Most chatbot UIs today feel flat and uninspiring. The future isn’t just chat—it’s agentic copilots that break out of the text box, give humans visibility and control, and quietly do as much work as possible in the background. Cedar-OS is an open source React framework that helps you build those interfaces, used by many YC companies in production.

​
What are AI-native applications?

AI-native applications are software applications that have AI workflows and agentic interfaces embedded deeply in the application itself. The agent should be able to interact with the source application in the same way the user does, and more.
AI interfaces currently have several problems that AI-native software applications solve.
  1. AI doesn’t have the context needed to do useful work.
Applications hold that context. A user might have a particular email, excel sheet, or a blueprint schematic open. The content of what the user is looking at provides a lot of context. Applications also hold an implicit context: if a user has an excel sheet, you can infer a lot about their current goal, and make agents that are proactive and meet users where they’re at.
  1. We can’t trust agents to execute sensitive/important tasks autonomously.
Imagine if an agent could merge PR’s directly to prod, or use your credit card for purchases. Even if they can be trusted 99% of the time, the 1% error would be unacceptable. But if humans have to review everything, they need to be able to effectively see what they agent is doing, what they changed, and most importantly remain in control. GUI’s are already set up for this, with auth, permissions, and functionality embedded into application logic. Think of how cursor doesn’t just change files, but shows you where and how it changes it and allows the user to granularly accept each diff.
  1. Human-AI interfaces are too hard to use.
Most users don’t even know what AI is, much less how to best prompt them. Writing out prompts can often be as tedious as just doing the task yorurself. It’s also hard to know what the agent is capable of, especially with tool calling. AI-native software should allow users can run complex agentic actions through simple commands?

​
What You’re Building With

Cedar-OS is a React framework specifically designed for building AI-native applications. It provides components, hooks, and patterns that make it easy to create interfaces where AI agents can interact naturally with your app’s state and UI. Mastra is an AI framework that handles the backend complexity of building AI agents - from LLM orchestration to tool calling and memory management. The hackathon starter combines Cedar-OS (frontend) with Mastra (backend) for a complete AI application stack.

​
What We’re Judging

You don’t need to hit every category. Pick the one that excites you most. Creative Interface – Interfaces that go beyond chat: sound, video, spatial, tactile, new UX patterns (see Spells in our docs as an example) etc. Commercial Potential – A product that could be venture scalable. Solving a real problem in a large market. Technical Depth – Complex, seamless, or cutting-edge integrations.

​
Getting Started

1

Create your project

Use the Cedar CLI to create a new project with the hackathon starter template:
npx cedar-os-cli plant-seed
This command will:
  • Create a new project with the Cedar + Mastra hackathon starter template (selected by default)
  • Set up a complete monorepo with Next.js frontend and Mastra backend
  • Include pre-configured Cedar components and basic AI features
2

Install dependencies and add API key

Navigate to your project and install dependencies:
cd your-project-name
npm install && cd src/backend && npm install && cd ../..
Create a .env file in the project root:
echo "OPENAI_API_KEY=your-api-key-here" > .env
3

Start development

Make sure you’re in the main project directory and run the development server:
cd path-to-root-project  # if not already there
npm run dev
This starts both the Next.js frontend and Mastra backend. Cedar chat should work immediately!
New to Next.js? Check out the Next.js Documentation to learn more about the framework powering your frontend.

​
Time to Explore :)

Now that your project is running, explore these Cedar-OS features that will make your hackathon project stand out:

​
Agent State and Actions

Let AI agents read and modify your React state in real-time.

​
Spells

Spells are Cedar-OS’s powerful system for creating contextual, gesture-based interactions that enhance your AI-native applications. They provide a unified way to bind keyboard shortcuts, mouse gestures, and selection events to custom UI behaviors and actions.

​
Voice

Add natural voice interactions to your app with built-in recording and AI response playback. Voice makes any interface feel futuristic and accessible.

​
For the Vibe Coders

Cedar & Mastra both have MCP servers you can hook up to Cursor and more!

​
Need Help?

If you have any questions about Cedar OS or Mastra, join our Discord community and post in the hackathon channel: Join Discord → Post in #hackGT We’ll be actively monitoring the channel and we’d love to see what you’re building!