See exactly what your agent did.
Your agent prepares the action. You decide. Every completed action leaves a receipt.
Agents do the work. You stay in charge.
VZN Labs is an agentic platform for your industry: an agent prepares the work, a person approves it, and nothing happens without a record.
Agents prepare the work
Replies, quotes, campaigns, pages and code are drafted by agents, each one inside the limits you set.
Meet VZN Agents →You decide
Nothing is sent, changed or billed until a person on your team approves it.
Try an approval →Every action leaves a receipt
Each approved action is kept with who approved it and the proof that it happened.
How a pilot is billed →
The agentic platform for teams that approve first.
- 12products on one platform
- 10with something built today
- 2planned, and shown as planned
- R 0.00billed without an approval
Twelve products. One platform.
Each product page says what works today and what is still planned. A dashed card is not built yet.
- VZN AgentsAgents that propose the work and wait for your APPROVE.The agent runtime: an agent loop, installable packs, a workflow engine and a Hub where every outbound action waits behind APPROVE. Live voice agents and personal assistants belong here too.Partly live
- VZN UIThe design language, as code.React components, a native sibling and a token core, all built on one design system, with five styles (VZN, Glass, Liquid, Neu and Minimal) in light and dark, Inter and Geist Mono.Partly live
- VZN Agent StudioCampaigns and content, written from your own facts.Plan a campaign, write ad, WhatsApp, SMS, email and social copy yourself or with AI, and check every line against your facts before you approve it.Partly live
- VZN MarketingWrite once. Post from accounts you connect yourself.Approved text posts go to seven networks, now or later. You connect the accounts; nothing posts without you.Partly live
- VZN ConnectYour desk is an MCP server, and it reaches other servers.Connect an AI client to your desk over MCP with a scoped key, and allowlist the tools your agents may call on other MCP servers.Partly live
- VZN SkillsTeach an agent a recipe once. Reuse it everywhere.Tenant skill recipes with validation, sharing and an audit trail, plus a first-party knowledge-base skill.Partly live
- VZN Bench (Evals & Observability)Evals and observability you can rerun.A deterministic evaluation harness: eight suites and 72 cases, run weekly, with reference answers that must pass.Partly live
- VZN Sales & ServiceCustomer records that agents propose changes to.A CRM where agents propose the write and you APPROVE it. The records, the screens in the desk, the sales and service processes, the reports, the dashboards, extra access, sharing, teams, page layouts, record types, a seat limit and a last-active time per user are built.Partly live
- VZN OperationsStock, orders and invoices, with agents doing the work.Agentic operations over MCP connectors. Agents propose stock, order and invoice work; a write still waits for APPROVE.Planned
- VZN SDKA typed client for the VZN API.TypeScript first: run agents, stream their steps, read and fork sessions, with types generated from the engine's own contract.Partly live
- VZN CLIA coding agent for your terminal.vzn works in any project folder: you chat, it reads and edits code and runs commands, and you approve what matters. It runs on free providers or a paid model.Partly live
- VZN ExchangeIndustry packages you install like an app.A marketplace of industry packages: skills, a control card and defaults. It is not the connector catalogue.Planned
Every action takes four steps, and the last one is yours.
A request arrives
A customer message, a form, or a task you give an agent.
An agent prepares the action
A reply, a quote or a change to a record, with what it relied on.
A person approves
Nothing outbound happens without an approval from your team.
A receipt is kept
What was done, who approved it, and the proof that it happened.
Built for the work your industry does.
Each industry has its own workspace of agents, connections and approvals.
Measured against what it replaces.
Result against control, with the size of the test beside it.
The easy calls, made in microseconds.
A small model reads every inbound message and routes it: answer from a template, hold for a person, or hand to a larger model when it isn’t sure. It runs on a CPU, with no GPU and no model host, and legal, medical and irreversible requests always go to a person.
Try the decision console →Held-out set of 136 synthetic WhatsApp-style messages. No pilot has produced a live result yet; shadow testing on real, consented traffic comes before it answers customers.
No approval, no bill.
In a pilot, a result is an evidenced action a person on your team approved. You review a monthly statement of every result, with who approved it and the proof, before any invoice is sent.
What counts
An action your team approved, with proof it happened: a response code, not a status flag. Nothing is billed without an approval.
What it costs
A price per result, agreed with you for your industry before the pilot starts. R 35.00 in the examples on this page.
Your ceiling
A monthly cap you agree up front (R 500.00 in the examples). Once it is reached, further results are not billed.
Every price on this page is an example. Pilot terms are agreed with you before anything starts. Ask about a pilot.
Log in or create your account.
Use the account you already have, or make one now. Everything your agents do stays yours, with a receipt.
Bring us the problem.
Tell us your industry and what you want an agent to take off your team's plate. We will propose a pilot, with terms agreed up front.