CopilotKit builds software that lets AI agents live inside apps and respond with actions, state changes, and interactive interfaces instead of a plain chat box. In May 2026, the Seattle startup announced a $27 million Series A. The round matters because a lot of companies are done with bolt-on chatbots that feel detached from the product itself. Founded in 2023 by brothers Atai Barkai and Uli Barkai, CopilotKit is trying to turn AG-UI — its open protocol for agent-to-interface communication — into a standard layer for enterprise software teams building agentic apps.
What is CopilotKit and how do its AI agents work?
CopilotKit is an agentic frontend stack. A developer connects an agent backend or direct model to a React app and wires in CopilotKit’s client libraries. That gets them chat and generative UI. It also adds in-app actions and human-in-the-loop workflows that stay synced with the application itself. Under the hood, AG-UI acts as the event layer between frontend and backend, using web-native transport like HTTP and WebSockets to stream state, actions, and interface updates back and forth.
A lot of teams usually build that part by hand — badly. CopilotKit’s framework includes real-time context awareness and bidirectional app-agent state sync. It also offers headless UI options for custom design systems, plus observability hooks so teams can inspect what the agent did and why. There’s also an Inspector tool for debugging interactions in real time and historically. That matters once a demo becomes a production workflow.
The most interesting bit is the UI layer. Through A2UI and MCP-style generative UI patterns, an agent can render structured components instead of dumping text into a chat bubble. CopilotKit supports both a dynamic schema approach — where a secondary model generates the interface schema and data — and a fixed schema approach, where teams predefine the component tree and let the agent stream data into it. That’s how Barkai gets from a vague prompt like “show me revenue by category” to a chart, a card, or some other interactive element that fits the app.
On top of the open-source core, the company is now pushing CopilotKit Enterprise Intelligence. That product adds persistent threads and analytics. It also includes continuous learning from human feedback, self-hosted deployment, air-gapped support, SSO, and role-based access control. In plain English: it’s the stuff a big company needs after the prototype works and security, compliance, and continuity show up to ruin the party.
Who founded CopilotKit and why did they start it?
The founding story
CopilotKit was founded in 2023 and publicly launched in 2024. Atai Barkai is CEO, and Uli Barkai runs growth. Their starting point was simple: chat-only AI inside software feels clunky, especially when the real job isn’t conversation but interaction — approving something, editing something, taking an action, or moving through a workflow with the UI reacting in real time.
That view shows up all through the company’s pitch. Atai’s core argument is that agents shouldn’t answer only with text; they should use interfaces defined by the product team. He described that shift as “not just with blocks of text, but with interactive UIs,” which is basically the whole thesis in 10 words.
Why the founders fit this market
Atai Barkai looks like the technical half of the pair. Before CopilotKit, he worked at Meta on media SDK and infrastructure problems. He also spent time at Doximity as a senior software engineer, helped build FermiCloud, and later founded tawkitAI, which produced PodcastGPT. He studied physics at the University of Pennsylvania. That doesn’t make a founder special by itself, but he’s spent years close to developer tooling, infrastructure, and productized AI.
Uli Barkai’s background is more on the distribution side. Before co-founding CopilotKit, he led marketing at tawkit and worked on PodcastGPT’s go-to-market story. His academic background runs through financial economics at Columbia and philosophy at Tel Aviv University. It’s a slightly unusual mix, but useful if your job is translating technical ideas into a market people will actually buy from.
Traction and fundraising
The early signals are strong, even if startup traction claims deserve a raised eyebrow. CopilotKit’s libraries are now reaching more than 4 million weekly downloads. Its open-source projects have more than 40,000 GitHub stars and 150 contributors. Its tooling also powers millions of agent-user interactions in production. The company says customers include DocuSign, S&P Global, Cisco, and Deutsche Telekom, while its team sits at about 25 employees.
The new money is a $27 million Series A. Glilot Capital, NFX, and SignalFire led the round. CopilotKit says the capital will help expand the enterprise toolkit around AG-UI, support self-hosted deployments, and grow the team. Open standards don’t pay the bills by themselves.
How CopilotKit compares with Vercel AI SDK and OpenAI alternatives
CopilotKit isn’t alone. Vercel’s AI SDK helps developers build AI apps. Assistant-ui focuses on interface components for chat-style assistants. OpenAI’s Apps SDK gives developers richer UI options — but only inside ChatGPT. The older alternative is even less elegant: teams bolt a chatbot onto an app, then custom-build the plumbing to connect it to backend actions and frontend state.
Its pitch is that it’s more horizontal than those stacks. CopilotKit wants to sit across whatever model provider, cloud, or agent framework an enterprise already uses. It also offers self-hosting and stricter control over how much the agent can change the UI. Atai summed that up cleanly: “enterprises want optionality and they want self-hosting.”
Why does the CopilotKit funding round matter?
Because this isn’t just a bigger balance sheet. It’s the point where an open-source protocol bet turns into an enterprise software company.
AG-UI already has adoption across major vendors and frameworks, and CopilotKit now wants to capture the paid layer on top of that usage: support and hardened deployment. It also wants persistence, analytics, and governance. If that works, the company doesn’t need to beat every model provider. It just needs to become the default frontend and orchestration layer teams reach for when they want agents inside real products.
It also matters for customers. A lot of enterprise AI pilots stall when teams hit security reviews, audit requirements, or the messy reality of multi-session workflows. CopilotKit’s new enterprise push is aimed right at that gap. That’s why features like air-gapped deployment, persistent threads, and role-based access control matter more than another flashy demo.
How big is the enterprise AI agents market?
Pretty big already, and growing fast enough to explain why investors are leaning in. Grand View Research estimates the global enterprise AI agents market hit $3.67 billion in 2025 and could reach $83.4 billion by 2033, a projected 48.4% CAGR. North America was the largest revenue-generating region in 2025, which lines up with where most enterprise software budgets and AI infrastructure vendors are concentrated.
The timing also makes sense. Developer tooling for agents is fragmenting into protocols and interface layers — AG-UI for frontend-backend communication and A2UI for declarative UI rendering. There are also MCP-style app patterns for tool use, plus a growing list of supported frameworks from LangGraph to Google ADK and Microsoft Agent Framework. That sounds wonky. It is. But it’s also the kind of plumbing layer that becomes valuable once teams stop treating AI as a novelty and start embedding it into core software flows.
Final take on CopilotKit funding
The easiest way to read CopilotKit funding is as a bet on the layer between the model and the screen. Not the model. Not the chatbot shell. The connective tissue that lets an agent understand what a user is doing, take action inside the app, and present something usable back. The next thing to watch is whether CopilotKit can turn protocol adoption into durable enterprise standardization before larger platforms absorb the same territory.
Read how a16z crypto raised a $2.2B crypto-only fund to back blockchain and web3 startups, doubling down on long-term crypto infrastructure and digital asset investing even as AI pulls venture capital away from the sector.
FAQ
– What is the CopilotKit funding round? CopilotKit raised $27 million in a Series A announced on May 5, 2026. Glilot Capital, NFX, and SignalFire led the round, and the company says the money will go toward expanding its enterprise product and hiring beyond its current team of about 25 people.
– How does CopilotKit work inside an app? CopilotKit gives developers a frontend stack that connects an AI agent to the app’s own state, actions, and UI components. Instead of returning only text, the agent can trigger workflows and stream updates. It can also render structured elements like cards or charts through AG-UI and A2UI-style generative UI.
– Who founded CopilotKit? CopilotKit was founded in 2023 by brothers Atai Barkai and Uli Barkai. Atai came in with engineering experience from Meta, Doximity, FermiCloud, and tawkitAI, while Uli had already worked on marketing and growth at tawkit before helping build CopilotKit’s commercial side.
– Is CopilotKit part of the enterprise AI agents market? Yes — it sits in the enterprise AI agent infrastructure layer, specifically around agentic frontends, generative UI, and production deployment tooling. That’s a category Grand View Research sized at $3.67 billion in 2025, which helps explain why investors are backing platforms that make agents usable inside business software instead of just chat windows.
