Muhammad Arslan

Posted on Feb 13 • Edited on Mar 3

HazelJS Beta: Why We Built It — Design, Purpose & Benefits

#ai #backend #javascript #node

A different kind of framework announcement.
If you like it don't forget to Star HazelJS repository

The Problem We Kept Hitting

Every Node.js backend we built felt like we were assembling the same puzzle from scratch. Need AI? Wire it manually. Need dependency injection? Pick a library. Need authentication? Integrate another package. Need agents, RAG, or vector search? Build it yourself.

The ecosystem had great pieces — Express, NestJS, Prisma, TypeScript — but nothing that treated AI as a first-class citizen while staying lightweight, modular, and TypeScript-native. We wanted one framework where the boring stuff was solved and the creative part — building AI-powered products — could start on day one.

So we built HazelJS.

Why We Designed It This Way

1. AI-Native, Not AI-Bolted-On

Most Node.js frameworks treat AI as an add-on. You add a provider, wire a route, and hope it scales. HazelJS is built around the idea that AI is part of the architecture, not an afterthought.

@hazeljs/ai — One interface for OpenAI GPT-4, Anthropic Claude, Google Gemini, Cohere, and Ollama (local LLMs). Docs →
@hazeljs/agent — Production-grade agent runtime with stateful execution, tool calling, memory, and human-in-the-loop. Docs →
@hazeljs/rag — Retrieval-augmented generation, vector embeddings, support for Pinecone, Weaviate, Qdrant, Chroma. Docs →

You don't glue AI onto your app. You design your app with AI from the start.

2. Modular: Pay for What You Use

We didn't want a monolith. You install only what you need — no bloat, no hidden npm dependencies:

npm install @hazeljs/core                    # Core framework (required)
npm install @hazeljs/ai @hazeljs/agent        # AI and agent runtime
npm install @hazeljs/rag                      # RAG and vector search
npm install @hazeljs/kafka @hazeljs/cache    # Event-driven, caching — as needed

No giant framework. Just composable npm packages that work together.

3. Decorator-First, TypeScript-First

We believe a clean, declarative API beats configuration files. Decorators like @Controller, @Get, @AITask, and @RAG keep your code readable and your intentions explicit. TypeScript gives you compile-time safety so you catch bugs before they reach production.

4. Production-Ready Out of the Box

Security headers, graceful shutdown, health checks, structured logging — these shouldn't be projects. HazelJS ships with production defaults so you can focus on features, not infrastructure.

What We're Releasing in Beta

We're launching HazelJS v0.2.0-beta — a framework refined over months of real-world use. All packages are published on npm:

Category	npm Package	Purpose
Core	@hazeljs/core, @hazeljs/cli	Framework, DI, routing, CLI scaffolding
AI	@hazeljs/ai, @hazeljs/agent, @hazeljs/rag	LLM integration, agent runtime, RAG
Infra	@hazeljs/cache, @hazeljs/websocket, @hazeljs/serverless, @hazeljs/kafka, @hazeljs/discovery	Redis caching, WebSocket, AWS Lambda & GCP Functions, Kafka, service discovery
Data & Auth	@hazeljs/prisma, @hazeljs/auth, @hazeljs/swagger, @hazeljs/config, @hazeljs/cron	Prisma ORM, JWT auth, OpenAPI docs, config, cron jobs
Ops	@hazeljs/ops-agent	AI-powered Jira + Slack assistant for DevOps

Browse all packages on npm → | Documentation →

Benefits: What You Get

For Backend Developers

Less boilerplate — Decorators, dependency injection, and validation built in
Type safety — End-to-end TypeScript
Familiar patterns — Controllers, services, modules — like NestJS but lighter
One command — hazel new my-app and you're running (CLI guide)

For AI Engineers

Agent runtime — Stateful execution, tool calling, memory
RAG ready — Vector stores, embeddings, retrieval
Multi-provider — Switch between OpenAI, Claude, Gemini without rewriting
Streaming — Native support for LLM streams

For Teams

Modular — Add npm packages as you grow
Testable — First-class Jest testing utilities
Documented — Swagger/OpenAPI auto-generated
Deployable — AWS Lambda & Google Cloud Functions adapters out of the box

Real Use Cases

HazelJS is already powering:

AI chatbots — Agent runtime + RAG for context-aware conversations
Internal tools — @hazeljs/ops-agent for Jira + Slack automation (see blog post →)
Event-driven APIs — @hazeljs/kafka for event-driven architecture
Serverless backends — Deploy to AWS or GCP with @hazeljs/serverless
Microservices — @hazeljs/discovery for service discovery and load balancing

Who It's For

HazelJS is for developers who:

Build APIs and backends and want a clean, maintainable structure (vs. Express spaghetti)
Work with AI, agents, or RAG and want a framework that understands them (vs. bolting OpenAI onto Express)
Prefer TypeScript and decorators over config-heavy setups
Want modularity — install only what you need
Care about production — security, logging, health checks from day one

If that sounds like you, try HazelJS.

If you like it don't forget to Star HazelJS repository

Why Beta?

We've used HazelJS internally for real projects. The core is stable. The AI stack works. But we want feedback before we call it 1.0.

Beta means:

We're committed to fixing bugs quickly
We may refine APIs based on your feedback
We're not promising 100% stability for edge cases — yet
We are promising a framework we're proud to ship

Your input will shape the 1.0 release.

Get Started

Install the core framework and AI packages from npm:

npm install @hazeljs/core @hazeljs/ai

Or scaffold a full app with the CLI:

npx @hazeljs/cli@latest new my-ai-app

Resources:

Installation Guide — Get up and running
Documentation — Guides, API reference, examples
GitHub — Source, issues, contributions
Discord — Community and support
Blog — Ops Agent (Jira + Slack), Agent runtime, RAG

The Bigger Picture

We built HazelJS because we believe the next wave of applications will be AI-native. Not "an AI feature," but apps where intelligence is part of the architecture. HazelJS is our contribution to that future — a framework that treats AI as a first-class citizen while staying lean, modular, and developer-friendly.

We're excited to share it with you. Give it a try, share your feedback, and help us make it better.

— The HazelJS Team

DEV Community