21st Builds the Production Layer for AI Agents
21st, the YC W26 startup behind the largest React component registry for AI applications, is launching a new SDK that lets developers deploy frontier AI agents inside their products without building the underlying infrastructure from scratch. The company's component registry is already used by 1.4M developers. 21st is backed by Pioneer Fund.
The new 21st Agents SDK targets a specific pain point: AI agents like Claude Code and OpenAI's Codex aren't simple API calls. They're full runtimes that need persistent sandboxes, filesystem access, and bash execution. That means you can't just host them on platforms like Vercel and call it a day. The SDK lets developers define an agent in TypeScript, deploy in one command, and get built-in streaming, session management, usage billing, and observability.
Out of the box, the SDK provides sandboxed code execution powered by E2B, a credential security proxy layer, SSE streaming compatible with Vercel's AI SDK, pre-built chat UI with themes, and usage-based billing. It supports React, Next.js, Node, and Python.
The product grew directly from the team's own frustrations. Like many teams, they started building agents from scratch, but when Claude Code arrived, its reasoning, context handling, and developer workflows were far ahead of anything they had built themselves. That was the good news. The bad news was that getting Claude Code into a production environment took months of infrastructure work. They decided to turn that infrastructure into a product for other teams hitting the same wall.
Co-founders Serafim Korablev and Sergey Bunas have a track record of shipping developer tools quickly. The company rapidly built and launched 9 products in its first 10 months, amassing over 1 million users, and their open-source work has accumulated over 15,000 GitHub stars. Bunas was previously a senior software engineer at Deel (YC W19). Korablev previously co-founded Via, a cross-chain crypto routing protocol with $1.5B in gross transaction volume.
The timing of the launch reflects a broader shift in how AI agents are being built. Rather than creating agent logic from scratch, more teams are building on top of frontier agent runtimes from Anthropic, OpenAI, and others. But getting those agents to work reliably for end users remains a significant engineering challenge, one that 21st is now positioning itself to solve.