Cloudflare’s Vinext Recreates 94% of Next.js API in a Week, Boosting Vite Deployments

Vinext’s rapid development showcases how AI can compress a traditionally months‑long engineering effort into a single workweek. Cloudflare engineer Vinext directed Anthropic’s Claude, spending roughly $1,100 on token usage, to re‑implement 94 % of the Next.js API on top of Vite, according to a report by The Register. The resulting open‑source plugin—hosted on GitHub under the cloudflare/vinext repository—automatically detects a project’s pages or app directory, reads next.config.js, and generates a Vite configuration without requiring a custom vite.config.ts for basic use cases. By swapping the standard next scripts for vinext commands (dev, build, start, deploy), developers can run a development server with hot‑module replacement, produce production builds, and even deploy directly to Cloudflare Workers with a single CLI call.

The experiment was motivated by a long‑standing friction point highlighted by Cloudflare engineering director Steve Faulkner: Next.js’s tooling is “entirely bespoke,” making it difficult to run the framework on platforms other than Vercel without reshaping the build output. Faulkner noted that Vercel’s upcoming deployment adapters—still in an “early effort” stage—do not fully address the problem because Next.js relies on Turbopack, Vercel’s proprietary bundler, for both development and production pipelines. Existing community solutions such as OpenNext, a joint effort by SST, Cloudflare, and Netlify, have been described by Faulkner as “difficult and fragile,” requiring a cumbersome conversion of Vercel‑specific build artifacts into a format that other edge runtimes can execute.

Vinext’s approach sidesteps the Turbopack dependency entirely by leveraging Vite’s native ES‑module dev server and Rollup‑based production bundler. The plugin’s “Agent Skill” integrates with a suite of AI coding tools—including Claude Code, OpenCode, Cursor, and Codex—to automate migration steps. A single command, npx skills add cloudflare/vinext, installs the skill; the AI then scans the project, flags incompatibilities, installs missing dependencies, and generates the necessary configuration files. Developers can also run vinext check to pre‑flight compatibility or invoke vinext init to automate a full migration. The repository’s README makes clear that most of the code, tests, and documentation were authored by AI, with human engineers providing architecture guidance but not performing line‑by‑line reviews, a disclaimer that underscores the experimental nature of the tool.

Early adopters who have tried Vinext report that the plugin preserves the familiar Next.js directory structure and public assets while delivering a markedly faster development experience on Cloudflare’s edge network. Because Vite serves modules directly from the source during development, developers can see changes reflected instantly, avoiding the latency introduced by Turbopack’s incremental compilation. Moreover, the vinext deploy command packages the build output as a Cloudflare Workers script, eliminating the need for a separate adapter layer. This streamlined workflow could lower the barrier for teams that want to leverage Cloudflare’s KV storage, Durable Objects, or R2 bucket services without resorting to Vercel‑specific APIs.

While Vinext is still labeled “experimental” and its codebase has not been fully vetted by human reviewers, the project demonstrates a proof‑of‑concept that AI‑driven development can accelerate the recreation of complex framework APIs. If the tool matures, it may pressure Vercel to accelerate its own adapter roadmap or inspire other edge providers to adopt similar AI‑assisted migration pipelines. For now, the open‑source community can clone the repository, test the migration on a sandbox Next.js app, and contribute fixes to the remaining 6 % of the API surface that Claude did not yet cover.