← Back to feed

CopilotKit raised $27 million in a Series A on May 5, led by Glilot Capital, NFX, and SignalFire. The round combines $20 million in new capital with $7 million in a previously unannounced seed. The company has roughly 20 employees and is headquartered in Seattle.

CopilotKit is the company behind AG-UI (Agent-User Interaction), the open protocol that defines how AI agents communicate with software applications. Since launching the protocol roughly a year ago, it has been adopted by Google, Microsoft, Amazon, Oracle, LangChain, Mastra, PydanticAI, Agno, AG2, and LlamaIndex.

The Protocol Stack Picture

AG-UI sits in a specific layer of the emerging agentic infrastructure stack. MCP (Model Context Protocol, from Anthropic) handles agent-to-tool connections. A2A (Agent-to-Agent, from Google) handles multi-agent coordination. AG-UI handles the third leg: agent-to-human-user interactions inside software applications.

The distinction matters for implementation. Agents don’t behave like traditional software. They stream, pause, resume, wait for confirmation, and persist across sessions. Conventional UI frameworks were designed for request-response patterns. AG-UI is designed for the streaming, stateful, interruptible flow that agentic systems actually produce.

Current capabilities in CopilotKit’s stack include generative UI (agents rendering their own interface components at runtime), human-in-the-loop workflows (agents pausing to request user input or confirmation), shared state between agent and UI, and MCP Apps integration.

Adoption Numbers

• 4 million weekly downloads across CopilotKit and AG-UI open-source SDKs
• 40,000+ GitHub stars
• Named enterprise customers: Deutsche Telekom, DocuSign, Cisco, S&P Global, Function Health
• Fortune 500 and Global 50 customer base (majority, per company)

All four major hyperscalers have adopted the protocol. That’s meaningful not because hyperscalers will use CopilotKit’s specific tooling, but because protocol adoption at that level signals ecosystem convergence. Developers building agent-native applications can assume AG-UI compatibility across cloud stacks.

What the Round Funds

The company is building in three directions with the new capital.

First, generative UI across the full spectrum: controlled UI through AG-UI, declarative UI through Google’s A2UI, open-ended UI through Anthropic’s MCP Apps. CopilotKit intends to support all three specifications in a single stack rather than forcing developers to pick one.

Second, Continuous Learning from Human Feedback (CLHF): treating every agent interaction signal (accepted, rejected, edited, redirected) as training data to improve agents from production usage without explicit fine-tune cycles. This ships next, per the announcement.

Third, the just-launched CopilotKit Enterprise Intelligence platform: a self-hosted Kubernetes product with persistent conversation threads, cross-device sync, analytics, and the self-improvement layer coming shortly. Named customers are already in production on it.

Who Invested and Why

Glilot Capital’s managing partner Lior Litwak said the speed at which cloud providers and open-source frameworks converged on CopilotKit and AG-UI is what drove conviction. NFX’s Gigi Levy-Weiss described enterprise AI agent adoption as being “slowed by a lack of tooling,” and AG-UI as “the front-end stack that closes it.”

SignalFire, which has around $3 billion in AUM and backed Grammarly and Frame, also participated, along with several smaller funds.

The founders are brothers Atai Barkai (CEO, former Meta and Doximity) and Uli Barkai (head of growth). They started as a podcast AI company (tawkitAI), open-sourced their internal infrastructure, saw strong developer traction, and pivoted to build CopilotKit. They went through Techstars Seattle 2023.

Competitive Context

CopilotKit competes with Vercel’s AI SDK, Assistant-ui, and OpenAI’s Apps SDK. The company differentiates on being vendor-neutral: it works with any agent framework (LangGraph, CrewAI, AutoGen), any cloud, any backend. The AG-UI protocol is MIT-licensed and maintained independently of any single lab or hyperscaler.

The funding comes roughly a year after AG-UI launched and about a year after CopilotKit hit meaningful enterprise adoption. At $27 million, this is an early-stage round for the layer that most enterprise agentic applications will eventually need to address. The bet is that the interaction interface layer is as structurally important to the agentic stack as the compute layer — and that open-protocol infrastructure wins there the same way it won in networking.