We built Yuan because we believe that when a conversation with AI ends, you should end up with something.
Over the past year almost everyone has felt how capable AI can be: it helps you sort out an itinerary, organize your thinking for a candidate review, or talk through the logic of a small tool. But once the conversation ends, what you're left with is usually just a chat log. You can't really use it, share it, or pick it up again tomorrow.
We've gradually come to think the models are smart enough; it's the product shape that has room to go one step further. Yuan is our attempt at that.
Four problems we kept noticing
First, a chat box leaves nothing behind. It's a great surface for "conversation" and a poor one for a "product": no data state, no separate sessions, no link you can hand to someone else. What you made with AI today is scattered through your history by tomorrow. We made this the core thing to solve: orchestrate AI with app logic, so results can be shared and reused.
Second, an agent that "can do anything" is one you dare hand nothing. When an AI can read your data, casually rewrite it, and call external APIs all at once, and you can't tell which step is deterministic and which is the AI improvising, you stop trusting it with anything that matters.
Third, you build something but can't get it actually running. For people without much technical experience, "deploy it, ship it, keep it running" is its own wall: environments, servers, domains, any step can lock you out. Yuan takes that part away. Once it's built you use it and share it directly, with no last-mile headache.
Fourth, getting the result you want often takes precise logic. Something like "pull in a data source and chart it in a specific format" sits on top of program logic that has to be written exactly right, a high bar for anyone without a coding background. That part can go to the AI Agent built into the app: you don't have to understand code like a programmer to build something genuinely useful.
Our approach: you build your own AI Native app
What we want to create is a visual AI Agent, an AI Native app. It belongs to a different category from chatbots and website builders.
Put simply, you get a web app with an AI Agent fused into it: one that can read the app's data, call its operations, and be published as a link you can share.
Here is what makes an app AI Native in our definition:
- The AI Agent lives inside the app. AI capability and app logic are natively fused, with the Agent woven into the UI in many forms.
- It runs around "sessions." Each trip, each stock-take, each review is its own session, with its own data state and AI memory, fully separate.
- It's a real app. With data tables, charts, a UI, and an address you can send out.
Build a real app in one sentence
We want the bar for creating to be as low as a single sentence: describe the app you want in plain words, and Yuan decides whether you need data tables, charts or an AI assistant, then publishes it for you. No technical background required.
You could build things like:
- A daily finance brief: every morning at nine it pulls the latest world finance news, AI summarizes and interprets it, you browse by category, and the important stories land straight in your inbox. Set the schedule once and it runs itself every day.
- AI Gomoku: the AI is your opponent. You place a stone, it thinks, plays its move, and explains the reasoning behind it. App logic orchestrating AI is an experience a chat box can't reproduce.
- A trip planner: a trip planned in ChatGPT is gone once the chat scrolls away. In Yuan every trip is its own session, with itinerary, budget and AI memory all kept, and shareable with your travel companions.
- An internal HR review: connect your company's employee data, track headcount and moves by department, and let AI write the biweekly report and the attrition-risk review. We've always believed AI shouldn't be a privilege reserved for "super-individuals"; everyone should be able to put it to good use.
All four cases have demos on our home page. And do browse the Gallery to see what people have already built.
How we make it trustworthy
Letting people hand work to AI with confidence comes from a few deliberate engineering choices; a smarter model alone doesn't get you there. This part is a little more technical, but it explains why a Yuan app won't "work today and break tomorrow."
Deterministic first, AI last. A generated app behaves, wherever possible, like a hand-written ordinary app. The built-in AI Agent is one capability among many, reserved for work that genuinely needs reasoning: conversation, summarizing, classifying. What should be deterministic stays deterministic; only what should be left to AI is.
AI can read, but not write freely. Every operation is classified by effect: only read-only operations are exposed to the AI as tools, while operations that write data or change the business run only from fixed flows you've defined. Which step is deterministic and which is the AI improvising stays clear.
Versioned, and you can return to any version anytime. Every round snapshots a version; what's live is the version you published. A broken draft doesn't affect production, and returning to any past version is one click, without deleting any later history.
After it's built
Making an app is only the beginning; we also care that it keeps working over time and that you can genuinely hand it to someone else:
- Ding! One click to let AI take over. Can't quite describe it, and edits keep missing the mark? Press Ding! It re-reads your conversation and the current app, diagnoses the problem itself, makes the changes itself, then actually runs your app to verify the result. You just wait for the Ding!, like sliding cookies into the oven.
- Connect to external systems. Ready-made MCP connections plug your app into outside services: send email, post to Discord or Telegram, fetch weather and crypto prices, hook up Notion or Airtable; add your key and it works. Or use a Skill to teach the app your domain knowledge and your team's private APIs.
- Scheduled runs. Give an app a schedule and every day it fetches data on its own, has AI organize it, and writes it into your tables. That's exactly how the finance brief above runs, with nobody opening it.
- Share what you make, casually. Keep an app private, share it by link, or publish it to the Gallery; you can even turn one session's results into a read-only snapshot link and send just that.
- Themes and languages. Switch between visual styles in one click, and publish your app's content in multiple languages.
Build one for free first
Yuan bills by the tokens the AI actually consumes. You get free credits on signup, no card required, enough to build your first app. Every model tier has a free daily token quota that resets each day, and monthly credits are only tapped once the day's quota runs out, so light daily use barely costs a thing. One more thing we care about: when other people use the apps you've published, their AI usage draws from the platform's public pool, not from your balance.
Say what you have in mind, and minutes later you can use it and share it.
Say it in one sentence. Get your own AI Native app.
FAQ
Q: How is Yuan different from an ordinary AI chatbot?
A: A chatbot gives you a conversation that scatters once it's over. Yuan gives you a real app: data, a built-in AI assistant, separate sessions, a shareable link, and the AI Agent fused into the app itself.
Q: Can I use it without writing code?
A: Yes. You just describe what you want in plain words, and Yuan decides which data tables, charts and AI capabilities you need and publishes it for you. No technical background needed.
Q: Are AI-generated apps reliable? Will they break over time?
A: A Yuan app follows "deterministic first, AI last," and every version is snapshotted so you can return to any point in history. It won't work today and break tomorrow.
Q: Can other people use what I make, and who pays?
A: Yes. One click to publish and share; when others use an app you've published, their AI usage draws from the platform's public pool, not from your balance.