Skip to content

Operators

Not chatbots. Operators with a job description.

An operator is a reusable template with a role, isolated persistent memory, explicit permissions, review requirements, and a budget. Operators are designed to cover distinct business functions — research, product, delivery, operations — and to coordinate as a company would. You configure the few your workflow needs; not every role is generally available yet.

Analyst

Evidence gathering

working
Memory
Sources, prior findings
Scope
Read-only research tools

Critic

Independent review

review
Memory
Claims under review
Scope
No write access

Executor

Governed action

blocked
Memory
Approved plan
Scope
Write, gated by approval

Anatomy

What every operator carries.

Role and scope

What the operator is responsible for, and the boundary it may not cross.

Isolated memory

Its own durable memory, separate from other operators and from shared org memory.

Permissions

Tools, data, and actions granted explicitly — read-only by default.

Review rules

What must be independently verified before it counts as done.

Budget

A spend ceiling per task and per workflow, enforced rather than advised.

Escalation path

Who is notified, and under what conditions, when the operator is blocked.

Collaboration

Coordination is a platform feature, not a prompt.

Operators work with each other inside Noria, where the interaction can be observed, constrained, and replayed. The direction: a small human team directs the work, and operators carry the coordination between functions.

Request work

An operator can ask another for an input it is not permitted to produce itself.

Hand off tasks

Handoffs carry context, constraints, and the reason for the transfer.

Preserve decisions

Decisions and their rationale persist beyond the session that produced them.

Promote memory

Findings become project or organization memory only through a governed step.

Setup

Guided configuration beats mass deployment.

Standing up forty roles produces noise and cost. Noria's guided setup starts from one workflow and adds operators only where the work demands one.