Building the AI-native backend
HazelJS is an open-source TypeScript framework where agents, RAG, workflows, and enterprise primitives live in one cohesive stack — not scattered across libraries.
One stack for the full AI backend lifecycle
Most teams building AI products end up stitching LangChain, Express or NestJS, vector DBs, observability, and auth into something that barely holds together. HazelJS started from a simpler question: what if all of that lived in one TypeScript framework from day one?
Today HazelJS ships 47 packages — from a two-line HCEL install for brownfield apps to a full agent runtime with durable workflows, guardrails, and OpenTelemetry built in. Open source under Apache 2.0, built with and for developers shipping agents to production.
Agent Runtime
Stateful execution with tools, memory, and human-in-the-loop.
HCEL Orchestration
Chain prompts, RAG, and agents in one fluent pipeline.
Enterprise Stack
Auth, audit, observability, and durable workflows built in.
47 Packages
Modular — install only what your app needs, scale up when ready.
Give every TypeScript team a production path for AI — from a two-package HCEL install in a brownfield app to a full agent runtime with memory, guardrails, and observability.
Become the default backend for teams shipping agentic software — where orchestration, state, and deployment are solved in one framework, not stitched together.
What we optimize for
Ship to production
Agents that pause, resume, and survive restarts — not demo scripts that break under real load.
Open by default
Apache 2.0, community-driven, and transparent. No black-box abstractions hiding what your agents do.
Developer-first
Decorators, DI, and TypeScript patterns you already know — with AI primitives that feel native, not bolted on.