HazelJS First Alpha Release

Today we're releasing the first alpha of the HazelJS ecosystem. This is a milestone for us: a modular, TypeScript-first set of packages built for AI-native applications, with a clear path from zero config to production.

What is HazelJS?

HazelJS is a modular framework for building backend and AI applications in Node.js. You install only the packages you need—core, AI, agent, RAG, memory, flow, auth, cache, and more—and compose them in your app. No monolith, no "all or nothing."

We've been shipping beta packages for a while. This alpha tag marks the first time we're inviting the broader community to try the stack as a whole: run it, break it, and tell us what to fix before we lock in APIs and behaviors.

What's in the alpha?

• @hazeljs/core — Application bootstrap, modules, dependency injection, graceful shutdown, health checks.
• @hazeljs/ai — LLM integrations (OpenAI, Anthropic, Gemini, Cohere, Ollama) with a unified provider API.
• @hazeljs/agent — Production-oriented agent runtime: stateful execution, tools, memory, approval workflows, observability.
• @hazeljs/rag — RAG pipeline, document loaders, vector stores, GraphRAG, and a memory system (conversation, entities, facts, working memory).
• @hazeljs/memory — Pluggable user memory with multiple backends: in-memory by default, optional Postgres, Redis, Prisma, vector. One model for profile, preference, behavioral, emotional, episodic, and semantic memory—shared by RAG and agents when you want it.
• @hazeljs/flow — Durable workflow engine: in-memory or Prisma, wait/resume, idempotency, retries, audit trail.
• @hazeljs/flow-runtime — Optional HTTP service for running and resuming flows.
• @hazeljs/auth, @hazeljs/cache, @hazeljs/prisma, plus config, CLI, cron, serverless, swagger, websocket, Kafka, resilience, gateway, discovery, and more. Install only what you use.

Why "alpha"?

• APIs may change. We'll avoid breaking changes where we can, but we're not yet promising full stability. Your feedback will shape the next iterations.
• We want real usage. Use the packages in side projects or internal tools. Report bugs, missing features, and rough edges on GitHub.
• Documentation and examples. We're improving docs and adding examples (including shared memory for RAG + agents) so you can get started quickly.

Highlights for AI and memory

RAG + Agent shared memory (in-process): One MemoryManager, backed by RAG's stores or by @hazeljs/memory via the adapter. Pass the same instance to RAGPipelineWithMemory and every AgentRuntime so they share conversation and context in a single process.

@hazeljs/memory: One store interface, multiple backends (in-memory, Postgres, Redis, Prisma, vector). Categories for profile, preference, behavioral, emotional, episodic, and semantic summary. Package docs and Memory guide are updated with purpose and examples for every store type.

Agent runtime: Tools, approval workflows, memory integration, and observability so you can build agents that act in the real world safely and auditably.

How to try it

npm init -y
npm install @hazeljs/core @hazeljs/ai @hazeljs/agent @hazeljs/rag

Or with the CLI:

npx @hazeljs/cli new my-app

• Docs: hazeljs.com/docs
• GitHub: github.com/hazel-js/hazeljs
• npm: npmjs.com/org/hazeljs

What's next

We'll iterate on the alpha based on your feedback, fix issues, and improve docs and examples. When we're confident in stability and migration paths, we'll move toward a stable release and clearer versioning.

Thank you for trying HazelJS. We're excited to have you in the loop.

— The HazelJS Team