Thesys

Your AI agent doesn't have to look like it was built in a weekend. We added new presets to Thesys Agent Builder! Now build agents that look like ChatGPT, Claude, Perplexity, or fully custom. Skip the frontend. Pick a preset. Ship your agent. 🚀 Try Agent Builder in link in comments.

Every time Claude ships something, our pitch gets simpler. Let me explain : ▶️ We launched Generative UI. We had to explain what Generative UI even means, how models shouldn’t return text but real UI components, how interfaces can be generated in real time. ❇️ Claude gets generative UI. Now we just say: “It’s in Claude. Want it in your product? Use Thesys.” ___ ▶️ We launched artifacts. We had to explain reports, slides, structured outputs inside an AI assistant, how responses can become usable deliverables. ❇️ Claude gets report-slide generator. Now we just say: “Claude can do this. Want it in your product? Use Thesys.” ___ ▶️ We launched GenUI dashboards. We had to explain why filters shouldn’t hit the LLM every time, why refinement shouldn’t cost tokens, why UI should persist and so on. ❇️ Claude launched dashboards. Now we just say: “See Claude’s live artifacts. Want it in your product? Use Thesys.” ___ There’s a pattern here. Claude is becoming one of the best designed AI products right now. But Claude ≠ Anthropic LLMs. Just because Anthropic, the company builds something in Claude, the app doesn’t mean you can plug it into your product with the LLM. You still have to build it.   And our pitch went from minutes to seconds. “If you want your product to be like Claude, you don’t need to build it from scratch.” That’s what Thesys does for you.

Traditional UIs wait for you to decide the next step. Agentic UI acts with you. We break down the full stack, from GenUI protocols to proactive design principles to real use cases like smart dashboards, adaptive onboarding, and automated QA. 🔗 Read the full guide in link in comments.

OpenUI is out in the wild and y'all showed up 🫶 We've been reading every tweet, every dm, every late-night build post. Watching what you're making with it is genuinely the best part of shipping. Keep cooking. We're watching 👀

Agent Builder now supports Snowflake, Supabase, MongoDB, Databricks, and more. Skip the manual exports. Skip the middleware. Just connect your database right inside Agent Builder. Describe what your agent does, hook up your data source, and deploy it. Then simply ask questions and get back insights as interactive UI, generated in real-time. Try out Agent Builder in link in comments.

Dashboards as you know them are done. And we're here for it! 🚀🙌

Nobody wants to copy-paste AI text into a slide deck. So we made it unnecessary. Build AI apps that skip the text dump and respond with interactive UI, including stunning, structured slides, generated right inside the app. Generate slides with a prompt. Edit them with follow-up prompts, make inline tweaks, and export as PDF or PPTX. All without leaving the conversation! Try it in link in comments.

Six months ago, asking a CloudThinker agent "show me Q4 EC2 spend by region" meant watching a spinner for 30-40 seconds. The agent would call a dashboard() tool, fill in a strict Pydantic schema field by field, wait for the entire JSON tree to validate, and then boom — a fully-formed dashboard would snap into place. Beautiful. $0.50 per report. Slow enough that most users tabbed away to Slack and forgot what they'd asked 😅 We'd been dreaming about a "tiny language for UI" for a while — something where the agent didn't have to emit a giant JSON tree, just a few lines of compact grammar. Cool idea, no time to build it. Then in early 2026 we looked up and the whole industry had walked into the same idea: Generative UI had become a real category — Vercel AI SDK, LangChain, Thesys C1, CopilotKit AG-UI, OpenUI Lang. We adapted the OpenUI vocabulary to CloudThinker's component palette and built our Artifact system on top. Results: 30-40s down to under 10s, $0.50 down to ~$0.08, streaming component by component, one DSL body that renders on React and PDF. But the part we treasure most isn't the numbers. It's a counterintuitive insight: "Constraints — not freedom — are the whole point of Generative UI." Give an agent freedom to write HTML and it'll inject XSS, break layouts, drift from your design system. Give it a small, rigid vocabulary and everything snaps into place. Both of our architectures worked for the same reason — they only differed in format, never in contract. The blog walks through both architectures, the two detours we deliberately skipped (matplotlib and HTML freestyle), and five production lessons ✨ 👉 https://lnkd.in/gZnV_FCA Nhat Tran Khải Trịnh Xuân #GenerativeUI #AIAgents #LLM #CloudThinker #FinOps #DeveloperExperience #AIEngineering #DSL #Pydantic #BuildingInPublic