HazelJSHazelJS
What Is an Agent OS?
An Agent OS is the production lifecycle for AI agents — runtime, orchestration, governance, security, durability, and observability. It is not a bootable host OS. HazelJS implements Agent OS for TypeScript inside your Node backend.
npm install -g @hazeljs/cli
hazel new agent support-agent
cd support-agent && npm install
hazel applyArchitecture
Inside HazelJS Agent OS
Define & Deploy
Version identity, prompt, skills, and policies as DNA. Declare desired state with Store/Apply — tools stay in TypeScript.
Run
Agent Runtime executes the loop. Agent VM classifies effects and enables speculation/undo. Organism self-organizes specialists from a mission and environment signals.
Decide
Bounded closed-set judgment before side effects. Evidence, judge, critic, and confidence inform policy — never authorize. Default provider is native hazel-agent, not an external-model wrapper.
Govern
Skillgate curates which APIs become capabilities. Gatekeeper authorizes each tool call — default deny, approvals, rewrite.
Survive
Durable runs, crash-safe HITL, leases, state, and recovery — so agents outlive process restarts.
TypeScript / Node.js Backend
Runs beside your APIs — local-first, no separate runtime plane required.
Define → Run → Decide → Govern → Survive
What you cannot get from LangChain alone
Packaging, durable approvals, and governed skills — inside the same Node process as your APIs.
01
DNA + Store
Version prompt and policies like libraries. Tools stay in @Tool code — DNA is not the implementation.
02
Crash-safe HITL
Refunds and freezes suspend on durable AgentRun. Restart mid-approval without losing the run.
03
Skillgate + apply
Curate REST into governed skills. Declare Definitions locally — apply does not restart Node. Kubernetes optional.
04
Agent Gatekeeper
Authorize every protected tool call before execution. Default deny, deterministic policies, and crash-safe approval — independent of the LLM.
05
Decision Runtime
Closed-set choose before side effects. Confidence informs policy; Gatekeeper still authorizes. Native hazel-agent pipeline — not an external-model wrapper.
06
Agent VM
Effect-typed tools (@Pure, @Read, @Reversible, @Irreversible) — speculative branches, transactional undo, automatic compensation on losing paths.
await runtime.execute('support-desk', message, {
loop: { maxIterations: 4, successScore: 90 },
contract: { name: 'support-slo', maxLatencyMs: 30_000, fallbackAgent: 'safe-desk' },
recovery: { maxRetries: 2, fallbackAgent: 'safe-desk' },
});DNA ≠ tools · Store ≠ platform ≠ run
Three layers, one backend. Overlay never replaces your @Tool handlers with stubs.
Package identity, prompt, policies, versions, lockfile
Desired Definitions / Deployments — local control plane
Real @Tool / Skillgate execute, HITL, timelines
Pick the right command
Apply declares desired state. Your app still runs the tools. Chat is always real execute.
hazel applyStart hazeld + register DNA — survives terminal close
hazel agent applyFiles-only desired state (no supervisor)
hazel agent runDNA smoke with stubs — not product money moves
Clone Meridian — see the whole Agent OS
One repo that teaches DNA vs tools, Store vs platform vs run, Skillgate, router, and HITL. Copy the pattern into your product.
01
store:sync
Publish DNA packages + lockfile
02
platform:sync
Apply Definitions / Deployments
03
npm run dev
Chat with real @Tool handlers
terminal
cd hazeljs-meridian-ops && npm i && npm run store:sync && npm run platform:sync && npm run devAgent OS, explained
Short answers for the search query — then clone Meridian or read the guide.
What is an Agent OS?+
An Agent OS is the production lifecycle for AI-native agents: durable runs, packaging, governance, recovery, and CI — not a bootable host operating system. HazelJS Agent OS runs that lifecycle inside your TypeScript / Node backend, in the same DI app as your APIs. Read the long-form article
What is HazelJS Agent OS?+
HazelJS Agent OS is durable AI agents inside your TypeScript backend: DNA packages, crash-safe HITL, Skillgate, Gatekeeper authorization, and local apply/reconcile. It is not a bare-metal OS and not a Python-only kernel.
Is HazelJS the same as Builder Methods Agent OS, Agno AgentOS, or agentos.sh?+
No. Builder Methods Agent OS is a markdown standards and spec system for coding agents (Claude Code, Cursor). Agno AgentOS is a Python FastAPI runtime for serving agents. Framers AgentOS (agentos.sh) is a separate TypeScript agent framework. HazelJS Agent OS is the production lifecycle for durable agents inside your TypeScript / Node backend — DNA, Skillgate, Gatekeeper, crash-safe HITL, and local apply in the same DI app as your APIs.
Is DNA the same as tool implementation?+
No. DNA owns identity, prompt, tool names, and policies. Real side effects stay in TypeScript @Tool handlers or Skillgate. Overlay never replaces tools with stubs.
Does apply restart my Node app?+
hazel apply ensures the local hazeld node, registers DNA, and returns — agents keep running after you close the terminal. The older hazel agent apply path only writes desired state files without starting a supervisor.
Can I undo agent side effects or speculate across plans?+
Yes — with @hazeljs/agent-vm. Tools declare effect classes (@Pure, @Read, @Reversible, @Irreversible). @Speculate forks K reasoning branches; the winner commits and losers roll back via @Compensate handlers. @Atomic enables transactional undo of entire runs.
What is an Agentic Organism?+
With @hazeljs/organism you deploy a mission, genes, and constitution — not a fixed agent graph. Environment signals drive spawn, reuse, specialization, survival, and a resource economy under hard safety limits. Embed with createOpsOrganism and OrganismHostRegistry; a durable OrganismRepository hydrates the same organism across process replicas.
What is Decision Runtime?+
@hazeljs/decision is the native Decision Runtime for Agent OS: closed-set choices, confidence, and deterministic policy before Gatekeeper authorizes side effects. Confidence never means permission. The default provider is hazel-agent — not a wrapper around an external decision model.
Do I need Cloud or Kubernetes?+
No. Local registry and local backend are enough. Hosted registry and K8s are optional.
Run locally (5 min)
hazel new agent → apply → still Running
What is an Agent OS? (article)
Long-form narrative
Agent OS guide
DNA, HITL, Skillgate, apply
Agent DNA
Package contract
Agent Runtime
Execution kernel
Durable Kernel
Runs + HITL
Skillgate
Governed capabilities
Gatekeeper
Tool authorization
Decision Runtime
Closed-set decide
Agent VM
Effect types
Control Plane
hazeld + apply
vs Builder Methods Agent OS
Specs vs runtime
vs Agno AgentOS
Python vs TypeScript
vs agentos.sh
Name collision
vs LangGraph
Graphs vs Agent OS
Glossary: Agent OS
Canonical term