DefiKids

Building a better world for kids

[DefiKids @defikids_]

@nathan_tarbert @X @mastra Nice!!
Just starred ⭐️ the repo!

[DefiKids @defikids_]

Open Teach just launched 🤯

[Atai Barkai @ataiiam]

Introducing 𝙾𝚙𝚎𝚗 𝚃𝚎𝚊𝚌𝚑

Better Claude Teach that works with any agent, any model, and with your own data.

Features:
-> Embed Open Teach into your own custom agents
-> Learn from in-browser activity and from agent interactions
-> Self-hosted for full data sovereignty
-> Future proof and cross-stack

Get early access: https://t.co/YBEBxkzkbO

[CopilotKit🪁 @CopilotKit]

Voice input, now in CopilotKit 🎙️

Speech-to-text is built into the chat. Your runtime transcribes the audio and passes it to any AG-UI agent.

Works with any provider. OpenAI, Deepgram, AssemblyAI, or your own.

Docs ↓
https://t.co/uWwgp4sngp

[Nathan Tarbert @nathan_tarbert]

We built an internal tool that immediately started saving us money on API costs. Now it's open source.

LLMock is a deterministic mock LLM server—works exactly like OpenAI or Claude, but local and free. Perfect for testing demos without burning tokens on flaky tests.

`pnpm install @copilotkit/llmock`
https://t.co/zW8yipwtV7

[Nathan Tarbert @nathan_tarbert]

This was fun to put together- a quick tutorial on the new Vibe-Coding MCP Server

Powered by @tadata_team, our latest step forward in making agentic application development 10x faster!

In this walkthrough, I cover:

- Give coding agents structured access to docs + code examples from CopilotKit’s repos
- Trigger an action and show it in real time
- Expose live state so agents always stay in context
- Use a CopilotKit pre-built, drop in chat component
- Generate clean, high-quality code directly from the server

By combining MCP with @CopilotKit, the Vibe-Coding Server removes the friction of tagging docs or wiring context manually.

Agents fetch exactly what they need, respond faster, and stay tightly aligned with your app.

If you’re curious about how to vibe code your first agentic app- this is the perfect place to start!

Watch the full tutorial here: https://t.co/RKdXQ6mdm1

[Nathan Tarbert @nathan_tarbert]

The look of someone who just saw their AI agent update the UI in real time without rewiring their entire app

⭐️@CopilotKit ⭐️

[CopilotKit🪁 @CopilotKit]

Why structured actions beat raw prompts

LLMs are great at suggesting things, but struggle to execute anything predictable or safe within a real product.

Copilot Actions: a structured way to define what your assistant can do, and how it should do it.

You define:

- The schema for input/output
- The logic it triggers
- Where and how it runs in the app

The LLM calls these like functions. It doesn't guess, it invokes.

This is what turns a suggestion engine into an actual product copilot.

Copilot Actions: https://t.co/YpLkNTvA0n

[DefiKids @defikids_]

@nathan_tarbert @benln @iangcarroll @levelsio @dannypostma @thedankoe @marclou @BrettFromDJ @thejustinwelsh @nevodavid_ +1

[Igor Lessio - Robots/acc - AIFlow Labs @AIFlow_ML]

Will be integrating @CopilotKit in all my new @AgnoAgi and @NVIDIAAIDev agents workflows from now on.

[CopilotKit🪁 @CopilotKit]

How should an LLM trigger things inside your app?

- CopilotKit gives you 𝘂𝘀𝗲𝗖𝗼𝗽𝗶𝗹𝗼𝘁𝗔𝗰𝘁𝗶𝗼𝗻:
- Define actions in your frontend
- Map them to natural language triggers
- Let the agent call those actions programmatically

This gives your agent real power to interact with your UI safely and predictably.

→ Docs: https://t.co/WpwLCImwAQ

[Think critically and un-boil your inner frog @dihaynes]

Tonight’s intro to the AG-UI Protocol was a nice way to make up with my neglected Linux laptop. Good vibes have me looking forward to diving in soon!

[CopilotKit🪁 @CopilotKit]

Thanks for everyone who came out to the AG-UI workshop tonight 🦾

[CopilotKit🪁 @CopilotKit]

👾Introducing the AG-UI CLI

An easy flow that guides you through the process of building AG-UI powered applications.

Pick your agent, client, and get started with a simple starter repo with all the building blocks you need.

Get started📟
npx create-ag-ui-app@latest

[Nathan Tarbert @nathan_tarbert]

5 years ago, I was working nights and learning JavaScript from scratch.

Today, I support devs from solo builders to Fortune 100 teams building with AI.

The learning curve was brutal, but worth it.

If you’re in that stage now, don’t quit.

You’re not behind. You’re building momentum.

[Nathan Tarbert @nathan_tarbert]

One command to rule them all! 🧙‍♂️

npx create-ag-ui-app

[CopilotKit🪁 @CopilotKit]

One of the most important, but least talked about, problems in agentic UI design:
𝗛𝗼𝘄 𝗱𝗼 𝘆𝗼𝘂 𝗸𝗲𝗲𝗽 𝗔𝗜 𝗮𝗻𝗱 𝘆𝗼𝘂𝗿 𝗮𝗽𝗽 𝗶𝗻 𝘀𝘆𝗻𝗰?

A traditional UI has clear state
--> React manages it
--> State management stores it,
--> The user interacts with it.

But when you introduce an LLM into your frontend, the complications multiply.

That's where CopilotKit shines, and it manages the complexity of shared state effectively.

You can expose React state directly to the LLM. The LLM always knows the latest value.

And it’s bidirectional:

- Your UI state updates → LLM sees the new value immediately
- The LLM generates an action → It can update the UI state directly

That means you can build interfaces where AI and the user truly collaborate on the same canvas, in real-time.

Think of it like `useState()`, but for humans and AI to share.

If you're building AI-native apps, sharing state is crucial for a seamless user experience.

Learn more about how CopilotKit handles shared state: https://t.co/cahK4ib2TP

[CopilotKit🪁 @CopilotKit]

Are the rules of UI quietly being rewritten?

Traditional buttons, dropdowns, and forms are being replaced by something more fluid: 𝗻𝗮𝘁𝘂𝗿𝗮𝗹 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 + 𝗶𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝘁 𝗰𝗼𝗻𝘁𝗲𝘅𝘁.

Our teammate at CopilotKit, @RanST91, published an insightful analysis:

"The Death of Buttons: How AI is Killing Traditional UI"
https://t.co/lTQcyJr9Op

It breaks down why "prompt as UI" isn't just a trend, it’s the beginning of a deeper shift in how users and software communicate.

Ran covers:

- Why rigid UIs can’t keep up with real-world complexity
- How agents make decisions with you, not for you
- What it means to replace "clicks" with "conversations"

If you’re building interfaces with AI agents, you'll enjoy this article.

Post your thoughts below.

[CopilotKit🪁 @CopilotKit]

Solving the M×N Problem for AI Agent Interactions

Agent frameworks are moving at the speed of light, but integrating them with every possible client interface has become an M times N nightmare.

Think of the dozens of frameworks out there, and then the countless clients they might need to connect to (ei, React front-ends, Slack, SMS, WhatsApp, and beyond).

Without a standard, every framework has to build a custom integration for every client, and vice versa. That’s M×N complexity.

[CopilotKit🪁 @CopilotKit]

CopilotKit created the AG‑UI protocol to address an internal need for seamless communication with any agent framework.

We chose to open source it, build in public, and collaborate with contributors to help drive broader adoption and improvement.

You can build apps where the UI streams:
- tokens
- tools
- state
- handoffs

And even supports user interruptions (Human in the Loop)-all with just HTTP events.

Start building in minutes with one command:
**npx create-ag-ui-app**

[CopilotKit🪁 @CopilotKit]

Copilots don’t need to return just text.

With CopilotKit’s Generative UI, your agent can return UI components directly:

- Rich media
- Visual previews
- Buttons that let the agent know how to proceed
- Interactive dashboards that allow for real-time data
- Charts that enhance data visualization

This opens up a new interaction model-agents that present rich, structured responses embedded in your app.

→ Explore the docs: https://t.co/srLtxKIpbr