We use a lot of open source to build @thesysdev
Given our experience maintaining OpenUI, we know how much time and effort goes into the projects the whole internet quietly depends on.
Today we're taking one small step toward giving back - sponsoring pnpm, one of the core build tools we rely on every day.
What project should we support next?
We took a long time before we decided to open source OpenUI
Not because we didn't want to but because we realized that once we open sourced it and people start depending on it, it would be our duty to maintain it and ensure its the best in class as long as @thesysdev exists
Since the field of Generative UI barely existed when we began we had to move fast and break things before we knew this is the right abstraction. In that moment it didn't feel right to open source it as we were still figuring out what "it" means
Even today we are very careful in what we open source vs what we don't so that we never have to pull this shit.
Looking forward to this!
We’ll be in Tokyo on July 2 talking OpenUI, Generative UI, and AI-native interfaces. Come say hi if you’re around.
Registration link: https://t.co/MURoxktz2f
Also SSR is quite unnecessary in most applications - the only good non marketing use case I can think of is ecommerce.
For everything else SPA served over CDN is the right choice
@badlogicgames 's PI is the harness of choice for full control and zero bloat. Terminal-native, text-pure.
We just took it out of the terminal.
Pi + OpenUI - your agent now responds with real interactive UI in the browser.
Charts, forms, dashboards, streamed live.
Know an awesome developer? Refer them and win a Mac Studio
At @thesysdev we are hiring across multiple roles. If you know someone cracked, send them our way and walk away with a Mac Studio to build your own personal AI lab!
Does @grok bott have even the most basic context window
> what's Le chaton fat joke
>> *gives an answer*
> How it did start
>> Goes off on a philosophical detour of how did the universe start
Got a chance to represent @thesysdev and OpenUI at the Agent Harness Hackathon as a sponsor judge! Amazing to see brilliant use cases - from FDA drug analysis to fun projects like using OpenUI to make music!
AWS loft served some amazing 🍕
@metrics_co - the folks powering portfolio analysis for 150+ VC/PE firms, GC, Accel, 8VC - just wrote a blog post on how they shipped Generative UI within their AI Analyst and why they chose OpenUI over JSON / Raw HTML
The blog post is a treasure trove of information with good architectural patterns on how they integrated Generative UI capabilities within their agent that already shipped markdown without breaking backward compatibility
Go read!
https://t.co/o1SaOLHQ8E
Most debates about Generative UI are actually two decisions being confused as one.
1. Transport: where does the UI appear? (AG-UI for your app, MCP-Apps for ChatGPT/Claude)
2. Generation: what does the agent emit? (Static, Declarative, or Open-Ended)
Any pairing is valid. Once you separate them, most disagreements dissolve.
Wrote a full report covering the protocols, the formats, the token benchmarks, and everything you should consider before choosing your stack.
A startup just dropped a free open-source spec that generates UI components with 67% fewer tokens than JSON.
It's called OpenUI. One protocol. No SDK lock-in. No proprietary runtime. You plug it into any LLM and the model starts shipping clean UI instead of bloated JSON blobs.
Every "AI generates UI" framework before this made you pick one trade-off. Spend the tokens and get rich components. Compress the output and lose half your interactivity. Skip the schema and ship broken renders. OpenUI does all of it inside a single open spec.
Here's what makes it different from every LLM-to-UI pipeline that came before:
→ 67% token reduction vs JSON for the exact same rendered component, which means your bills drop by two-thirds on every UI generation call
→ Model-agnostic, works with Claude, GPT, Gemini, DeepSeek, Llama, anything that can output structured text
→ Component-first instead of element-first so the LLM thinks in buttons, cards, and forms instead of nested div soup
→ Streamable by design, components render as the tokens arrive instead of waiting for the full response
→ Framework-neutral output that compiles to React, Vue, Svelte, or plain HTML without rewriting the prompt
→ Built-in primitives for forms, charts, tables, modals, and dashboards so you stop reinventing the same 40 components every project
→ Type-safe schema that catches malformed LLM output before it ever hits the renderer
→ Zero runtime dependency, the spec is just text the model already knows how to produce
Killed: $200/mo "AI UI builder" SaaS, every prompt-to-React wrapper sitting on top of a bloated JSON schema, the 4000-token system prompts trying to teach GPT what a button is.
Works with every major LLM provider. No API key lock-in. No vendor.
MIT License. 100% Opensource.