Skip to content
Projects

hazel-js/hazeljs has 58 GitHub stars and 6 forks. GitHot measured 0 stars in 24 hours, +7 in 7 days, and +8 in 15 days. Last checked 2026-11-15 23:11 UTC. Last push 2026-10-26.

How these counts are measured

Measured star growth

in 24h
0
in 7 days
+7
in 15 days
+8
in 1 month
Update delayed

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

HazelJS Agent OS — durable TypeScript agents (DNA, Skillgate, Gatekeeper, crash-safe HITL) in your Node backend

AI

README.md

Original

HazelJS

The Agent OS for TypeScript.
Operate agents like production systems.

GitHub stars CI npm version npm downloads License: Apache-2.0

Frameworks help you build agents. HazelJS helps you run them — durable, governed, local-first.

Get started · Agent OS · Docs · Meridian


Why an Agent OS?

Frameworks help you build Prompt loops and tools get you to a demo.
Agent OS helps you operate Durable runs, Gatekeeper, DNA, and HITL keep agents running for real.
Local-first TypeScript hazel apply on your laptop. Cloud and Kubernetes are optional.

You do not assemble Nest + LangGraph + a worker fleet. Agents live in the same TypeScript backend as your APIs.

Define → Run → Decide → Govern → Survive

Name collisions: HazelJS Agent OS is not Builder Methods Agent OS, Agno AgentOS, or Framers AgentOS.


Get started

Four commands. Install → Apply → Running — survives closing the terminal.

npm install -g @hazeljs/cli
hazel new agent support-agent
cd support-agent && npm install
hazel apply

Guides: Local Agent OS · Agent OS overview

Prefer a full teaching app?

Clone Meridian — DNA, Store, Decision, Skillgate, HITL, and local apply in one repo.

git clone https://github.com/hazel-js/hazeljs-meridian-ops.git
cd hazeljs-meridian-ops
npm install
npm run store:sync      # DNA packages + lockfile
npm run platform:sync   # Apply Definitions / Deployments
npm run dev             # Chat with real @Tool handlers + HITL

HTTP + AI app (ai-native template)

Scaffold a server on port 3000 (chat, agents, Prisma RAG, ContextService) — not hazeld. Use this when you want an HTTP/HCEL backend first; use Get started or Meridian for local Agent OS (apply / HITL).

npm install -g @hazeljs/cli
hazel new my-app --template=ai-native
cd my-app
npm install
cp .env.example .env    # add OPENAI_API_KEY
npm run dev

Then open:

Quick checks:

curl -X POST http://localhost:3000/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "What is HazelJS?"}'

Skeleton API only (no AI scaffold): hazel new my-api or hazel g app my-api.


What you get

Layer Primitives Job
Define & Deploy DNA · Store · Apply Version identity, prompt, skills, and policy. Declare desired state locally.
Run Runtime · Agent VM · Organism Execute the loop. Speculate / undo. Self-organize specialists from a mission.
Decide Decision Runtime · Confidence · Policy Closed-set choose (retry | rollback | escalate). Confidence informs — never authorizes.
Govern Skillgate · Gatekeeper Curate which APIs become skills. Authorize every tool call (default deny).
Survive Durable runs · HITL · Recovery Crash-safe human approval. Runs outlive process restarts.

Built on your TypeScript / Node.js backend. Local first. Kubernetes optional — never the product.


Agents in code

import { Agent, Tool } from '@hazeljs/agent';

@Agent({
  name: 'support-agent',
  systemPrompt: 'You are a helpful customer support agent.',
})
export class SupportAgent {
  @Tool({
    description: 'Look up order by ID',
    parameters: [{ name: 'orderId', type: 'string', required: true }],
  })
  async lookupOrder(input: { orderId: string }) {
    return { status: 'shipped', trackingNumber: 'TRACK123' };
  }

  @Tool({
    description: 'Process a refund',
    requiresApproval: true,
    parameters: [
      { name: 'orderId', type: 'string' },
      { name: 'amount', type: 'number' },
    ],
  })
  async processRefund(input: { orderId: string; amount: number }) {
    return { success: true, refundId: 'REF123' };
  }
}

Decision Runtime — choose, then govern

LLMs reason. Decision Runtime judges from a closed set. Gatekeeper still authorizes.

import { createDecisionRuntime } from '@hazeljs/decision';

const decisions = createDecisionRuntime({ gatekeeper });

const result = await decisions.decide({
  objective: 'Choose the safest remediation',
  state: { errorRate: 0.38, failedHealthChecks: 8 },
  choices: ['retry', 'rollback', 'escalate', 'ignore'] as const,
  risk: 'high',
  execute: true,
});
// result.decision is typed — Gatekeeper still gates side effects

Skillgate — OpenAPI → governed skills

import { Skillgate } from '@hazeljs/skillgate';

const gate = Skillgate.fromOpenApi(spec, {
  include: { tags: ['agent'] },
  classify: { writeRequiresApproval: true },
  invoke: { baseUrl: process.env.API_BASE_URL! },
});
gate.register(registry, 'api-concierge');

More: Decision · Skillgate · Gatekeeper · Agent VM


CLI

# Agent OS
hazel new agent my-desk
hazel apply
hazel agent run | doctor | logs
hazel agent runs list | inspect | resume | approve
hazel store publish | install | list
hazel skillgate from-openapi ./openapi.yaml
hazel gatekeeper validate | simulate
hazel decision run | lab | compare

# HTTP app scaffolding (secondary — not hazeld)
hazel new my-app --template=ai-native
hazel g app my-api
hazel add @hazeljs/auth --setup

Examples

Repo Learn
hazeljs-meridian-ops Flagship Agent OS — DNA, Store, Decision, Skillgate, HITL
hazeljs-skillgate-agent-starter OpenAPI → governed skills + MCP
hazeljs-mcp-starter @Tool as an MCP server
hazeljs-csr-agent Support agent example

Packages

Agent OS

Package Role
@hazeljs/agent Runtime — @Agent / @Tool, DNA, HITL, policy
@hazeljs/decision Closed-set Decision Runtime
@hazeljs/skillgate OpenAPI → governed skills
@hazeljs/agent-gatekeeper Fail-closed tool authorization
@hazeljs/agent-vm Reversible tools, speculative undo
@hazeljs/organism Mission-defined agent societies
@hazeljs/testing · @hazeljs/eval · @hazeljs/benchmark CI, golden sets, regression
@hazeljs/inspector · @hazeljs/mcp Timelines · MCP from @Tool

Around Agent OS (optional — they support it, they do not compete with it)

Area Packages
AI / RAG / Flow @hazeljs/ai · @hazeljs/rag · @hazeljs/flow · @hazeljs/prompts
Backend @hazeljs/core · @hazeljs/auth · @hazeljs/prisma · GraphQL / gRPC / queues
npm install @hazeljs/agent @hazeljs/decision @hazeljs/skillgate @hazeljs/agent-gatekeeper
# CLI
npm install -g @hazeljs/cli

Do not add reflect-metadata yourself — @hazeljs/core loads it.


Why not Nest + LangGraph?

HazelJS Nest + LangGraph
Agents in the same app as APIs Yes Glue
DNA / Store / local apply Yes No
Closed-set Decision Runtime Yes DIY
Governed OpenAPI skills Skillgate DIY
Fail-closed tool auth Gatekeeper DIY
Crash-safe HITL Yes Partial

Docs & community


License

Apache 2.0 — free for commercial and open-source use.


Operate agents like production systems.

Get started · Meridian · Discord · ⭐ Star