Getting started
Setting up DocuMind is a sequence: the categories come first, then the scheme that groups them, then the project that uses that scheme. Automation is the last step and it is optional.
1Open the admin area
You will find four pages: AI Settings, Categories (labelled Kategorien), Teamwork Graph and Audit Logs. Only a Jira administrator can reach them.
2Create your categories
On the Categories page, select Add category. Start with the document types your teams actually receive: invoices, contracts, offers, test evidence. You can add more at any time.
3Group them into a scheme
Switch to the Category schemes tab and create a scheme. A scheme is the set of categories one kind of project is allowed to use, so a finance project and a support project can work with different lists.
4Assign the scheme to a project
Pick the scheme this project should use. Until a project has a scheme, DocuMind has no categories to offer there and the panel stays empty.
5Decide about automation
DocuMind is fully usable at this point: open a work item, attach a file, and assign a category by hand. If you want files sorted automatically, continue to AI providers. Automation is switched off until you turn it on.
Categories
A category is a label you can put on an attachment, plus the instructions that tell automation when that label fits.
| Field | What it does |
|---|---|
| Name | What people see on the attachment, for example Invoice & Billing. |
| Key | The stable identifier used in JQL and in the Rovo agent. Lowercase, hyphenated, for example invoice-billing. |
| Prompting | A description of when this category applies. Written for people, and read by the AI provider when automation is on. |
| Colour and icon | How the category is shown on the attachment. |
| Active | Deactivate a category to stop it being offered without deleting the assignments already made with it. |
Writing good prompting text
Prompting text is the single biggest influence on how well automatic sorting works. Describe the document and name the words that give it away:
Documents arrive in more than one language, so list the terms in every language your teams receive. DocuMind matches meaning rather than exact words, but naming the real vocabulary makes it considerably more reliable.
Extracting fields
A category can do more than label a file. Turn on extraction and list the values you want pulled out of the document, and DocuMind reads them from the file itself.
An invoice category might extract:
- Invoice number
- Customer / account name
- Contact person
- Contact email
- Due date
- VAT
- Currency
Field names are treated as business concepts, not literal labels. A field called Company name will match Rechnungssteller, Lieferant, supplier or vendor depending on the document. Formatting that matters, such as IBANs, invoice numbers and VAT IDs, is preserved as printed.
Extracted values appear under the attachment in the work item, behind Show extracted fields and values.
Writing values into Jira fields
A category can also map its extracted values onto fields of the work item itself. Turn on field mapping for the category and pair each extraction field with the Jira field that should receive it. DocuMind then fills those fields when it processes an attachment, so the information is available to boards, filters and automation rules like any other field.
Schemes and projects
Categories are defined once for the whole site. A scheme decides which of them a given project may use, and in what way.
| Per category in a scheme | Effect |
|---|---|
| Order | The order categories are listed in when someone assigns one. |
| Required on upload | The attachment is expected to carry this category; uncategorized files are flagged. |
| Allow multiple | Whether one attachment can hold more than one category from this scheme. |
One scheme can serve many projects. Changing the scheme changes every project mapped to it, which is the point: you maintain one list instead of twenty.
Working in a work item
The DocuMind panel sits below the attachments on the work item. Every attached file is listed with the categories it carries.
Assigning by hand
Select the attachment and choose a category from the scheme mapped to the project. If the scheme allows multiple categories, add as many as apply. You can clear categories from a file at any time.
Reviewing suggestions
When automation is set to Suggest only, DocuMind proposes categories and marks them as suggestions. Nothing is applied until someone accepts it. Each suggestion can be accepted or rejected on its own, or you can accept every suggestion on the work item at once.
When automation is set to Auto confirm above threshold, a suggestion the provider is confident enough about is applied directly, and anything below the threshold still waits for review.
Finding documents
DocuMind indexes what it knows about each work item's attachments so you can search for documents the way you search for anything else in Jira.
| JQL field | Searches |
|---|---|
documindCategories | Category names on the attachments |
documindCategoryKeys | Category keys, for exact matching |
documindAttachmentNames | Attachment filenames |
documindExtractedText | Values extracted from the documents |
documindSearchText | All of the above at once |
Examples
documindCategories ~ "Invoice" documindCategoryKeys = "invoice-billing" documindExtractedText ~ "EU-2026-1188" documindSearchText ~ "Acme GmbH" AND project = FIN
There is also a read-only field called DocuMind Search Text that you can add to a board or a filter view to see at a glance what a work item's attachments contain.
AI providers
DocuMind ships with automation switched off. When you turn it on, you choose which engine does the work, and that choice decides whether anything leaves your Atlassian instance.
| Provider | How it works | Data leaves Atlassian |
|---|---|---|
| Disabled | No automation. Categories are assigned by hand. | No |
| Rule-based | Matches on keywords, file extensions and MIME types that you define. No AI. | No |
| Atlassian Rovo | Uses Rovo in your own Atlassian instance. | No |
| Forge LLM | Uses the language model capability built into Atlassian Forge. | No |
| OpenAI | Your own OpenAI account and API key. | Yes — api.openai.com |
| Anthropic | Your own Anthropic account and API key. | Yes — api.anthropic.com |
| Google Gemini | Your own Google AI account and API key. | Yes — generativelanguage.googleapis.com |
Settings
| Setting | Meaning |
|---|---|
| Assignment mode | Suggest only waits for a person to accept. Auto confirm above threshold applies confident results directly. |
| Confidence threshold | How certain the provider must be before a category is applied without review. The default is 0.85. |
| Max categories | The most categories that may be applied to one attachment. The default is 3. |
| Allow no category | Lets the provider return nothing when no category genuinely fits, instead of forcing a weak match. |
Use Test provider after entering a key to confirm the connection before you rely on it.
The API key is yours. DocuMind does not resell AI capacity. You create the key in your own provider account, usage is billed to that account, and the key is held encrypted in Forge secret storage inside your Atlassian instance.
When you select one of the three external providers, the attachment's content, its filename and type, the work item and project identifiers, and your category instructions are sent to that provider so it can classify the file. Nothing is sent while automation is disabled or set to rule-based.
Rovo agent
If your site has Rovo, DocuMind installs an agent called DocuMind Agent. Open it from a work item and ask about the files attached to it.
- Review the attachments and get category recommendations, using only the categories the project's scheme allows
- Ask what a specific document contains and follow up on the answer
- Apply a category to an attachment through the agent
- Create a new Jira work item from a document, with the summary, description, type, priority and labels drawn from its content
The agent shows you what it proposes and creates a work item only after you confirm.
Audit log
Every change to categories, schemes and project mappings is recorded, together with automation results and provider tests. Each entry carries who acted, what changed, the work item involved and the time.
For automated assignments the entry also records which provider ran, how confident it was, and which categories it proposed, so a decision made weeks ago can still be explained.
Where your data lives
DocuMind is a Forge app. Everything it stores — categories, schemes, project mappings, assignments, extracted values and audit records — is held in Atlassian Forge storage inside your own Atlassian Cloud instance. teams2tech keeps no copy and has no access to it.
The files themselves are never copied. They stay as Jira attachments, where your existing project permissions already govern who may see them.
The one case where content leaves your instance is an external AI provider you have selected and supplied a key for. How that works is set out in the teams2tech privacy policy.
Support
Raise a request in the teams2tech support portal, or write to kontakt@teams2tech.com. Support hours are Monday to Friday, 08:00–18:00 CEST.
When reporting a problem with automatic sorting, the work item key and the attachment name let us find the matching audit entry quickly, which usually explains what happened.