--- title: OpenUI 1.0 wants agents to stop answering in walls of text description: Thesys released OpenUI 1.0, an open-source framework that lets AI agents respond with generated interactive interfaces (cards, maps, comparison widgets)... canonical: https://daily.dev/posts/openui-1-0-wants-agents-to-stop-answering-in-walls-of-text-cfvo5omcm twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: OpenUI 1.0 wants agents to stop answering in walls of text | daily.dev og:description: Thesys released OpenUI 1.0, an open-source framework that lets AI agents respond with generated interactive interfaces (cards, maps, comparison widgets)... og:url: https://daily.dev/posts/openui-1-0-wants-agents-to-stop-answering-in-walls-of-text-cfvo5omcm og:image: https://api.daily.dev/og/posts/CfVO5oMcm.png og:image:alt: OpenUI 1.0 wants agents to stop answering in walls of text og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # OpenUI 1.0 wants agents to stop answering in walls of text **[Trends](https://daily.dev/sources/trends)** · 2 min read · 0 upvotes · 0 comments ## Summary Thesys released OpenUI 1.0, an open-source framework that lets AI agents respond with generated interactive interfaces (cards, maps, comparison widgets) instead of plain text. Version 1.0 adds custom actions that let generated UI buttons trigger an app's own code, support for any model including open-weight ones, and streaming UI built from a team's own components rather than a fixed widget set. Thesys' own benchmark claims OpenUI uses 58% fewer output tokens than Vercel's json-render and 52% fewer than Google's A2UI across 46 interface briefs and six models, though this is a vendor-run comparison rather than an independent verdict. Early reactions from AI practitioners are largely positive, with some predicting generative UI could become a standard way agents interact with users. ## Content Thesys just shipped OpenUI 1.0, an open-source framework that lets agents answer with working interfaces instead of paragraphs. The reaction so far is mostly developers nodding along, with one pitch repeated everywhere: ChatGPT already has generative UI, and now everyone else can have it too. @rohanpaul_ai frames it as "good news for front-end engineers" and puts the idea in one line: "ChatGPT has generative UI built in, and OpenUI 1.0 brings it to every other app and agent." His point is that the UI layer isn't tied to one provider. It works with GPT, Claude, or an open model running on your own hardware. @omarsar0 is the most enthusiastic. He already renders actionable output as small interfaces in his own agent orchestrator, and says "I can easily see this becoming a standard way to interact with agents in the future." @svpino sells it with a concrete picture: a travel assistant that replies with a hotel card, a map, and an interactive comparison of plane tickets instead of a wall of text. ## How it works The division of labor is the draw. You register your own components (hotel cards, maps, charts, booking forms). OpenUI turns that library into a system prompt. The model replies in OpenUI Lang, a compact markup naming the components and filling them with data. The renderer validates that markup against your registry as it streams, then draws the screen in your own design system. As @svpino puts it: you decide the components, the model decides how to combine them, OpenUI renders. ## The numbers The efficiency claim comes from Thesys' own benchmark, so read it as vendor data. Across 46 interface briefs and six models, OpenUI used 58% fewer output tokens than Vercel's json-render and 52% fewer than Google's A2UI. Because the language is compact, the UI can render while the answer is still being generated. 1.0 also adds custom actions, so buttons inside a generated answer can trigger your app's own code. That's what moves it from pretty cards to something an agent can actually drive. The GitHub repo sits around 11K stars, per @rohanpaul_ai. ## Questions this post answers ### What new features does OpenUI 1.0 add for AI agents generating interfaces? OpenUI 1.0 adds custom actions that let buttons inside a generated interface trigger an app's own code, support for any model including open-weight ones running on self-hosted hardware, and streaming UI built from a team's own components rather than a fixed widget set. This moves generated interfaces from static displays toward functional product surfaces. _Teams weighing generative UI frameworks for their agents can follow feature breakdowns like this on daily.dev._ ### How does OpenUI's token usage compare to Vercel's json-render and Google's A2UI? According to Thesys' own benchmark of 46 interface briefs across six models, OpenUI used 58% fewer output tokens than Vercel's json-render and 52% fewer than Google's A2UI. This is a vendor-run comparison rather than an independently verified result, so it should be treated as a claim rather than a settled verdict. _Developers comparing generative-UI frameworks before committing can track benchmark claims like this on daily.dev._ ### What problem does generative UI for AI agents solve compared to chat text responses? Generative UI replaces walls of chat text with working interfaces, such as a travel agent returning a hotel card, a map, and an interactive flight comparison instead of a paragraph. The model picks from a constrained set of developer-defined components and decides how to combine them per request, which keeps output compact and prevents generated UI from going off the rails. _Developers exploring agent UX beyond chat text can follow this design pattern's progress on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 4 discussions and 32 comments across x (as of 2026-10-16). **TL;DR:** Reaction is largely enthusiastic about agents returning rendered UI components instead of text walls, with the token-savings benchmark cited approvingly, though several replies probe safety and reliability gaps like unregistered components, untrusted link props, and validation of model output. **Sentiment:** 55% positive · 35% mixed · 10% skeptical **The case for** - Generated UI components (cards, buttons) feel more actionable and usable than plain text replies. - The reported 58% reduction in output tokens is seen as meaningfully cutting cost and latency, not just a marketing number. - Streaming the UI as it's generated is viewed as a genuine improvement over waiting for a full response. - Restricting the model to a fixed, registered component set is seen as the sensible way to keep generated UI predictable. - MIT licensing is seen as likely to help adoption. **The pushback** - If the model invents a component that wasn't registered, the experience may silently fall back to a text wall. - Concern that model-filled props (e.g. link destinations) could be manipulated by injected content to point a trusted-looking button somewhere unsafe. - Unclear how predictable or safe the output is when the model freely picks which components and combinations to render. - Generating the UI is seen as only solving half the problem in enterprise contexts, since the underlying data still needs reliable sourcing and permissioning. - Skepticism that a self-reported, self-benchmarked comparison is an inherently biased way to prove superiority. **By community** - x (positive): Replies mostly welcome agent-generated UI and the claimed token savings, with a vocal subset raising practical concerns about component validation, security of model-filled props, and output predictability. **Hottest debate:** Whether constraining agents to a fixed, validated component catalog is enough to prevent unsafe or broken output, versus the risk of models inventing components or filling props with untrusted data. **Open questions** - Can developers limit which actions a generated button is allowed to call, or does the model choose freely? - Is the layout streamed progressively as it's generated, or does rendering wait for the full response? - How is predictability maintained when the model selects components and combinations itself — is there a validation layer, or do bad layouts surface only as bug reports? - Are streamed props validated against a schema before rendering? **Highlights** > @testingcatalog the custom actions are what i'd want to try. can the dev limit which actions a generated button is allowed to call, or does the model pick freely? > — [daniiie7 on x](https://x.com/daniiie7/status/2110765711983587333) > @svpino the model composing your components only works until it invents one you never registered. then you probably get the same text wall back > — [ragzoi on x · 1 comments](https://x.com/ragzoi/status/2110746397683470472) > @svpino The part I'd lock down is model-filled props. If an injected web page talks the model into putting its own URL behind the "Book now" button, users see a trusted-looking card. Allowlisting link domains per component lets the model pick layout, not destinations. > — [hackrowd on x](https://x.com/hackrowd/status/2110746005624738113) > @svpino When the model picks the components for each request, how do you keep the output predictable enough to ship? Do you constrain which combinations are allowed, or does a bad layout just show up as a bug report? > — [AthenaAI\_org on x](https://x.com/AthenaAI_org/status/2110757993591836771) > @svpino The shift from text-based responses to dynamic interfaces is compelling. But in enterprise applications, generating the UI is only half the challenge. The data behind each component still needs reliable sources, traceable evidence, and clear permission boundaries. A polished > — [OMID\_0909 on x](https://x.com/OMID_0909/status/2110774238894235656) **Source threads** - [x](https://x.com/testingcatalog/status/2110765439798087835) · 0 points · 12 comments - [x](https://x.com/svpino/status/2110743040323379705) · 0 points · 14 comments - [x](https://x.com/omarsar0/status/2110776734509359427) · 0 points · 6 comments - [x](https://x.com/rohanpaul_ai/status/2110826849249149020) · 0 points · 0 comments ## Similar posts on daily.dev - [A2UI](https://daily.dev/posts/a2ui-fov2gqezm) · Hacker News · 2 upvotes · 0 comments - [Google Releases A2UI v0.9: Portable, Framework-Agnostic Generative UI](https://daily.dev/posts/google-releases-a2ui-v0-9-portable-framework-agnostic-generative-ui-nhjivffxg) · InfoQ · 1 upvotes · 0 comments - [\[New\] Generative UI for Agents](https://daily.dev/posts/new-generative-ui-for-agents-7mg572qny) · Daily Dose of Data Science \| Avi Chawla \| Substack · 110 upvotes · 4 comments - [Generative UI in 2026: A Developer Guide That Ships](https://daily.dev/posts/generative-ui-in-2026-a-developer-guide-that-ships-didpe5oji) · Alex CloudStar · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#generative-ui](https://daily.dev/tags/generative-ui) [View this post on daily.dev](https://daily.dev/posts/openui-1-0-wants-agents-to-stop-answering-in-walls-of-text-cfvo5omcm) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"OpenUI 1.0 wants agents to stop answering in walls of text","url":"https://daily.dev/posts/openui-1-0-wants-agents-to-stop-answering-in-walls-of-text-cfvo5omcm","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/openui-1-0-wants-agents-to-stop-answering-in-walls-of-text-cfvo5omcm"},"datePublished":"2026-10-15T16:56:26.581Z","dateModified":"2026-10-16T14:03:02.046Z","description":"Thesys released OpenUI 1.0, an open-source framework that lets AI agents respond with generated interactive interfaces (cards, maps, comparison widgets)...","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/openui-1-0-wants-agents-to-stop-answering-in-walls-of-text-cfvo5omcm","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,generative-ui","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"OpenUI 1.0 wants agents to stop answering in walls of text"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/openui-1-0-wants-agents-to-stop-answering-in-walls-of-text-cfvo5omcm#faq","mainEntity":[{"@type":"Question","name":"What new features does OpenUI 1.0 add for AI agents generating interfaces?","acceptedAnswer":{"@type":"Answer","text":"OpenUI 1.0 adds custom actions that let buttons inside a generated interface trigger an app's own code, support for any model including open-weight ones running on self-hosted hardware, and streaming UI built from a team's own components rather than a fixed widget set. This moves generated interfaces from static displays toward functional product surfaces. Teams weighing generative UI frameworks for their agents can follow feature breakdowns like this on daily.dev."}},{"@type":"Question","name":"How does OpenUI's token usage compare to Vercel's json-render and Google's A2UI?","acceptedAnswer":{"@type":"Answer","text":"According to Thesys' own benchmark of 46 interface briefs across six models, OpenUI used 58% fewer output tokens than Vercel's json-render and 52% fewer than Google's A2UI. This is a vendor-run comparison rather than an independently verified result, so it should be treated as a claim rather than a settled verdict. Developers comparing generative-UI frameworks before committing can track benchmark claims like this on daily.dev."}},{"@type":"Question","name":"What problem does generative UI for AI agents solve compared to chat text responses?","acceptedAnswer":{"@type":"Answer","text":"Generative UI replaces walls of chat text with working interfaces, such as a travel agent returning a hotel card, a map, and an interactive flight comparison instead of a paragraph. The model picks from a constrained set of developer-defined components and decides how to combine them per request, which keeps output compact and prevents generated UI from going off the rails. Developers exploring agent UX beyond chat text can follow this design pattern's progress on daily.dev."}}]} ```