Agent · 01
ATLAS
Full-stack app builder.
engineering · VZU Build
VZU OS · The Agent Operating System
An operating system for the work.
12 specialized agents. One orchestrator. MCP-native integrations. SOC 2 / HIPAA / ISO 27001 posture. Audit trail by default. The runtime that powers every VZU capability, every VZU brand, every VZU engagement.
VZU OS is the agent operating system that powers the VZU family. It runs 12 specialized agents — Atlas, Vega, Kai, Nova, Pixel, Echo, Sketch, Launch, Sentinel, Oracle, Quill, and Hunter — under a central orchestrator. The agents communicate via MCP (Model Context Protocol), enforce RBAC at the agent level, and produce an audit trail by default. VZU OS is operated by Virtual Zero Unbound Inc. and runs on the NetWit Agentic OS runtime in cloud, hybrid, or air-gapped on-prem.
The orchestrator
One orchestrator. Twelve agents. Every brief.
The VZU orchestrator is a multi-agent runtime. It takes a brief, decomposes it into subtasks, routes each subtask to the right agent, observes the result, and replans if needed. The orchestrator enforces RBAC at the agent level and produces a complete audit trail for every brief.
Click any agent to see its execution trace.
How VZU OS works
Four stages. Every brief.
-
Step · 01
Brief
A real brief from a real operator. The goal, the data, the regulator, the deadline. The brief is the contract.
-
Step · 02
Orchestrate
The orchestrator decomposes the brief into subtasks, routes to the right agents, and sets the audit posture.
-
Step · 03
Execute
The agents run. Each one calls MCP servers, observes, replans. Every call is logged in the audit trail.
-
Step · 04
Audit
The orchestrator reviews the work, the audit trail, and the operator's RBAC. The result is signed, sealed, delivered.
Orchestration patterns
Three patterns. One orchestrator.
The VZU orchestrator supports three core patterns. The right pattern depends on the brief.
-
Pattern
Relay
A → B → C. One agent hands off to the next. The cleanest pattern. The audit trail is a clean chain of custody.
Use case
Brief → BriefAnalyzer → DomainAgent → Output. Each step is signed, audited, replayable.
-
Pattern
Parallel
A, B, C in parallel. Results are combined. Used when the brief has independent subtasks.
Use case
Brief → [Atlas, Vega, Nova] in parallel → Orchestrator combines → Result. Each parallel branch is independently audited.
-
Pattern
Supervisor
A delegates to B and C, observes, re-delegates if needed. The most powerful pattern. The most audited.
Use case
Brief → Supervisor → [Worker A, Worker B, Worker C] → Supervisor reviews → Result. The supervisor re-plans based on observations.
The runtime, visualized
The orchestrator, in motion.
Every brief, every audit trail, every MCP call — visualized below. The orchestrator moves between relay, parallel, and supervisor patterns based on the brief. The agents are alive, the audit trail is on.
The 12 agents
Meet the agents.
Each VZU capability is powered by a dedicated agent. The agent owns the goal, the toolset, the memory, and the audit trail. Click any agent to see its execution trace.
Agent · 02
VEGA
React/Next.js engineer.
engineering · VZU Web
Agent · 03
KAI
Backend + infra engineer.
engineering · VZU Stack
Agent · 04
NOVA
Editorial UI designer.
design · VZU Surface
Agent · 05
PIXEL
Motion + WebGL designer.
design · VZU Canvas
Agent · 06
ECHO
WCAG accessibility auditor.
design · VZU Design
Agent · 07
SKETCH
Wireframe + prototype designer.
design · VZU Sketch
Agent · 08
LAUNCH
Landing page composer.
design · VZU Launch
Agent · 09
SENTINEL
Code reviewer + security auditor.
review · VZU Audit
Agent · 10
ORACLE
Production prompt engineer.
review · VZU Prompts
Agent · 11
QUILL
Spreadsheet engineer.
data · VZU Sheets
Agent · 12
HUNTER
Resilient web scraper.
automation · VZU Scrape
An execution trace
What an agent run looks like.
Below is the actual execution trace from a recent VZU OS run by Atlas (VZU Build). The agent was briefed: build a multi-tenant booking dashboard. Here is every step it took.
atlas@vzu-os — trace.log
- 01 ◉ atlas > read brief
- 02 ◐ atlas > thinking: brief is "build a multi-tenant booking dashboard". scope: 2 weeks.
- 03 ◇ atlas > mcp.data.schema > generate schema: tenants, services, customers, bookings, payments
- 04 ◐ atlas > thinking: stripe connect requires 2-day payout. need to scope webhook handler.
- 05 ◇ atlas > mcp.code.generate > next.js 15 + prisma + stripe-connect-express
- 06 ◇ atlas > mcp.code.generate > component library: Calendar, CustomerList, BookingForm
- 07 ◇ atlas > mcp.deploy.staging > ship to staging.atlas.vzu.ca
- 08 ◇ atlas > mcp.test.run > 47 integration tests passing
- 09 ⏵ atlas > ship. 14 days. 4 deploys. audit trail: 23 entries.
atlas>
Audit trail · 8 entries
- T+00:00:00 brief.received
- T+00:01:12 orchestrator.decomposed
- T+00:04:33 mcp.data.schema.generated
- T+02:18:07 mcp.code.generate.api.completed
- T+05:42:19 mcp.code.generate.ui.completed
- T+08:11:55 mcp.deploy.staging.completed
- T+11:33:48 mcp.test.run.completed
- T+14:00:00 brief.completed
The integration story
MCP-native. The cleanest way to call an action.
Every VZU agent calls its work through MCP — Model Context Protocol. MCP is the open standard for tool calling. Each MCP server advertises a manifest of actions (name, description, input schema, output schema). The agent receives the manifest, decides which action to call, calls it, and continues reasoning.
This is the cleanest way to call an action. Stateless. Structured inputs. Structured outputs. Explicit error semantics. Fast response times. The patterns that work in production. The patterns that fail in production are also narrower — stateful actions that hold long-running connections, actions that depend on user-specific context the agent doesn't have, ambiguous error responses, actions that are too slow for an interactive loop.
The VZU OS enforces the patterns that work. Every MCP action is stateless. Every action has a single responsibility. Every action returns a typed result. Every action has a fast response time. The audit trail records every call.
The compliance posture
SOC 2. HIPAA. ISO 27001. Audit trail from day one.
VZU OS is the compliance boundary. Access control is enforced at the agent level, not the user level. Change management is enforced on prompts, models, and action definitions, with every change versioned and auditable. Monitoring is enforced on every action call, every model call, every agent decision, with the full conversation history retained for the auditor.
The runtime ships as a binary. Cloud (Canadian, US, or EU region). Hybrid (runtime on operator hardware, model API-called). Fully on-prem (runtime, model, and tools on operator infrastructure, including air-gapped). The same code, the same audit trail, the same RBAC, regardless of where it runs.
The shape
VZU OS, by the numbers.
12
Specialized agents
one orchestrator
3
Orchestration patterns
relay · parallel · supervisor
50+
MCP actions
across the runtime
1
Runtime
NetWit Agentic OS
Featured traces
Side-by-side execution traces. Scroll to see them all.
Agent
ATLAS
Full-stack app builder.
01 atlas > read brief
02 atlas > thinking: brief is "build a multi-tenant booking dashboard". scope: 2 weeks.
03 atlas > mcp.data.schema > generate schema: tenants, services, customers, bookings, payments
04 atlas > thinking: stripe connect requires 2-day payout. need to scope webhook handler.
05 atlas > mcp.code.generate > next.js 15 + prisma + stripe-connect-express
Agent
VEGA
React/Next.js engineer.
01 vega > read brief
02 vega > thinking: refactor admin console. stack: react 19 + next.js 15 + tailwind v4.
03 vega > mcp.code.read > scan components, find 47 client components
04 vega > thinking: 23 of these can be server components. move state to URL search params.
05 vega > mcp.code.write > server component: <BookingList />
Agent
KAI
Backend + infra engineer.
01 kai > read brief
02 kai > thinking: multi-tenant saas. cloudflare workers + hyperdrive. 8-12 weeks.
03 kai > mcp.db.schema > tenants, users, organizations, plans, subscriptions, events
04 kai > mcp.db.migrate > applied 12 migrations to staging
05 kai > mcp.api.deploy > hono + trpc + zod, deployed to workers
Agent
NOVA
Editorial UI designer.
01 nova > read brief
02 nova > thinking: uae real estate marketing site. bilingual. editorial. <2s load.
03 nova > mcp.figma.write > design tokens: ink, paper, graphite, clay, mist
04 nova > mcp.figma.write > components: Hero, Bento, Marquee, TestimonialGrid
05 nova > mcp.code.write > astro 5 + gsap + lenis
Agent
PIXEL
Motion + WebGL designer.
01 pixel > read brief
02 pixel > thinking: 3d product viewer. three.js + react-three-fiber. 6-8 weeks.
03 pixel > mcp.three.scene > camera, lights, model, controls
04 pixel > mcp.shader.compile > pbr + reflections + soft shadows
05 pixel > mcp.motion.choreograph > 12-second orbit, 8-second reveal, 4-second configurator
Agent
ECHO
WCAG accessibility auditor.
01 echo > read brief
02 echo > thinking: wcag 2.2 aa audit. 47 pages. 2-4 weeks.
03 echo > mcp.a11y.scan > axe-core: 12 violations, pa11y: 8 violations, wave: 15 issues
04 echo > thinking: 6 critical (color contrast), 4 serious (keyboard nav), 7 moderate
05 echo > mcp.screenshot.compare > reference: 2 patterns need rework
The full cast
Twelve agents. Every capability. In motion.
Get in touch
Brief VZU OS.
The runtime handles the orchestration, the audit trail, and the enforcement. The operator owns the goal.