Compliance support
The Compliance module turns what MitDesk records into documents you can give to an auditor, a works council or a data subject.
Who this is for: administrators and compliance officers.
You need: Edition: Business. It needs the module minddesk_compliance. See Editions.
Not a certificate
These documents help you prepare evidence. None of them is a compliance certificate. Compliance depends on your legal basis, your contracts and your processes, not on software. Every document says this in its own text.
Who can use it
- MitDesk administrators can use everything below.
- The group MitDesk Compliance Officer can read the posture and evidence, verify the audit chain and create evidence exports. It cannot change retention policies or legal holds. It is a separate role. An administrator does not have it automatically.
The menu is Governance ▸ Compliance. The live overview is in Cockpit ▸ Compliance.
The compliance overview
Cockpit ▸ Compliance shows the current state, built from the live audit log. It has the buttons Refresh, Verify chain now and Accept gap and rebuild chain.
- Integrity shows the state of the audit chain: Not verified yet, Verified, New rows since verification, Rows awaiting chaining, Rows written during an upgrade or Integrity failure. Each state has a sentence that says what to do. The tab also shows when the last verification ran and how many times the guards blocked something in the last 30 days.
- Residency inventory counts your active AI models by hosting class: on-premises, EU cloud, non-EU cloud. See Sovereign routing.
- Retention & holds shows the retention for each kind of data and the active legal holds.
- Evidence documents links the latest stored documents and opens the exports below.
- Controls & fraud shows the result of an optional scan (see the end of this page).
If nothing has been verified, the overview says so. It does not assume the best. Run Verify chain now first.
Create a document
Open the export from the Compliance menu or from the Evidence documents tab. Each export is recorded in the audit log.
Processing-record support (Art. 30)
Choose the company and click Export document. You get a Markdown file built from your live setup, to help you prepare your record of processing under Article 30 of the GDPR. It lists:
- the agents as processing purposes, with their model, autonomy level and run-as account,
- digital employees, when the Workforce module is installed,
- what governs an AI action,
- the AI models as processors, with hosting region and a link to each provider's data-processing agreement,
- data categories, when Sovereign routing is installed,
- models outside the EU, as third-country transfers to review,
- the retention setting and the technical measures.
The document says that it is not a finished record and not legal advice. You decide who is controller and processor, and you need a legal review.
DSAR export
Choose the Data subject and click Export document. You get a PDF to help you answer an access request under Article 15. It has two parts:
- Operator side: the subject used the AI. You see their conversations (with message text), audit rows, jobs and approvals.
- Reference side: the AI worked on the subject's records inside other people's conversations. You see metadata only. Taking the subject's data out of someone else's conversation needs a manual check, so that other people's rights stay safe.
A text scan looks for mentions in other conversations. It is only a hint. It reads at most 200 messages and says when it hit that limit.
The document says that a person must review it before you give it out. It also repeats that the audit log stores a hash of the tool arguments, so it shows what the subject did but not which records each action touched.
AI governance report
Choose the company and a date range. The default is last month. Click Export report. The Markdown report is built only from the audit log:
- governance outcomes: executed, waiting for approval, refused, blocked from leaving,
- where model calls went, by hosting class and data class,
- activity by event,
- consent: how many approvals were granted and denied, and how long they took,
- guardrail stops,
- AI-content marking for Article 50 of the EU AI Act, grouped by source (own agents or external AI tools),
- usage and estimated cost by agent and by user. Costs are estimates from the prices you entered for the models.
If the marking field is missing, the report says so. It does not claim zero AI content. The report says that it is not a certification, and that its numbers are only as reliable as the part of the audit chain you can verify.
Egress attestation
Open Governance ▸ Compliance ▸ Egress attestation, choose the company and dates, and export. This document says what is known about data leaving your system, using positive evidence:
- a fresh verification of the whole audit chain, with the result,
- the guard setup that was active: the cloud deny list, and whether Sovereign Mode and its EU allowance are on,
- the number of guard events that fired: egress blocks, sovereign blocks and routing blocks,
- where the recorded model calls went, by hosting class and data class. MitDesk says which part of the period the records do not cover.
It does not promise that nothing left. The guard counts show that the guards work. They are not a count of what left, and no blocks does not prove that nothing went out. The model call records show where recorded calls went, but only from the date when they were first written.
If an operator signing key is set up, the export is a JSON bundle with a signature. Without a key it is a Markdown file marked unsigned. A small script in the module's tools folder can check a signed bundle offline. A signature shows who signed the bytes. It is not a qualified signature or a trusted timestamp.
The evidence register
Exports you make by hand are not kept. The Evidence register keeps the monthly documents.
- A daily job creates the AI governance report and the Egress attestation for the previous month, for each company, if they do not exist yet.
- An administrator can click Generate missing previous month.
- Each record stores the file and its SHA-256. There is one per company, document type and period. Records and files cannot be edited or deleted.
- Download gets the file. Verify (or Verify file integrity) checks it against the stored SHA-256.
Erasure requests
Open Governance ▸ Compliance ▸ Erasure requests to remove a person from the copies MitDesk made. Deleting the contact in Odoo stays your own process. This feature removes the person from MitDesk's own stores: chat lines, tool results, queued arguments, knowledge entries, memory and notes of digital employees.
- Create a request, choose the Data subject (a person, not a company) and write the Legal basis.
- Click Preview. MitDesk shows how many rows each store would change.
- Click Erase. MitDesk changes those rows and keeps the counts, not the removed text.
Good to know:
- A job in the queue or an approval that names the person stops the erasure.
- A legal hold keeps its rows. Those rows are left as they are and counted. The rest goes on.
- The audit log is not rewritten, because it proves that the erasure happened. MitDesk writes one
erasure_donerow. - Running a finished request again removes only what arrived since then.
Optional scan: controls and fraud
The Controls & fraud tab shows an optional scan of Accounting and Point of Sale data, if those apps are installed. It uses fixed rules, not guesses: duplicate vendor payment references, large posted journal entries, cash-count differences, large refunds and high discounts. It shows the number of findings and a risk band (low, medium, high). It reads only what the user may read. Set the look-back period and the thresholds in the settings (block Controls & fraud scan). The scan helps a reviewer. It does not decide that fraud happened.
Related
- Audit and evidence: the record behind every document here.
- Sovereign routing: the data classes and routing decisions.
- Running it on your own hardware: the guards that the attestation reports.
- How consent works: the approvals that the report counts.
- Autonomy policies: where the autonomy level in the record is set.
- Permissions and groups: the groups behind these menus.