ai</a>
- [3] Open-source GitHub repository: <a href="https://github.com/UpstreetAI/upstreet-core">https://github.com/UpstreetAI/upstreet-core</a>