Today, we're shipping OpenUI Dashboards.
Your users are taking their data to Claude to build dashboards. Now your agent can build them - right within your product. Your own components and APIs, any model.
Claude's best feature, now for every agent. https://t.co/HGmq1R6Gcg
Summary
@rabi_guha announces OpenUI Dashboards. The pitch: users already export their data to Claude to build dashboards. Now a product's own agent can build them inside the product, using the product's own components and APIs, with any model.
It matters because it keeps dashboard building native. Output follows the host design system instead of whatever code a model writes. In replies, the author says dashboards are live: they refresh with new data without calling the model again.
The thread
1. **Launch post.** OpenUI Dashboards ships. Agents build dashboards in your product, with your components and APIs, on any model. Includes a 1:07 demo video.
2. **Link post.** Points to the product page at openui.com/dashboards.
Replies worth keeping
**@rabi_guha:** Design system grounding is a core principle of OpenUI, so output uses the developer's own components.
**@rabi_guha:** Dashboards are live. They support sort and filter. A page refresh days later pulls fresh data without involving the model. There is also an auto refresh setting the developer configures.
**@rabi_guha:** Custom functions are written by the model and run in a server-side sandbox. They can chain and compose any MCPs you have.
**@rabi_guha:** On why not raw React: React needs strict sandboxing, has a large token footprint, renders slowly, and does not follow your component library.
**@rabi_guha:** It can't render inside Claude itself. MCP apps exist but don't give enough control over dashboard rendering. Layout customization is "as much as you want" and fully developer controlled.
**@karansingh010:** Asked whether dashboards can be saved and edited, and wants to replace embedded Grafana in their app. The author said "Yes and yes" and moved to DMs.
**@Arpit39036248:** Asked about group-by and joins across two APIs, and whether the model is like a code interpreter that outputs your components instead of a PNG. The author answered the first part only. The second was not answered.
Key Takeaways
Consider it if your users already export data to a chat model to build dashboards.
Output is grounded in your own design system and components, which avoids off-brand generated UI.
Dashboards stay live after creation, with developer-set auto refresh and no model call per refresh.
Data logic is generated as sandboxed server-side functions that can compose multiple MCPs.
Check the open questions yourself: joins, group-by, and saving or editing dashboards were only briefly addressed.
Links
[OpenUI Dashboards](https://www.openui.com/dashboards): product page
[Launch post](https://x.com/rabi_guha/status/2118376353750045066): original post with demo video