ACAgentic Craft

Advanced~120 minHazelJS

Design a Production Agent Runtime

Multi-page Meridian Ops flagship guide: App code, DNA, Store, platform, and AgentRuntime — with HITL, Skillgate, and what to copy into your product.

Authors
editorial-team
Published
Last reviewed

Start Part 1 · Boot the Flagship

Before you start

You will leave with

  • Boot Meridian and explain store:sync vs platform:sync
  • Draw the five-layer model: App / DNA / Store / Platform / Runtime
  • Contrast CLI stubs with real /api/chat execute paths
  • Run a HITL refund pause/resume
  • Fill an Absorb Meridian memo for your product
On this page

Direct answer

A production Agent Runtime on HazelJS is not a chat widget with tools bolted on. It is AgentRuntime living in the same TypeScript backend as your HTTP APIs — with durable runs, policies, HITL, timelines — while DNA / Store / platform sit beside the process to version contracts and declare desired state.

The canonical lab is Meridian Ops (hazeljs-meridian-ops/). This guide is a multi-page series — start Part 1 after reading this hub.

Pattern: /patterns/production-agent-runtime-blueprint. Docs: Agent OS.

Tagline (Meridian): Ship AI workers as part of your backend — versioned like packages, governed like APIs, declared like infrastructure.

Why Meridian

AssetWhat it showsGap
hazeljs-agent-os-starterSupport desk + HITL + DNA exportLittle Store / Skillgate / multi-DNA
hazel agent runDNA smokeStub tools
Meridian OpsFull story in one backendThis guide’s lab

First @Tool only? Use /guides/build-your-first-tool-using-agent.

Journeys

JourneyGoalParts
AClone → sync → tour → chat1–3
BHITL refund pause/resume4
CSkillgate + MCP7
DRemote registry + Phase 68 (skip on first pass)

Honesty

  1. Commerce/risk data is in-memory.
  2. HITL defaults off (AGENT_OS_HITL=0).
  3. Platform APIs are v1alpha1.
  4. hazel agent run uses stub tools — product truth is /api/*/chat.
  5. Phase 6 labs are opt-in (Part 8).

Clone one-liner

bash
cd hazeljs-meridian-ops
cp -n .env.example .env
npm install

Then open Part 1 — Boot the Flagship.

Parts in this series

The part list appears below the hub content on this page (and as prev/next on each part). Lab time ~90–150 minutes with Meridian running. Pin: @hazeljs/*@2.0.1.

Parts in this series

  1. 1. Boot the Flagship~15 min

    Clone Meridian Ops: env, store:sync, platform:sync, tour, and dev server with Skillgate + DNA overlay.

  2. 2. Five-Layer Mental Model~12 min

    App code, DNA, Store, Platform, and Runtime — and why DNA alone cannot add cancelOrder.

  3. 3. Runtime Core: Execute, Tools, Timelines~15 min

    AgentRuntime.execute via chat APIs, real tools against the demo store, Inspector timelines, and eval smoke.

  4. 4. HITL, Policies, and Contracts~20 min

    Journey B: requiresApproval, durableSuspend, AGENT_OS_HITL=1, approvals API, and stub vs real anti-confusion.

  5. 5. DNA, Store, and Promotion~15 min

    Multi-DNA packages, change prompt without rewriting tools, and grow from thin starters toward Meridian.

  6. 6. Control Plane Beside the Runtime~12 min

    What platform apply does and does not do; packageRef vs nested DNA; events and reconcile (F19–F21).

  7. 7. Multi-Agent and Skillgate~18 min

    Journey C: ops-router, fraud path, Skillgate report/tools, and MCP export.

  8. 8. Absorb Meridian: Production Design~20 min

    Copy vs leave checklist, Phase 6 labs, troubleshooting, and the Absorb Meridian memo capstone.

Sources

  • Meridian Ops README — hazeljs-meridian-ops/README.md — mental model + quick start
  • Meridian TOUR.md — F1–F22 curl walkthrough
  • Flagship project plan (21) — docs/agent-os/21-flagship-project-plan.md
  • HazelJS Agent OS guide
  • HazelJS Agent package

Continue learning