Introduction

Nodalite is a small, runtime-agnostic TypeScript API framework. The same App instance runs unmodified on:

• Node.js (18+) — via @nodalite/adapter-node
• AWS Lambda — via @nodalite/adapter-lambda
• Cloudflare Workers — via @nodalite/adapter-edge
• Bun and Deno — directly, no adapter needed

What makes it different?

Framework	Runtimes	Serverless-native	Built-in security	Background jobs	ML inference
Express 5	Node only	Bolt-on	None	None	None
Fastify 5	Node only	Bolt-on	Plugins	None	None
NestJS	Node only	Poor cold starts	Guards/interceptors	@nestjs/schedule	None
Hono	Node/Bun/Deno/Workers/Lambda	Yes	Middleware ecosystem	None	None
Elysia	Bun-first	Bun only	Plugins	None	None
Nodalite	Node/Bun/Deno/Workers/Lambda	Yes (first-class)	Built in	Yes	Yes

Genuinely unmet needs

1. Background threads alongside your API — @nodalite/workers' runDetached() runs a supervised worker_thread (bot, poller, watcher) in the same process, with automatic crash recovery. See Background Threads.

2. Lightweight ML inference on serverless — @nodalite/ml's Model class disk-caches model bytes across cold starts, deduplicates concurrent load requests, and is engine-agnostic so it doesn't force a heavy native dependency on you. See ML Inference.

Everything else — routing, middleware, validation, auth, rate limiting — is a well-executed implementation of table-stakes features, built on the standard Fetch API so the exact same code genuinely runs everywhere.

Design principles

• Fetch API foundation — Request/Response/Headers/ReadableStream are the only primitives. No runtime-specific shapes.
• Zero dependencies in core — @nodalite/core relies on nothing but the JS runtime. This keeps bundles small and supply-chain risk minimal.
• Independent versioning — each package ships independently. Adapter fixes don't bump core. ML's heavy native deps stay optional.
• Honest about limits — runDetached() can't work on FaaS, and the docs say so explicitly. No marketing over truth.