Who this is for. Super Admins configuring an Agent Co-worker. No other role has access to the setup wizard. Managers still own the decisions these settings record, and are the people to ask when a setting is wrong. Requires the AP Helpdesk function enabled on the agent. See the Roles and Access.
Autonomy Level is the first capability group on the AP Helpdesk tab of step 3 when the autonomy-level feature is enabled for your instance, or when the Agent Co-worker already runs in Collaborative Mode. It asks: “How much of the helpdesk-processing work should this agent attempt on its own?” The answer decides which of the two autonomy models the rest of the configuration follows.
If the feature is not enabled, the group does not appear, Contact Management is the first group, and the agent runs with Governed Autonomy.
The autonomy level you select rebuilds the capability rail beneath it. The two models list different sections, some of them common to both, so the configuration articles are split by model. For the section-by-section difference, see Collaborative Mode Overview.
Navigate to Autonomy Level
On the Define capabilities step, select the AP Helpdesk tab.
In the left navigation, select Autonomy Level. The page asks: How much of the helpdesk-processing work should this agent attempt on its own?
Figure 1. The Autonomy Level section on the AP Helpdesk tab.
Mailbox analysis
The Mailbox analysis panel sits at the top of the Autonomy Level page, above the autonomy question. It samples the most recent 30 days of inbound email from the mailbox connected to this agent, up to 500 emails, and reports what that mail looks like. The level you choose is then measured against your own traffic rather than the shipped defaults.
The panel needs a mailbox before it can run. Until one is selected in the Mailbox field on the Work scope step, the button is unavailable and the card reads “Select a mailbox in the working-scope step to enable the analysis.”
The sample is read-only. No mail is sent, changed, or forwarded by the analysis.
Run the analysis before you work through the rest of the AP Helpdesk rail, so the categories and contacts you review reflect real traffic.
Select Run mailbox analysis to start a run. A progress bar reports the phase it is in, from sampling the mailbox, through classifying the emails, to discovering new email categories, with a running count of emails processed. Stop analysis ends a run that is still going without changing any categories.
When a run finishes, a line reports how many emails were sampled and the window they came from, and the button changes to Run analysis again. A collapsible Mailbox & SOR analysis summary opens beneath it with five tiles: Emails sampled, High-confidence classification, Known-sender share, Business emails and Non-business, all computed across the connected mailbox and the system of record.
The run also feeds the sections that follow. Contact Management shows a sender and message count on each sender-type card, and selecting a count opens a read-only list of the emails behind it. Only the most recent completed run supplies those figures, so they are not shown while a re-run is in progress, or after a run fails or is stopped.
Figure 2. The Mailbox analysis panel.
Where the summary sits
Open the agent’s Define capabilities step, select the AP Helpdesk tab, then select Autonomy Level. The Mailbox analysis panel sits above the autonomy choice.
Select Run mailbox analysis to sample the most recent 30 days of inbound mail, up to 500 messages. Once a run has finished the control reads Run analysis again, and a line beside it reports the outcome, for example whether any new categories were found. The results open under Mailbox & SOR analysis summary, which collapses, so on a returning visit it may be closed.
The sample is read-only. No email is sent, changed, or forwarded by the analysis.
Figure 3. The Mailbox analysis card, after a run.
The five tiles
Each tile summarizes one finding from the sample. Read them as evidence, not as something to configure.
Table 1. Mailbox analysis summary
Tile |
Reports |
|---|---|
Emails sampled |
How many messages the analysis read, over a stated window of days, across the connected mailbox and the system of record. |
High-confidence classification |
The share of sampled mail where an intent was detected above the confidence threshold. The threshold itself is printed under the figure. |
Known-sender share |
The share classified as internal, vendor, or system senders. |
Business emails |
The share falling into actionable helpdesk request categories. |
Non-business |
The remainder: marketing, events and system noise. |
Read the figures
Each tile answers a different setup question, and a low number is not automatically a problem.
Emails sampled near zero. Either the mailbox is genuinely quiet or the connection is not returning mail. Confirm the mailbox on the instance row on the Agent Co-workers page, and ask the Super Admin who created the connection whether it is returning mail, before reading anything else. The other figures are computed from this sample, so a small sample makes them meaningless.
Low known-sender share. Expected before you have configured contacts and system senders. That is what Contact Management is for, and it should rise as you work through that section.
Low business-email share, high non-business. Usually means the mailbox carries a lot of newsletters, notifications and internal traffic. It is a signal about which categories to disable.
Low high-confidence classification. The agent is seeing mail it cannot place. Look at the Suggested Email Categories tab, where the patterns behind a low figure often surface.
Known-sender share is computed from Auditoria’s own per-message sender typing rather than from the system sender configuration alone. That matters on a new tenant: a figure derived from the configuration would read as 100% known when no configuration exists yet.
Two models
The product’s own descriptions make the positioning plain: Governed Autonomy carries the Recommended badge, Collaborative Mode the Legacy one.
Table 2. Autonomy models
Model |
Badge |
Described in the product as |
|---|---|---|
Governed Autonomy |
Recommended |
An intelligent AI agent architecture that replaces deterministic, rule-based automation with agentic reasoning. The agent earns autonomous authority through demonstrated performance, evolving from a system that follows fixed scripts into a configurable partner that adapts to your operations. |
Collaborative Mode |
Legacy |
A maturity phase in the Helpdesk automation framework where the AI performs complex processing and generates a complete response or action plan, but requires final human review and approval before every execution. |
Collaborative Mode is being deprecated. The product states it directly: “Collaborative Mode is being deprecated. Existing configurations will continue to run, but new setups should use Governed Autonomy.” Choose Governed Autonomy for anything new.
Select an autonomy level
-
Select one option:
Governed Autonomy, labeled Recommended. An intelligent AI agent architecture that replaces deterministic, rule-based automation with agentic reasoning. The agent earns autonomous authority through demonstrated performance, evolving from a system that follows fixed scripts into a configurable partner that adapts to your operations.
Collaborative Mode, labeled Legacy. A maturity phase in the Helpdesk automation framework where the AI performs complex processing and generates a complete response or action plan, but requires final human review and approval before every execution.
Select Next.
Which to choose
Choose Governed Autonomy for any new Agent Co-worker. The application recommends it, and it is the only model with Email Categories, Processing Rules and Task SLA available.
Choose Collaborative Mode only when maintaining an existing configuration that already uses it. The application displays a deprecation warning on this option.
What the selection changes
Your choice determines which sections appear in the left navigation.
For which sections each model lists, see Collaborative Mode Overview.
Level changes on an active Agent Co-worker
Switching autonomy level changes which sections are available, and reply content does not transfer between the two models. Collaborative Mode templates are literal text with variables; Governed Autonomy replies are composed from response guidance. Review what needs rewriting before switching an active Agent Co-worker.
For more information, see Maintain Agent Co-workers.