Agent data and privacy

The Monitoring Agent uses the same Azure OpenAI Service as the Polysync Copilot, under the same commitments. This page lists exactly what it sends, what it stores, and the controls you have. For the legal wording see the Privacy Policy.

What is sent on each sweep

Only a fixed set of data classes is read, and each is either always on or switchable:

Data class	What it contains	Control
Run summary	Task and platform names, statuses, timings, run ids and retry history	Always on
Error message	The platform's error text, after known secret patterns have been removed	Can be switched off
Configuration metadata	Non-secret task and platform settings, pipeline and job names, attribute values	Can be switched off
Platform health and contention	Queue and lock state, platform availability	Always on
Platform error detail	Activity-level error detail and run output fetched from the platform after a failure, for the failing run only, cleaned like error messages and not kept beyond the insight it supports	Off by default. An administrator can switch it on, for platforms that support it
Earlier statistics	Counts and outcomes of earlier similar failures in your organisation	Always on
Documentation excerpts	Polysync's own product documentation	Always on

What is never sent

Parameter values passed to a task.
Resolved secret vault values, and any credential.
Anything identifying a person: no user names, no e-mail addresses, no identifiers.

The copilot and the agent never send these, whatever the settings say. The error-message switch exists because some platforms echo fragments of connection details into their error text; removing known secret patterns is best effort, so a tenant that would rather not take that risk can switch error messages off for the agent entirely.

What agent chat sends

When you ask the agent a question, your message is sent to the model, together with the same kinds of data the sweep reads, fetched only when the assistant decides it needs them: run summaries, run statuses and timings, sanitised error text, task and platform configuration metadata, platform health, contention state, earlier similar failures, your organisation's active agent rules and the product documentation. Parameter values, resolved secret values and anything identifying a person are never sent, and the same data-class switches apply.

Where it is processed

The agent calls a regional model deployment provisioned in Australia East, so requests made by the agent are processed within Australia. The Copilot and the Ask Polysync assistant use the same regional deployment. The Privacy Policy explains the detail.

Microsoft processes the requests as a sub-processor under the same terms as the rest of the service: the data is not used to train the model, and prompts and completions are retained for up to 30 days for abuse monitoring, then deleted.

What is stored, and where

Insights, rules, failure statistics and feedback are stored in your organisation's own data area. They are never placed in a shared or public location, and no other tenant can see them.
Run summaries and the action audit trail are stored alongside them. The audit trail records each agent action with who or what caused it and when.
Your agent preferences are stored against your tenant user id, so they follow you across devices: summary style, what the dashboard shows you, notification overrides, digest time and the evidence panel state. Your personal guidance is stored as a personal agent rule, so it appears on the rules page with its author and can be edited or deleted there.
One counter, no message text. A daily row per person holds the tenant id, the tenant user id, the UTC date, a request count and the time of the last request. It exists only to enforce the daily chat limit. It contains nothing you typed.
No conversation transcripts are stored. The Ask the agent panel and the Copilot both keep the conversation in the browser session only, and nothing is replayed on a later visit.

Insights are kept for the retention period stated on the agent card in Organisation Settings, which comes from your plan, and are removed by a daily retention pass. Feedback is removed with the insight it belongs to.

Your controls

AI features switch. One switch in Organisation Settings turns off the Copilot and the agent together. With it off, no sweep is claimed, no model call is made and no AI data is written, and anything already queued is abandoned rather than completed.
The agent switch. The agent has its own on/off control and is off by default. Enabling it requires recorded consent to scheduled AI processing.
Data classes. Error messages, configuration metadata and the optional platform error detail can each be switched off or, for the platform error detail, on.
Your own rules and preferences. You can edit or delete your personal guidance and change your preferences at any time on Rules and preferences. An administrator can disable or delete any shared rule.
Pause. Polysync can pause the agent for your organisation on request. A paused agent does not sweep until it is resumed, and no other setting changes.
Delete agent data. An administrator can delete all insights, feedback, learned rules and failure statistics at any time from the agent card. Deletion is permanent, happens immediately, and keeps only the run summaries and the audit trail that security and change control require.

Documentation