The agentic operating system, explained.
How the OS model works, how authority is delegated, and the six agents it runs today — for the evaluator who needs the mechanics, not the pitch.
One OS, four familiar concepts.
MyntriqOS borrows its model from the operating systems your infrastructure team already trusts.
= Processes
Each agent runs as an isolated unit of work, scheduled and supervised — not a script left running unattended.
= The scheduler
Work is routed to the right agent, sequenced, and retried under the same rules every time.
= Memory protection
One customer's process space cannot read or write into another's — enforced beneath the application, not by convention.
= The permissions model
What an agent may do alone, and what it must ask for, is configured — not hardcoded into its behaviour.
Four layers, each isolated for a reason.
Control plane
Policy, delegated authority and audit live here — separated so a compromised agent can't rewrite its own permissions.
Distribution
Routes work to the right agent and skill version, and enforces the tenant boundary on every request.
Model layer
Frontier, open-weight or locally hosted models, selected per workload — no single vendor dependency.
Connect layer
The only layer with credentials to your systems of record — scoped per agent, per action, per tenant.
A spectrum, not a switch.
Every agent action sits on one of three settings — configured per action, per agent, per tenant.
Ask before acting
The agent proposes the action and waits for a human to approve it.
Act and notify
The agent proceeds, and a human is informed immediately, with a clear undo path.
Act autonomously
The agent completes the action within limits set in advance, with no action required.
Some ceilings can't be configured away. Payments, invoices and admin changes never run autonomously — regardless of settings.
Six governed agents, out of the box.
Receptionist
Greets, qualifies and routes every inbound query to the right person or system, before a human ever sees it.
Sales & Marketing
Works pipeline and campaigns end to end — enrichment, follow-up and reporting — under the approval limits you set.
IT Support
Triages tickets, resolves the routine ones, and escalates the rest with full context already attached.
HR
Handles onboarding steps and policy questions directly, so people ops isn't answering the same question twice a day.
Personal Assistant
Manages scheduling and correspondence for an executive, within the boundaries their calendar and inbox already have.
Executive Reporting
Rolls up operational data into board-ready views on a schedule, not a request.
MyntriqFlow
An agent's capability is composed from versioned skills — not shipped as a fixed feature list. Skills are added, updated and rolled back independently, so an agent's abilities grow without a platform upgrade, and a bad change is reversible in minutes, not a release cycle.
Want to see the architecture on your own use case?
Run it in a synthetic-data evaluation environment before it touches anything real.