Who this is for. Everyone. The terms defined here are used throughout the guide.
Access. No permission required. See the Roles and Access.
The terms the rest of this guide assumes, in the order they build on each other rather than alphabetically. Read them once in sequence: each one depends on the one above it, and the chain is what makes the product predictable.
For the full alphabetical list, use the Glossary.
Terms, in dependency order
Table 1. Core vocabulary
Term |
What it means |
Why it matters to you |
|---|---|---|
Agent Co-worker |
The configured AI worker that reads mail and documents and acts within limits you set. |
Everything in the console belongs to one agent at a time. |
Work scope |
The system of record and the boundary an agent operates in. Fixed when the agent is created. |
It cannot be changed later. Getting it wrong means recreating the agent. |
Autonomy level |
How far the agent may go without a person. |
Decides whether you see finished work or a decision to make. |
Sender type |
How the agent classifies who sent a message: External, Internal, System or Unknown. |
Sets what the agent is allowed to do in reply. It is an authorization boundary, not a label. |
Email category |
What the message is about. |
Drives which processing rules apply. |
Processing rule |
A condition and an action. |
What the agent does once it knows the sender and the category. |
Task |
A unit of work the agent could not finish alone. |
Your queue is made of these. |
SLA |
The age threshold that makes a task at-risk and then overdue. |
Drives the priority order you work in. |
How they chain
Read the chain in one direction and most product behavior becomes predictable:
Work scope decides what data exists → sender type decides what the agent may do → email category decides which processing rules apply → autonomy level caps how far those rules may go → whatever is left becomes a task, aged against an SLA.
When something unexpected happens, walk the chain from the top rather than from the symptom. A reply that was not sent is usually stopped by the sender type or by the autonomy level rather than by the rule that drafted it.
Commonly conflated terms
Table 2. Commonly conflated terms
These look similar |
But |
|---|---|
Sender type and email category |
Sender type is who and governs permission. Category is what about and governs behavior. |
Autonomy level and confidence |
Autonomy is how far the agent may act, and you set it. Confidence is how well a value was read, and the agent reports it. A perfectly read invoice from a disputed vendor still reads as high confidence, so confidence is not a judgment about whether the invoice should be paid. |