Roles, file formats, limits and retention periods in one place. This page collects values that would otherwise appear separately in each chapter and drift apart there over time. You find explanations, settings and step-by-step instructions in the relevant feature chapter; this page only states what applies.
This is the only page in the guide where tables are intended. It is built for looking things up, not for reading through.
Roles and permissions
Datenmassiv AI has five roles. Owner and Administrator always have all permissions, regardless of the table below. Editor and Member have the permissions that your administrator assigns to them under Settings → Roles & permissions. Viewer has none of the permissions listed below and cannot be configured either: there is no toggle for this role.
| Role | Configurable permissions |
|---|---|
| Owner | All, plus actions reserved for owners: appointing further owners or changing their roles, and accepting the DPA. Deleting the organization is done by the operator (see Settings). |
| Administrator | All, fixed and cannot be restricted |
| Editor | See the table “Configurable” below |
| Member | See the table “Configurable” below |
| Viewer | None, cannot be configured |
Owner and Administrator only
These five permissions cannot be enabled for Editor and Member, so that nobody can grant themselves higher rights.
| Area | Permission (interface) | Key |
|---|---|---|
| Workspace | Manage settings | workspace:settings |
| Team | Manage members | team:manage |
| Team | View audit log | team:audit |
| Team | Configure roles | roles:configure |
| AI usage | View usage | usage:view |
Why “View usage” is locked: The organization-wide usage view shows AI activity per employee. In Germany that is an evaluation subject to co-determination by the works council (Section 87 BetrVG / Section 26 BDSG), and it is therefore limited to Owner and Administrator. Every member still sees their own figures under My usage.
Configurable for Editor and Member
Your administrator can switch these ten permissions on or off individually for Editor and Member under Settings → Roles & permissions. The columns show the factory settings of a newly created organization; your organization may differ.
| Area | Permission (interface) | Key | Editor (default) | Member (default) |
|---|---|---|---|---|
| Workspace | Edit company context | workspace:description | ✅ | ❌ |
| Knowledge bases | Create & manage knowledge bases | kb:manage | ✅ | ❌ |
| Knowledge bases | Upload documents | kb:upload | ✅ | ❌ |
| Chatbots | Manage chatbots | chatbots:manage | ✅ | ❌ |
| Automations | View workflows | workflows:view | ✅ | ❌ |
| Automations | Edit workflows | workflows:edit | ✅ | ❌ |
| Automations | Run workflows | workflows:run | ✅ | ❌ |
| Automations | Share workflows | workflows:share | ❌ | ❌ |
| Templates | Manage templates | templates:manage | ✅ | ❌ |
| Transcription | Delete other people's recordings | transcription:delete | ✅ | ❌ |
File formats
Company profile
The AI fill-in assistant allows each organization at most five requests per one-hour window and ten per 24-hour window. The time windows begin with the first counted call; they do not correspond to the calendar hour or the calendar day. Generation attempts that fail after approval also count; switching users or starting a new process ID does not reset the counter. During a pause the dialog shows the waiting time, while you can keep editing and saving the draft manually. The company context instructions explain how to maintain it later.
Document chat
| Format | Extension | Note |
|---|---|---|
| Only with readable text: scanned PDFs without a text layer are detected and rejected | ||
| Word | .docx | Microsoft Word |
| Text | .txt | Plain text files |
| Markdown | .md, .markdown | Markdown-formatted text |
Knowledge bases
| Format | Extension | Note |
|---|---|---|
| Scanned PDFs are read automatically by vision AI | ||
| Word | .docx | Microsoft Word (Office Open XML); legacy .doc (Word 97-2003) is not supported |
| Excel | .xlsx | Each worksheet as a Markdown table, formulas as their calculated result; legacy .xls is not supported |
| Text | .txt | Plain text files |
| Markdown | .md, .markdown | Markdown-formatted text |
| Image (PNG) | .png | Indexed by a vision model |
| Image (JPEG) | .jpg, .jpeg | Indexed by a vision model |
| Image (WEBP) | .webp | Indexed by a vision model |
Transcription
MP3, WAV, M4A, WebM, OGG and FLAC.
Languages: German, English (US), English (UK), French, Spanish, Italian, Portuguese, Dutch, Polish or Detect automatically.
Export: Word (.docx), PDF, text (.txt), subtitles (.srt, .vtt) and raw data (.json); transcript and summary also to the clipboard.
Limits
AI Chat (attachments)
| What | Limit |
|---|---|
| Images per file | 20 MB, automatically scaled down to the model's resolution before sending |
| PDF per file | 3 MB |
Microsoft 365 in chat, agents and workflows
The document integration uses personal business accounts and storage locations approved by owners or administrators; personal Outlook mail and your own Outlook calendar are each approved separately for read access. Connecting an account does not automatically add its OneDrive. The retention period for chats must be finite; only the owner may change it. Your organization's model approvals also apply to Microsoft content in chat, agents and workflows. EU providers only limits processing to approved models in EU regions; Global (incl. non-EU) also allows approved models outside the EU. You see the framework that applies on the integration page; no additional Microsoft approval is needed for global models. Personal workflows require the corresponding activation; if it is missing, the application shows a notice. In AI Chat and in workflows, suspicious content or content that cannot be checked reliably is held back and does not appear as a source used. In agent tasks the same applies to emails and appointments; as soon as a task has searched your Microsoft sources, it no longer fetches web pages and no longer searches the web. A personal workflow and its versions remain private even after the Microsoft step has been removed.
| What | Limit |
|---|---|
| Readable files | DOCX; text PDF only when activated, as stated on the integration page |
| Not included | Excel, PowerPoint, old DOC files, scanned and encrypted PDFs |
| Search terms | At most three per search step; each resolved search term contains 1 to 120 characters |
| Files per question | At most three |
| File size | 20 MiB per file, 40 MiB in total |
| Long documents | Limited text excerpts may be read instead of the complete document |
| Agent tasks | Choose the source explicitly when starting a new conversation; documents, your own Outlook mailbox and your own calendar with the same limits as in chat; follow-up questions stay private and read again |
| Workflows | Personal manual workflow with Microsoft documents and AI text; at most twelve steps with one Microsoft source step, no team sharing or automatic execution |
| Other tools | Public chatbots and automatic or shared workflows do not get personal Microsoft access |
For Outlook mail, each question allows at most two search rounds with up to three alternative word groups each, twenty hits in total, eight read attempts and 60,000 characters of processed mail context. The word groups are searched together; for an empty search with a message type and a received period, additional candidates may be searched based on these two limits alone. At most four search queries are run in total; candidates are not yet evidence for the topic you are looking for. Current individual messages are read from your personal business mailbox; a complete conversation is not guaranteed. Mail attachments, shared mailboxes, sending, mailbox changes and mail access in workflows are not included. The mail instructions explain the approval, choosing sources and taking over replies yourself.
For Outlook calendar, each question allows at most three calendar requests covering 1 to 31 whole days each, at most 366 days before or after today. One request lists at most 25 appointments, one question at most 30 in total; descriptions are read for at most six appointments, together with the appointment data up to 60,000 characters. For each appointment at most 30 attendees are listed individually; rooms do not count as attendees. Only your own default calendar is read; shared calendars, other people's availability, appointment attachments and any change to appointments are not included, and workflows do not read a calendar. The calendar instructions explain the approval and choosing sources.
The AI Chat chapter explains how to use it and how to check the original source. The agent instructions explain the selection and private follow-up questions. The workflow instructions describe personal manual use.
Document chat
| What | Limit |
|---|---|
| Documents per analysis | 5 |
| Total pages (across all documents) | 50 |
| File size per upload | 10 MB |
| Number of analyses | No hard limit; an unobtrusive clean-up notice from about 30 onward |
Knowledge bases
| What | Limit |
|---|---|
| File size per file | 20 MB |
| Storage per organization | 2 GB (default; the system admin can raise it for each organization) |
| Max. URLs per import job | 100 |
| Max. URLs that discovery finds | 500 |
Embed chatbots (monthly quota)
Every message a visitor sends to the chatbot counts; the quota applies per calendar month (German time) and unused messages expire. With booked packages the quota applies per chatbot; the trial quota applies to the whole organization and stays used up even if you delete or create chatbots.
| Package | Messages per month |
|---|---|
| No package (trial quota) | 250 per organization; only one active chatbot, the widget shows Powered by with the Datenmassiv AI logo |
| Package S | 4,000 |
| Package M | 12,000 |
| Package L | 32,000 |
Transcription
| What | Limit |
|---|---|
| Max. recording length, file and microphone | Up to 4 hours; the smaller of two limits applies: the length set in your organization and the technical maximum of 240 minutes |
| Max. file size | 500 MB |
| Max. length of a live session | 2 hours |
How much of the quota models use
Your plan includes a monthly quota. How quickly it runs down depends not only on how much you write but also on which model you choose. That is why a usage level may appear to the right of a model in the model picker.
The level always applies to your organization, the current quota period and the area in which you use the model. Chat and automations count against a shared AI quota; the last 10% is reserved for chat. In the current period the same model can still carry different levels in the two pickers, as long as your quota dates from the time before the shared quota. Without a usage rule that applies to this area, no level appears.
| Level | Usage weight in the applicable quota area |
|---|---|
| Low | below 5, including use without quota deduction |
| Medium | from 5 and below 15 |
| High | from 15 and below 45 |
| Very high | from 45 |
Your administrator sets up which models are available to you, so your list may be shorter.
Two things that often cause confusion:
The usage level has nothing to do with speed. In the model picker the models are grouped by suitability (Fast, Powerful); the Fast group also contains models with high usage.
A more expensive model does not deliver a better result for every task. For wording, summaries and standard questions an economical model is almost always enough; the strong models pay off for long documents, multi-step analyses and tricky specialist questions.
Owners and administrators see under AI usage → Monthly quota the quota, the used and reserved units and the remaining units, with the share of the chat reserve for the usage period actually provided. Subscription & billing additionally shows guide values for the remaining usage. When the organization approaches the limit, a ring with a percentage also appears right next to the model name in the chat.
Where your data is processed
This question comes first in every public authority and every company with a staff council, so the full answer is here in one place.
The principle: In standard mode your AI processing runs exclusively in EU regions. Database, file storage and email sending are located in the EU.
Whether models outside the EU are also available is set by Datenmassiv AI for your organization, not by your own administrator. Under Settings → AI models you see the framework that applies as a label: EU providers only or Global (incl. non-EU).
Three services work outside the EU regardless. None of them gets to see the documents you upload:
| Service | When it runs | What it receives |
|---|---|---|
| Web import into a knowledge base | only when you import an address | the address you enter and the publicly available content of that page. Read only |
| Calendar lookup by a chatbot | only when you have connected a calendar | free and busy times of the connected calendar. Read only, no bookings |
| Audience measurement of the public website | only on our website, not in the application, and only after your consent | visits to public pages |
None of these three services receives your documents, chat histories, recordings or knowledge bases.
Workflow steps
A workflow is built from steps. The palette lists 17 steps in four groups; the Code step among them is reserved for system administrators.
| Group | Steps |
|---|---|
| AI | Agent, Read document, AI text, Look up knowledge, Microsoft documents |
| Logic & flow | Condition, For each item, Repeat |
| Actions | Email, Web request (HTTP), Create Word document, Save knowledge, Code |
| Output & review | Create file, Guardrails, Review, Delay |
The “Code” step is reserved for system administrators. It appears in the palette, but a workflow that contains it cannot be saved or run by anyone else. If a workflow does not start for no apparent reason, this is the first place to look.
Inside a container, additional values are available under Variables.
In For each item these are {{loop.currentItem}}, {{loop.currentIndex}} (starting at 0) and {{loop.total}}.
In Repeat, {{loop.iteration}} is the pass (starting at 1) and {{loop.feedback}} is the review's comment from the previous pass, empty in the first pass; the variable picker offers it as soon as the container contains a review.
Values in text
The review note, the subject and text of an email, the Word document, the file (Word, PDF, text) and Save knowledge are texts for people.
There, numbers and {{sys.date}} appear in the Number and date format of the person who starts the run (My account → Language and format); for a schedule or a form, that is the person who created the workflow.
If “€” follows the variable directly or after a space, the result is an amount: with an English (UK) format, {{…betrag}}€ gives “€17,790.50”.
If “%” follows it, the result is a percentage: {{…mwst_satz}}% gives “19%”.
A yes/no field appears as “yes” or “no”, a list or field group with field names instead of as JSON; if it stands alone on a line, each entry gets a line of its own.
A field that a step could not fill appears as “not recognized”.
A value that a step delivers as text, such as a postal code, stays unchanged.
A form field of type Number arrives in the run as a number, one of type Yes/No as a yes/no field, even if the form also sends a file; an empty number field counts as not filled in.
Conditions, AI steps, search and web requests receive the value unchanged so that calculations and comparisons are correct; Excel and CSV keep numbers as numbers.
By default a review waits 72 hours for a decision. Under Waiting time (hours) you can set 1 to 168 hours. An old approval link does not extend this period.
Word template for quotes
For Create a quote from a customer inquiry, you upload your DOCX template during setup. Placeholders are written in the Word text in single curly braces. The mapping that ships with it uses the following names:
| Placeholder | Content |
|---|---|
{kundenname} | Customer name from the form |
{datum} | Quote date |
{gesamtbetrag} | Proposed net total amount |
{lieferzeit} | Proposed delivery time |
{gueltig_bis} | Proposed validity |
The quote template is designed for a list of finished line items.
With {.} you output the text of one such line.
Write {#positionen}, {.} and {/positionen} each in a paragraph of its own, so that every proposed line item gets its own line.
After the upload, the setup shows which placeholders of your template would remain without data.
If the template contains a placeholder at run time for which the step has no data, no document is created; the step names the placeholders concerned.
An arbitrary company template is not adapted automatically.
Check the generated document before approving it.
The agent receives the default “30 days from the quote date” if the price list does not state a validity; that does not replace your check of the terms.
Workflow runs
Schedules start at most once a day. Deferred runs are checked every five minutes, oldest first and at most 25 per organization per pass. After more than 60 days without a start they are discarded. A restart is therefore guaranteed neither at a specific time nor for every backlog. Discarding a backlog requires “Run workflows”. Only Owner and Administrator may set the organization-wide Emergency stop or lift it with Resume workflows.
Retention
| What | Period | Is it enforced? |
|---|---|---|
| Chat histories & agent tasks | Configurable per organization: 7 / 30 / 90 / 365 days or unlimited, default 365 days | Yes, daily |
| Agent attachments not yet sent | Immediately via Remove file; otherwise from 24 hours after the upload | Yes, daily |
| Sent agent attachments | As long as a saved task uses them; cleaned up afterwards, at the earliest 24 hours after the upload | Yes, daily |
| Audio files (transcription) | 30 days; the transcript itself is kept, only the audio file is removed | Yes, hourly |
| Workflow files from completed runs (intermediate files, form uploads) | 30 days, failed runs 7 days; if a run was retried, the end of the last retry counts | Yes, hourly |
| Form uploads that did not lead to a run | 1 hour | Yes, hourly |
| Result files from successfully completed workflow runs | Default 90 days; a different period per organization is possible | Yes, daily |
| Chatbot conversations | Configurable per chatbot: 7 to 365 days, default 90 days | Yes, daily |
| Security log (audit log) | 90 days | Yes, daily |
Knowledge base documents, document chat analyses and workflow templates have no automatic deletion period. They are kept until you delete them yourself. Workflow result files stored in Datenmassiv AI, by contrast, are subject to the period stated above. Copies already downloaded to your device are not removed by this.