The open-source layer that lets AI agents live inside your app — creators of the AG-UI protocol
HQ: Seattle, WA, United States | Founded: 2023 | Employees: ~25 (2026); engineering mostly Seattle-based | Stage: Series A ($27M, May 2026); total funding ~$27M | Website: https://www.copilotkit.ai
CopilotKit builds open-source infrastructure for putting AI agents inside applications — not as a chat sidebar, but as a working part of the product. Its SDKs (React, Vue, Angular) give developers prebuilt copilot UI and, more importantly, the plumbing behind it: shared state between agent and app, in-app tool calls, and generative UI where an agent can render a chart or a form instead of a wall of text.
In early 2025 the company shipped AG-UI, an open protocol for agent-user interaction. It sits alongside Anthropic's MCP (which connects agents to tools) and Google's A2A (which connects agents to each other) as the third leg of the agent stack: agents to humans. Adoption moved fast — Google, Microsoft, Amazon and Oracle support it, Microsoft contributed an official .NET SDK in September 2026, and the major agent frameworks speak it natively.
The company behind it is unusual: two Israeli brothers who originally moved to build an AI podcast tool (as tawkitAI, via Techstars Seattle 2023), open-sourced their internal copilot infrastructure, and watched that side project eclipse the original company. They renamed the company CopilotKit and went all-in on developer tools. Headcount stayed lean — roughly 25 people, engineering concentrated in Seattle — while the open-source footprint grew to the top tier of the agent ecosystem.
Revenue comes from CopilotKit Enterprise Intelligence, a self-hosted platform that hardens the open stack for enterprises: persistent conversation threads, cross-device sync, observability, and feedback loops that let agents improve from production usage. The bet is that enterprises want optionality and self-hosting rather than a single vertically-integrated AI vendor.
CopilotKit is betting on the least glamorous but most defensible layer of the agent stack: the handshake between AI agents and the humans using software. Models keep improving, agent frameworks keep multiplying, and every one of them eventually needs to talk to a user interface. CopilotKit's AG-UI protocol is an open standard for exactly that — streaming agent state into front-ends, rendering generative UI, and keeping a human in the loop.
The adoption numbers are the story. AG-UI has been adopted by Google, Microsoft, Amazon and Oracle, plus the major agent frameworks (LangChain, Mastra, Pydantic AI, Agno, AG2, LlamaIndex). Microsoft itself contributed a first-class .NET SDK to the protocol in September 2026. The company's libraries pull millions of installs per week, its repos have passed 40,000 GitHub stars, and CopilotKit claims a majority of the Fortune 500 use its tools in some form — mostly through the free open source, but increasingly as paying users of its self-hosted enterprise product.
The business is a classic open-core wager. The SDK and protocol stay free and vendor-neutral — a deliberate contrast to Vercel's AI SDK and OpenAI's Apps SDK, which tie you to their platforms. Money comes from CopilotKit Enterprise Intelligence: a self-hosted, Kubernetes-deployed layer adding persistent threads, analytics, observability and 'continuous learning from human feedback'. Named customers include DocuSign, S&P Global, Cisco, Deutsche Telekom and Function Health.
The risks are the standard ones for a protocol company: standards capture little value on their own, and giants adopting your protocol can also absorb your layer. At roughly 25 people and $27M raised, the company is small and early — which is either fragility or discipline, depending on how the enterprise motion converts. What's not in dispute is that it got to the layer first and got the ecosystem's biggest names to agree on it.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki