Release notes

​

Nobly Insight — mid-October update

​

✨ Admin assistant: ask about your configuration, approve what it changes

• An assistant in the admin settings (opt-in) — select Admin assistant on any admin settings page and ask how something is set up (“which workflows use this script?”), or ask for a change in plain words (“make Contract value required and searchable”). It starts from the page you are on and can work across users and access, document types, keywords, Caseflow, form templates, workflows, scripts, system searches, retention, reporting, AI settings, general settings, navigation and banners. See admin assistant.
• Nothing changes until you approve — a proposed change appears as a card showing what changes, before and after. Approve or reject it in the panel, or later on the new Approvals page under Settings → AI Admin Assistant. The assistant can only propose what you could do on the admin page yourself, every change passes the same checks as the page, and a change is refused rather than written over a colleague’s newer edit. See approving changes.
• You decide how far it may go — the assistant is off until an administrator switches it on under Settings → AI Admin Assistant → Settings, and it is offered only to user groups holding the new Use AI Admin Assistant permission. For every operation the environment chooses Off, Always ask or Auto-apply allowed; deletes, changes to access, security keywords, retention and script bindings start at Always ask. A change only applies without a card where the environment allows it and the person has also switched it on for themselves under Assistant rules. See what the assistant may do.
• Every change on record — changes made through the assistant are recorded in the configuration audit log as AI on behalf of the person who asked. To undo one, ask the assistant: it reads the values from before the change and proposes putting them back. Script and workflow secret values are never shown to it.
• Help in the editors, previewed before you approve — the workflow designer, the script editor and the Caseflow and form template editors open the admin assistant with Ask the admin assistant, and it sees the item you have open, including unsaved edits. Preview in editor loads a proposal into the editor as unsaved edits so you can check it before approving. This replaces the AI panels these editors had until now, so switch the admin assistant on if your administrators used them. See preview in the editor.
• The same tools for your own AI assistant (opt-in) — an AI assistant such as Claude, connected through AI assistant access, can now read and change configuration with the same operations, rules and approvals as the admin assistant. Its changes wait on the Approvals page marked MCP, and a small, simple batch can be approved in the assistant’s own dialog where it has one. It needs Use AI Admin Assistant and the configuration permissions for each area. See changing configuration.

​

⌨️ Keyboard & dates

• A command palette on every page — press Ctrl+K (⌘+K on macOS) anywhere in Nobly Insight to jump to a page, a saved search or an action instead of navigating menus. In Caseflow it also lists the application’s saved searches, and typing # looks up an object by its ID.
• Shortcuts you can change — press ? or choose Keyboard shortcuts in the user menu to see every shortcut with the keys it currently uses. Under Personal settings → General settings → Keyboard shortcuts you can change the keys for an action, switch a shortcut off or reset them all; your choices are saved to your user account and follow you to every browser you sign in on. See using keyboard shortcuts.
• Results from the keyboard — the arrow keys move through search results and load the preview, Enter opens the row, Space selects it, Shift with the arrow keys extends the selection and Ctrl/Cmd+A selects all. Ctrl/Cmd+S saves the Caseflow form or keyword panel you are working in, and Ctrl/Cmd+Enter sends a note or confirms a dialog.
• Type dates and times into every date field — date and date-and-time fields across Nobly Insight now take typed entry in your environment’s format, such as 03.04.2024 or 03.04.2024 14.30, with the calendar still beside the field. This covers search filters, upload rows and keyword values, Caseflow tables and templates, retention criteria, task forms, dashboards and mailbox settings. A date that cannot be read is flagged in place, and the search, save or upload it belongs to waits until you correct it or press Escape to discard it.

​

🕐 Dates on your calendar

• Today means your today — “Today”, “Yesterday” and the other date presets in document search now count from your environment’s own calendar, so the first hours of your morning no longer fall into yesterday. Date filters cover every hour of the day you picked, dates in shared and saved searches no longer shift by a day when reopened, and exports are named with your local date. A “today” default on a Caseflow attribute and date macros in Caseflow constraints resolve to the same day.
• The whole end day — a custom date range in a search now includes everything filed on its end date, not just the hours up to midnight. Searches run through the API and by an AI assistant on your behalf follow the same rule.
• Documents dated when they arrive — a document created without a date, whether uploaded, created from a form or built by document composition, now carries your local date and time, and the received and sent times the mailbox importer records are your local times.
• Stored dates on one clock — the stored date, revision dates and deletion time of a document now come from the same clock as the rest of your archive, and stored-date filters and the archive trashcan’s deletion-date filter compare against it, so a day’s documents are no longer missed or listed under the wrong day. Dates already recorded are unchanged.
• Auto-names with the right stored date and time — the stored-date and stored-time tokens in an auto-name now show your environment’s local time instead of a time one or two hours behind. Existing names keep what they show; a name written in another date format by an older client is rewritten in your configured format when its document type is recomputed, so check the affected count before recomputing a large type. See document types.
• Security keyword rules on your calendar (opt-in) — a security keyword rule using the Today placeholder now switches at your local midnight rather than in the middle of the night. See security keywords.
• One time zone per environment — your environment is set to the zone its users work in, Danish time unless it was configured otherwise. If your users work in a different zone, ask your consultant team to change it.

​

🎨 Caseflow

• Long text attributes — administrators can now create a Long text attribute for free text of any length, such as a decision rationale or a note. It takes no length or default value, templates edit it as a multi-line field, and it can be used in saved-search filters, sort order and custom searches. See applications and classes.
• Purge deleted objects — a class under Admin settings → Caseflow → Applications & classes now lists the objects deleted from it, with who created and deleted each one and when, and lets you purge one or the whole page for good. A class whose deleted objects are the only thing in the way can now be deleted, after confirming the purge. Purging keeps the documents themselves, is refused while an active object still relates to the deleted one, and cannot be undone.
• Search as you type in value lists (opt-in) — a bulk-edit value list taken from a Caseflow search can be set to Search as you type instead of loading the whole list, so very large value sets become practical. The same lookup serves the bulk edit field and search prompts built on that column. See Caseflow administration.
• Dropdowns in search prompts — when a saved search prompts for a value and that attribute has a value list configured, the prompt now offers the list as a dropdown instead of a text box. Prompts using contains, in or sounds like keep their text box.
• Richer templates (opt-in) — an ObjectSearch can act as the master list of a master/detail view, a DocumentSearch can follow the selection on its own, and its results can open a document preview over the view with the keyword panel beside it. ObjectSearch cards and lists can also show the values of the search’s display columns, including values reached through a relation. See data components.
• Ask AI about a template’s documents (opt-in) — a template can carry a chat panel that answers questions about the documents its document search matches, citing the documents and pages it draws on. The documents are fixed when the conversation starts, and the chat never changes documents, keywords or the case.
• The case as it is after a task (opt-in) — completing a workflow task on a Caseflow object now refreshes the values on the page, so you see what the task changed. If you have unsaved edits, they are kept and you are asked to save and reload.
• Classes that extend another class — such a class now shows its own attributes together with the inherited ones, a new or removed attribute is carried through the hierarchy straight away, and deleting the class leaves the classes beneath it working.
• Change tracking on new attributes — attributes created in Caseflow administration or through configuration promotion are now created with change tracking switched on, so their history is recorded from the start. Existing attributes keep their current setting; contact your consultant team if they need change tracking enabled.
• Smaller fixes — creating an object whose template sets a number no longer fails, and 0 is kept as 0; a slash or greater-than sign inside a template tag value, such as “Flow (m³/s)”, no longer breaks the template; and attribute values in templates are always shown as plain text, while templates using the retired HTML field setting keep working until you update them.

​

⏳ Workflows

• Tasks where the work is (opt-in) — a workflow author can switch on Also show on the document or object for a task, so it can be completed from the document page or the Caseflow object as well as from the Tasks page. The document page shows a Tasks button with a count, and Caseflow templates can place the new WorkflowTasks element. Who may see and complete the task is unchanged, and it applies to instances started after the version is published. See tasks.
• Caseflow deletions reach the right workflow — a workflow anchored to a Caseflow class now recognises its objects by the class the event carries, so renaming a class no longer makes its deletions go unnoticed and two applications with the same class name no longer start each other’s workflows. Trigger filters can read the class id and application id. See triggers.
• Safer email steps — every placeholder in To, Cc and Bcc must resolve to a single address, values inserted into an HTML message are encoded, and a value that looks like a template reference is shown as written. An email step can send through the shared email relay that comes with Nobly Insight, or through the servers your environment allows, and a refused server is named in the failure message.
• Designer fixes — the attributes on a Create object action now come from the class named in its Class ID, and the Document.Reindex action always shows its Target Document Type, so you can pick it in the form instead of editing the action’s JSON. See actions.

​

🧾 Scripts

• Office documents as PDF, without storing a copy — a script can now get a document revision as a PDF with get_document_pdf, which converts a Word, Excel or PowerPoint file on the fly, or convert Office content it already holds with convert_office_to_pdf. Nothing is stored, so it also works for document types that do not allow a PDF rendition, and older formats such as .doc are included.

​

📤 Upload

• Edit pages before you archive — a PDF or TIFF in the upload list now offers Edit pages: turn a page, leave one out, or split the file into several documents. Nothing changes until you apply, editing a signed PDF asks first because it removes the signature, and files from the pending uploads queue can be edited the same way. See uploading documents.
• Pick rows like a file list — Ctrl-click (Cmd-click) adds a row and Shift-click a range, without clearing shared keyword values you have typed. The shared keywords panel lists the keywords the selected files need under Required first, and double-clicking a file opens a larger preview.
• Always store chosen document types as a new document (opt-in) — under Admin settings → Upload → Upload behavior, administrators can list document types whose uploads are always stored as a new document, without asking whether to add the file as a revision. The setting applies to everyone. See defaults and actions.
• Large uploads on multi-server installations — files uploaded in several parts no longer fail now and then with “Refusing to commit an incomplete upload”.

​

📄 Documents & viewer

• Manage the file formats you upload — a new Admin settings → Documents → File types screen lists the file formats documents can be uploaded in. Add a format your uploads refuse today, such as a CAD drawing or a video, by giving it a name, an extension and its MIME type. A document type’s default file format is picked from the same list. See document types.
• OCR on large scans, with progress — OCR in the redaction viewer now runs in the background, so scans of up to 2,000 pages or 90 MB complete instead of being cut off. You see which step it is on and how many pages are done, can cancel it, and are told which pages could not be read. OCR now also works on JPG, PNG and multi-page TIFF images. See redacting a document.
• Redaction lands on the exact word — search highlights, marks and drag selections in the redaction viewer now cover exactly the characters you chose, also in justified text and converted scans.
• Large documents save after redaction and annotation — saving needs far less server memory, so large scans and long PDFs save again. If a document is too large to save at that moment, nothing is stored, your marks stay in the viewer, and you are asked to try again shortly.
• The viewer remembers your settings — your tool, zoom, side panels, search options and sort order come back as you left them, saved for each user in that browser.
• Unsaved work is protected — the browser’s Back button and links inside Nobly Insight now ask before leaving keyword edits, redaction marks, annotations, the workflow designer or unsaved administration settings.
• Safe replacement of a revision’s file — a replacement now switches the revision over in one step and leaves the original in place if the replacement is refused or fails.
• Readable mask warnings — when a value does not fit a keyword’s input mask, the message shows the mask with its fixed characters, such as DK-999999999.

​

🧰 Office

• Desktop saves are stored — saving from Word, Excel or PowerPoint after Open in app now always creates a new revision. Some saves were previously confirmed without being stored.
• Large documents up to 1 GB — saving back a document larger than about 28 MB no longer fails, and the document is not held locked while it transfers.
• Read-only opens release the document — a document opened read-only no longer holds the editing lock, so colleagues allowed to edit are not kept waiting.
• Documents open as complete — Office is now told the exact size of the file it downloads, so a document no longer looks incomplete or damaged in the application.

​

🔎 Search & folders

• Clear search in the field — clearing a search is now an X inside the search box, and the cursor goes back to the field so you can type your next query straight away.
• Your view stays when you delete — deleting a document or Caseflow object from the results no longer re-runs the search, so sorting, column filters, Find in results and your scroll position are kept.
• Group access for many system searches at once — on the system-search administration page, pick one or more user groups and tick the searches and folders they should reach, with a preview of what gains and loses access before you save. Every changed search and folder is recorded in the audit log. See system searches.
• External datasets you can search — keywords whose values come from an external dataset now look values up as you type, say when more matches exist, and keep what you typed if the source cannot be reached. Lookups return complete, distinct values, characters such as % and _ in a search term are matched literally, and Test Query previews exactly what a dropdown shows. See external datasets.
• External datasets from Script Engine scripts (opt-in) — an external dataset can now take its values from a Script Engine script instead of a SQL query, with a new External Dataset script type and a starter template.

​

🤖 AI

• Stop an answer (opt-in) — a Stop button replaces Send while AI Chat is writing. What was already written is kept and marked Stopped.
• See how full a conversation is (opt-in) — a ring beside the send button fills as the conversation uses the assistant’s context window and warns before it is full. It also appears in Caseflow document chat.
• Choose where each agent is offered (opt-in) — under Admin settings → AI Assistant → Agents, a search agent’s Offered in setting picks AI chat, Caseflow templates or both. Agents you configured before stay offered in AI chat. See agents and settings.
• Classification trees that stay valid — saving a tree now checks that every allowed keyword is a keyword type of the branch’s document type; configuration promotion holds back or blocks a tree, or a change, that would leave a branch the target environment refuses; and a document type cannot be deleted while a branch still stamps it. See configuring the tree.

​

📊 Dashboards & reporting

• Results cut by a row cap are flagged — when a data source returns more rows than its row cap, the cards, charts and tables built on it say that only the first rows are included, exports warn that the workbook is incomplete, and dashboards with parameters suggest narrowing them. See viewing dashboards.

​

🛠 Administration & access

• Saving no longer overwrites a colleague’s change — every admin settings page now saves against the version of the item it loaded. If someone else saved the same item in the meantime, your save is refused and the page asks you to reload instead of silently replacing their change, and what you typed is kept.
• You cannot lock yourself out — you can no longer disable your own account or remove it from a user group, on the Users screen, through the API or through the admin assistant; another administrator has to make that change. See users and groups.
• Keep user accounts in step with Microsoft Entra ID (opt-in) — where you sign in through Microsoft Entra ID, accounts can be created, updated and deactivated as people join, change and leave. Leavers are deactivated rather than deleted, at most 25 accounts can be deactivated in an hour, and administrator, service and integration accounts are never touched. Ask your Nobly contact to set it up.
• Filter users by status — Admin settings → Access → Users can now list disabled accounts on their own, with a Reset filters button and steadier paging. See users and groups.
• Document Composer permissions (opt-in) — the Permissions screen has a Document Composer area for granting each user group what it needs. See application permissions.
• Mass value editing that finishes — a value replacement with auto-name recompute now renames only the documents it updated and reaches Completed when the replacement is done. See mass value editing.
• Document type revision settings — rendition-only document types are no longer shown as revisionable, and editing a document type keeps its revision and rendition settings intact.
• Retention policies promote correctly (opt-in) — a promoted policy that counts from a date keyword now uses the keyword of the same name in the target environment, and the editor keeps the keyword the policy really evaluates. See retention policies.

​

🔒 Security

• Safer previews — HTML files and mail messages now open in an isolated preview where nothing can run scripts or load content from the internet, SVG images are shown as plain pictures, and HTML stored in a Caseflow attribute stays within its field. When something fails, you see plain wording and a request ID to quote to support.
• Hardened services — the Nobly Insight API, the identity service, the Script Engine and ContentFlow send the standard browser security protections, run current versions of their web servers without advertising them, and keep internal details out of health checks, error messages and refusals.

​

🔌 For integrations

• Configuration changes through the API — configuration writes now share one path with the admin pages and the admin assistant. Reads return a version you can send back so an update cannot overwrite a newer change (409 when it is out of date), every refusal uses one format with an errors list, and a write that meets another configuration change being saved answers 503 with Retry-After. Five requests changed shape and refuse the old one: a user group’s feature access, a system search’s user groups, publishing and rolling back a script, updating a script, and a new draft of an existing workflow. The editor AI endpoints are removed. Check integrations and scripts that change configuration against adapting an integration built on an earlier release.
• Update many Caseflow objects in one call — POST /api/v3/caseflow/{application}/{class}/bulk-update sets the same attribute values on up to 5,000 objects of a class, with the same permission and attribute security checks as updating them one at a time. Objects that already hold the values are left untouched, and the response counts the objects updated, unchanged and skipped. Long text, formatted text, relation, document, encrypted and calculated attributes cannot be set this way.
• AI extraction names files it cannot read — POST /api/v3/ai/llm/extract now answers a password-protected PDF with the problem type urn:nobly:insight:problem:password-protected-file, and a file with no readable text, such as a blurred photo, with urn:nobly:insight:problem:no-text-found, both as 400; a file without text used to get a 500. Retrying does not help with either: ask for an unprotected file or a sharper picture.
• File types by MIME type — the v3 API now identifies file types by their MIME type everywhere and no longer returns file type numbers. If an integration reads a file type name or number from an upload’s status or a document’s renditions, update it to use the MIME type. See integration contracts.
• Input errors, not server errors — invalid ids, values that do not match a Caseflow attribute’s type, and unknown keyword types are now answered as invalid input or not found, and nothing is written.
• External dataset value limit — a lookup can ask for between 1 and 1,000 values, 1,000 by default.
• Refused tokens — the Script Engine now answers every refused token with the same short message. If an integration reads the wording of a refusal rather than the status code, review it.

​

⚡ Reliability & platform

• Two tabs no longer sign each other out — a Caseflow object, document, dashboard link or redaction view opened in a new tab now signs in on its own, instead of sharing a sign-in that ended the session in both tabs.
• Faster account lookups — sign-in, account lockout checks and the permission check behind every document you open now find the account directly, which keeps them quick and the database calm on large user directories. Account names are matched exactly as your source system matches them.
• Recent searches recorded reliably — the same search run in two places at once is counted once in your search history instead of one run being dropped.

​

📥 ContentFlow

• A new workflow canvas — steps are cards with an icon for their type, you can pan, zoom and use a minimap on large flows, and the canvas opens on the flow. Arrange lays a tangled flow out left to right or top to bottom in one click, with exception steps in a lane beside the main flow, and one Undo puts it back. See design and publishing.
• Find runs by document ID — the States page search now takes a Nobly Insight document ID as well as a trace ID and finds every run that created, revised, updated the keywords of or deleted that document, failed attempts included. See states and recovery.
• A mapping row can add its value — a field row in a document mapping can now add its value to a keyword an earlier row or rule already wrote, so a keyword can carry every value that applies. Tick Add its value when an earlier row already wrote this field on the row. See document mappings.
• Plain-text email by default — the Send email step now sends its body as plain text unless you turn on the new IsHtml option. Workflows that send an HTML body should switch it on and wrap values from the item in the new HtmlEncode(...) helper. See steps and variables and tools.
• Clearer failure messages — a failed run’s message names what failed and gives the trace ID; the full error is in the run’s log.
• Browser access limited to your ContentFlow pages — the ContentFlow APIs now answer browser requests only from your environment’s ContentFlow screens. Integrations that authenticate with a token are unaffected.
• Assistant sign-in — an AI assistant connecting to ContentFlow is now sent to the sign-in realm that holds its client, where staff and customer identities are kept apart. See authoring with an AI assistant.
• Clearer wording — ContentFlow screens use the ContentFlow name throughout, examples no longer carry customer-specific names, and a step without an on-error target says what will actually happen.

​

Nobly Insight — early-October update

​

💬 Interaction rules: ask before a document is stored

• A question at upload, an answer for your workflow (opt-in) — administrators can set up interaction rules under Admin settings → Workflow. A rule picks the uploads it applies to by document type, keyword values and document date, with the same criteria editor as retention policies; it puts a question to the user with up to six answer buttons, and starts the document workflow you choose with the answer passed into its variables. See interaction rules.
• Answered before anything is archived (opt-in) — the question appears once any duplicate choice is settled and before the file is stored; adding a revision never asks. Backing out abandons that upload so you can adjust it and try again, and if a rule changes while you are answering, you are asked again with the current choices. Document types without rules are not slowed down.
• Managed like the rest of your configuration — rules can be reordered, disabled, deleted and restored, two administrators cannot overwrite each other’s edits, and every change is recorded in the configuration audit log. New view, manage and administration permissions decide who may change them, granted automatically to the standard manager groups.

​

📤 Upload

• A queue for scanned and integration files (opt-in) — scanners and other sources can now hand files to Nobly Insight, where they wait on a new Pending uploads tab in the upload tool until someone indexes and archives them or removes them. Start upload loads an item’s files into your upload list and marks the item as yours; colleagues see who has it and can take it over. Files are stored encrypted while they wait, items never expire on their own, and every item that leaves the queue is recorded in the configuration audit log. Two new permissions govern it: Prepare Pending Uploads for the account a source signs in as, and Handle Pending Uploads for the people working the queue. See the pending uploads queue.
• Type the document date — the document date in the upload list can now be typed in your environment’s date format, with the calendar still beside it. A date that cannot be read is flagged in place and that row will not upload until it is corrected, and a date picked close to midnight is now stored as the day you chose, in every time zone. See uploading documents.
• Files filed by what they are — when the browser cannot tell a file’s format, the upload is now filed by its extension, falling back to the file types your organisation has configured, instead of being stored as plain text. .json files are accepted as text documents.
• Uploads that complete — uploads and new revisions no longer fail part-way with a message that the file’s details are not available.
• AI Index on a full tray (opt-in) — AI Index now analyses a few files at a time instead of all of them at once, waits and retries when the service is busy or your AI request limit is reached, and says on the row why an analysis could not run. A file stays locked while its analysis is pending, so it cannot be uploaded or have shared keyword values applied over the result in the meantime.

​

⏳ Workflows

• Scripts that run for up to 30 minutes — a Script.Run action can now run in the background instead of being cut off after 10 seconds, so a workflow can split or merge PDFs, sweep a set of documents or run other slow logic. The node waits on a Join transition and carries on with the script’s output once it finishes. A run that fails, times out, is cancelled or is interrupted by a restart takes the node’s error path and is never repeated, and a set-up mistake is caught when you publish. See running a script in the background.
• Lists and records reach scripts as data — a list or record passed in a Script.Run input now arrives in the script as a list or dictionary, and numbers and true/false values keep their type. If one of your scripts expects the JSON text instead, review it.
• Scripts no longer set off workflows — what a workflow’s own script writes to Nobly Insight no longer starts workflows, and documents it creates are no longer added to lifecycles, so a script that updates the document it is working on can no longer trigger itself again. Its calls to other systems are unchanged.
• Triggers that keep to their scope — a workflow anchored to a Caseflow class now starts only for objects of that class, including for deletions, and a document, Caseflow or mail trigger that carries a schedule is refused when you publish instead of the schedule being silently ignored.
• Filters built from the event itself — when you write a trigger filter you are now offered the values that trigger’s own event carries, such as the details of an imported mail, and a filter on a delete trigger that reads the deleted item is flagged as always empty. The workflow authoring assistant (opt-in) follows the same rules, so the filters it drafts match real events. See trigger filters.
• Group tasks regardless of capitalisation — a task aimed at a user group now appears on the Tasks page for its members however the group name is capitalised. Existing tasks are corrected for you.
• Duplicate protection now also covers an interrupted start — if the engine stops between creating an instance and running its first step, the repeated event now completes that start instead of leaving an instance that never moves.

​

🧾 Scripts

• Script variables — keep non-sensitive settings such as base addresses, identifiers and environment names under Admin settings → Script Engine → Variables, for every script or for one, and read them in a script the same way as secrets. Administrators can see the values, so you can check what a script receives; keep credentials in secrets. Variables travel with configuration promotion (anything that looks like a credential is never exported), and the script authoring assistant uses them in the code it suggests. See secrets and troubleshooting.
• Work with PDFs — scripts can open a PDF, extract its text, split it and merge pages, check a signature against a trusted issuer certificate kept as a secret, and read the PDF rendition of any document revision. A script can also turn HTML or Markdown it composes into a PDF, with page size, margins, headers and footers, for a letter, a receipt or a printout of a case. See script examples.
• Bundles list what they leave out — a document that cannot be converted no longer fails a whole bundle, whether a script or an integration builds it: it is listed under a “not included” heading at the end of the table of contents, and the result names each one and why. A bundle with nothing to merge is still refused.
• Maintain autofill keyword sets from scripts — a script can list your autofill keyword sets and add, change or remove their rows, to keep a lookup table in step with a system you trust. The account the script runs as needs the new Keyword Sets privilege. See autofill keyword sets.
• Long runs for your integrations — an integration can now queue a script through the API, get a reference back straight away, follow or cancel the run and read its result. A run may last up to 30 minutes, and sending the same request again never starts a second one. See integration contracts.
• Errors you can catch by name — a failed call to Nobly Insight from a script now raises ApiError with the status, method and address, so an expected refusal can be told apart from a real failure.
• Services behind private endpoints (opt-in) — scripts can now call a service published through a private endpoint in your network, once that destination is allowed for your environment.
• Event hooks and service-account runs reserved for the platform — an event-hook script now runs only from its document events, and a run that acts with the Script Engine’s service account can be started only by the platform itself or by someone with Script Engine management permission. If a custom integration of yours starts such runs under another identity, give it that permission. See bindings and event hooks.

​

📄 Documents & viewer

• Video and audio in the viewer — recordings start playing as soon as the first part arrives and you can jump to any point; QuickTime .mov and .m4v files play too, and anything the browser cannot play falls back to a download. Your access is re-checked while a recording plays, and playing it is recorded once in the document’s history. See working with documents.
• More files preview properly — AutoCAD drawings open on the download card instead of a broken image, large CSV exports preview their first 1,000 rows with a download for the rest, and files separated with semicolons, tabs or pipes split into columns. Excel files converted for viewing, download or redaction now show dates and times in your locale.
• Filled-in PDF forms — values typed into a PDF form now show in the viewer and in redaction, are found by Search document, and redacting removes only the values you mark instead of the whole form.
• Form data on legacy form documents (opt-in) — values a legacy electronic form stored outside its keywords now appear beside the document’s keywords, and where you may change the document, an update template can let you edit them with the new FormDataField and FormDataValue components. A change made elsewhere at the same time is never overwritten. See form components.
• Document Information on small screens — the details cards now stay inside the dialog in narrow windows.

​

🏷️ Keywords

• Check-digit validation — a text keyword type can now require a valid Norwegian national identity number (fødselsnummer, including D-numbers), a Norwegian organisation number, or either. A value that fails is flagged under the field and refused — at upload, in the keyword panel, in bulk edit and on form documents — while empty values and values already stored are left alone. A default value for the keyword has to pass as well, the validator travels with configuration promotion, and Caseflow template fields can use the same national identity check. See check-digit validation.
• Older values no longer block a save — now that input masks are enforced on save, a value stored before its mask was added or tightened no longer stops you saving a document’s other keywords. Only the values you change or add are checked, surrounding spaces are ignored, and a refused save tells you why.
• One keyword-edit window per document — you can no longer start editing a document’s keywords in a second window, tab or device while you are editing them elsewhere, and warnings now say when pending changes are your own. The keyword panel’s heading, its edit controls and the document actions stay in view while the list scrolls. See updating keywords.
• Autofill sets no longer hide keywords — a keyword assigned to a document type and also filled by an autofill keyword set bound to that type now stays on the type, so uploads of such types are no longer refused with “keyword(s) do not exist on document type”.
• Duplicate checks stay quick — a duplicate check on a case-sensitive keyword, such as an organisation name, no longer slows down storing a document.

​

🧰 Office

• Office editing out of the box — Open in app, Open in Office Online and the Office add-in no longer need an administrator to switch editing on. The usual document type and user group rules still decide who may edit, and environments that switched editing off stay off.
• Read-only where there is nothing to save — a document whose type keeps no revisions now opens read-only: in Office Online without an Edit button, and in the desktop app after a confirmation. The add-in disables Save Revision for it and explains why. To keep your changes, use Upload New Document in the add-in, or Save As and upload the copy. See document types.
• Names that used to stop editing — documents with Nordic or accented characters, or very long names, now open for editing, and reopening a document right after saving it from Office gives you the revision you just saved. Accounts without an email address can now edit in Office Online too.
• Save Revision tells you what happened — the add-in now confirms that a save really produced a new revision, keeps warnings that need you to act on screen, uploads your edits as a new revision when a save cannot be written back, and shows its messages in your language.
• Files keep their real format — a document archived from the add-in, including a macro-enabled workbook or one created from a template, is stored in its real format, so it opens again in Office.
• Signing out means signed out — signing out of the add-in also ends your session with your identity provider, so the next person on that device has to sign in.
• Staged documents protected — documents the add-in hands over to the upload tool are encrypted while they wait and cleared away once they are no longer needed.

​

🔎 Search & folders

• Bulk delete from search results — select rows and use Delete (n). The count includes only documents you may delete, deletion runs in small batches you can stop, and the summary names any document that could not be deleted. See searching for documents.
• Clear search — a new button beside the search bar empties results, preview, selection and filters in one go, and asks first if you have unsaved keyword edits.
• Fixed filters out of the way — in the input dialog of a search that asks for values, the fixed criteria now sit in a collapsible Fixed filters section below the fields you fill in, and a field that is optional because of a fixed criterion says so. See system searches.
• Folder views with masked keywords — folders built on a masked keyword are named and found the way the value is displayed, however you type it. A view with more than 50 folders asks you to search instead of listing thousands, and the folder tree now has the same grey panel as Search, with a clearer highlight on the selected folder. See folder views.
• Search prompts open straight away — a search covering document types whose keywords you may not read no longer holds its prompt back for several seconds.
• Search auditing under load (opt-in) — with the tamper-evident search audit switched on, two searches run at the same time no longer risk one of them failing while its record is written.

​

🤖 AI

• A guided editor for extraction profiles — each profile is now edited on its own screen, with validators picked from a list instead of written as JSON, and every save becomes a version you can compare and restore; two people can no longer overwrite each other. Test a draft against a sample document before you save it to see the extracted values, their confidence and the evidence behind each one, and follow every extraction a profile has run in its run log. See extraction profiles.
• Clear answers when a document cannot be analysed — analysis, extraction, summaries, chat and redaction now say when a file is damaged, password-protected, too long or has no text to read, and the run log records File could not be read rather than an extraction failure. A request that runs too long gets a timeout message with what to try, a busy service is told apart from a slow one, and retrying a slow document reuses the text already read.
• Select and export AI search results (opt-in) — AI search results can now be selected, downloaded as a zip, exported and bulk edited like any other results, with each document counted once. See AI search.

​

🎨 Caseflow

• No more overwriting each other’s templates — if someone saves a Caseflow or form template after you opened it, your save or restore is refused instead of replacing their work: Reload to see their version, Overwrite to put only your changes on top, or keep editing. Leaving the editor with unsaved changes now asks first, and reopening a Caseflow template in the same browser tab offers your unsaved work back. See when someone else saved first.
• See where an attribute is used (opt-in) — a new page under Admin settings → Caseflow shows which saved searches use each attribute and how — as a filter, column, sort or prompt, including through a relation — and deleting an attribute warns you which shared searches will stop working. A saved-search overview flags references that no longer resolve. See applications and classes.
• The templates screen remembers your place — it opens with the first application in your priority order selected, and keeps your application and class when you come back from the editor.
• Bulk edit wording — the attribute page is now Bulk edit attributes with an Editable in bulk edit column, Caseflow screens speak of saved searches rather than filters, and the legacy import is named Re-import legacy filter bars, with a plain warning about what it replaces.
• Changes reach only the object you mean — an update or delete addressed through a class now reaches only objects of that class or a class extending it, rights are checked on the object’s own class, and locked accounts are refused here as everywhere else.

​

📬 Mailbox importer

• An import history page (opt-in) — Admin settings → Mailbox importer → Import history lists what each mailbox has processed, filters to failures and retries them, and shows how many attachment documents each message produced and which rule handled it. See operations.
• More precise rules (opt-in) — rules can now also match attachment file names, each condition takes up to 20 values, and document types are picked by name. A policy can store a message’s attachments alone, skipping messages that have none. See archive rules.

​

📊 Dashboards & reporting

• A filter bar for the whole dashboard — dashboard builders can add dropdowns bound to a data source’s columns, so one choice narrows every element on that source. The selection is kept in the page address for sharing, an element can opt out, and the date range control can be hidden. See building dashboards.
• Document types as parameters — a data source parameter can now be a document type group or a document type, picked from what the viewer may see, with several selections allowed and the types narrowed by a chosen group. See data sources.
• Report on ContentFlow ingestion (opt-in) — a new Database picker lets a data source run against your ContentFlow database, so a dashboard can chart runs, batches, failures and retries next to the documents they produced.
• Value cards keep their settings — switching a value card’s display keeps its title, label and measure, and the multi-select pickers in administration are now translated.

​

🛠 Administration & access

• Choose your start page — Personal settings → General settings → Start page, or Use this page as start page in the user menu, opens Nobly Insight on the section, saved search, folder view, dashboard or case you use every day; Home stays the default. If you have not chosen a language yourself, the organisation’s language now keeps applying to you after you save other settings.
• The Keyword Sets privilege — Admin settings → Documents → Access rights (by group) can now grant a group the right to create, change and delete autofill keyword set entries, including from scripts and workflows. It needs Manage user groups and leaves the group’s document type rights untouched. See document access rights.
• Single sign-on only — where your organisation’s single sign-on is set up, the local username and password form is now switched off, so every interactive sign-in goes through your identity provider and its multi-factor and conditional access policies. Environments without single sign-on keep signing in as before, and if your identity provider is ever unavailable, your Nobly contact can restore access. See single sign-on.
• Single sign-on for your own applications (opt-in) — applications of your own, outside Nobly Insight and ContentFlow, can sign users in through the same Microsoft Entra ID federation, with access following your directory groups at every sign-in. Your Nobly contact arranges it.

​

📥 ContentFlow

• Document mappings — a new screen defines in one place what Nobly Insight receives: the document type, the document date and each keyword, with where its value comes from, any transformation, and what happens when it is empty. The document type rules introduced in the previous release now live inside the mapping, so the separate menu and the Derive document type step are no longer offered; workflows already using the step keep running. Preview any edit against a sample document — tests run the draft while live documents use the published version — and classic mappers keep working, opening as mappings. See document mappings.
• Checked against Nobly Insight before a run — Check against Nobly Insight on a mapping’s Try it tab shows whether the document type has every keyword the mapping writes and flags required keywords nothing fills. ContentFlow now places each value where the document type keeps it, in a keyword group’s record or as a plain keyword, so one mapping row can serve several document types.
• Several alternatives in one rule condition — a contains condition can now hold several values and matches when any of them is found. Lookup tables a mapping reads from can no longer be reshaped, renamed or deleted while a mapping depends on them.
• Keep the original metadata — a new Upload JSON metadata to Nobly Insight step stores what the run received as a readable JSON document next to the archived document, with its own document type, date and keywords from a mapping.
• A full rerun picks up your fix — rerunning from the first step now uses the workflow version published at that moment, while continuing from the failed step keeps the version the run started on. See states and recovery.
• Run data cleaned up for you — runs nothing has touched for 12 months are removed with their logs, an unfinished run’s file after 30 days, and run logs and permission audit entries after 60 days; each period can be changed for your environment. A rerun that needs a file already removed says so.
• Clearer refusals — a submission that names a workflow that does not exist is refused at once, a run whose workflow was deleted or renamed is marked Failed with the reason, and a file whose type cannot be determined comes with guidance on giving it a name or MIME type.
• AI assistants in ContentFlow — an AI assistant connected to ContentFlow can now read document mappings and, where its user has the rights, create, check, preview and publish them; look up document types and their keywords; follow runs and batches; and create and run guided tests. When it drafts a workflow it is told about unknown settings, expressions that do not compile, secrets written into a step and references to mappings or variables that do not exist. It can never do more than its user may in ContentFlow itself.
• Running test code needs the rights that publish it — running an expression in the tester, or a mapping preview that evaluates expressions, now requires Manage workflows or Manage mappers, the rights that already put code into live runs. Anyone else can still check that an expression compiles.
• Smaller fixes — a changed upload limit applies from the next run without a restart; a bulk rerun tries every run you selected and reports which started and which did not; the States list pages through runs without repeating or skipping any; the step editor no longer flags valid code or blocks Apply for a value converted when the workflow runs; and deleting a classic mapper that is still in use is refused, naming the workflows that use it.

​

⚡ Reliability & platform

• The Workflow Engine shares the database fairly — running steps no longer compete for database connections, idle queues are polled less often, and a workflow start that had to wait resumes as soon as capacity frees up.

​

Nobly Insight — mid-September update

​

📬 Mailbox importer: mail archived for you

• Connect your Microsoft 365 mailboxes (opt-in) — administrators can connect one or more mailboxes from the new Mailbox importer page and schedule imports, so incoming mail is stored as documents in Nobly Insight without anyone forwarding or dragging it in. See setting up a mailbox.
• Rules that decide what is stored (opt-in) — ordered rules route messages by sender, recipient or subject to different document types and keyword mappings, with a default for everything else. Each message is stored either as the original mail file, or as an HTML copy of the body with every attachment as its own document. See archive rules.
• Zip attachments unpacked (opt-in) — a policy that archives attachments separately can expand zip files so each file inside becomes its own document, limited to the file types you name if you wish, and keeping the zip itself if you want it. An archive that cannot be opened is stored as an ordinary attachment with a warning rather than failing the import.
• Mail can start a workflow — a new Mail imported trigger starts a document workflow when a message arrives, with sender, recipients, subject, timestamps, the rule that matched and the archived attachments available to the workflow’s filters and later steps. See mail and workflows.
• Imports you can follow — the import history shows what was archived and what failed, and an administrator can review and retry a failed message. See operations.

​

📝 Form documents: records that have no file

• Capture records without a file (opt-in) — inspections, registrations and requests can now be stored as form documents: keyword-only documents that are searched, governed and secured exactly like any other document, minus the actions that need a file. Users create them from New form in the sidebar. See form documents.
• Design the form per document type (opt-in) — a form template designer builds the preview, create and update forms for each document type with the same editor as Caseflow templates: a component palette, live preview, version history and inline checks. Where AI template generation is enabled, describe the form you want and have it drafted for you. See form templates.
• Who may design forms — new view, manage and administration permissions govern the designer, granted automatically to the standard manager groups where those exist.

​

🚚 Move configuration between environments

• Promote configuration instead of rebuilding it (opt-in) — export an environment’s configuration to a file and apply it to another: document types and keywords, Caseflow and form templates, scripts, workflows, navigation and banners, reporting and retention policies. You only see and change the configuration types you hold the matching administration permissions for.
• Preview before anything is written — every difference is listed as a create, update, delete or unchanged item with a before-and-after view, anything that cannot be applied is marked with the reason, and you choose what to include. Each group of changes applies in full or not at all, and stored secret values never leave their environment.
• Keep the differences you mean to keep — mark configuration this environment deliberately holds differently and it stops arriving as a pending change, and point user-group names from the source at the groups that exist here — groups themselves are never copied between environments.
• See where you stand — the status view shows the configuration version this environment is on, what has drifted since it was applied, and a history of earlier promotions. See configuration promotion.

​

🤖 AI assistants: Caseflow, keyword edits and authoring help

• Assistants reach Caseflow (opt-in) — a connected assistant can now list the Caseflow applications, classes and attributes you can reach, open a case object and search case objects by attribute, including attributes reached through a relation. Where you allow it, the assistant can also create and update case objects; deleting is never offered.
• Assistants can update keywords (opt-in) — an assistant can set, add to and clear keyword values on a document, including keyword groups, and says plainly whether anything actually changed. Keywords that are read-only or hidden for you are left untouched.
• Assistants can reindex a document (opt-in) — an assistant can move a document to another document type and re-map its keywords. It names the values the new type has no place for, asks you to confirm when a matching document already exists, and the change is recorded in the document’s history like any reindex you perform yourself.
• Four new permissions, granted per group — MCP: View Caseflow, MCP: Edit Caseflow, MCP: Edit Keywords and MCP: Reindex Documents join the existing MCP Access permissions on the Permissions screen. MCP Administration now covers them too, so review which groups hold it. An assistant is offered nothing its group has no permission for, it always acts as the person using it and never exceeds what they may see or change, and its searches are recorded in the audit trail. See granting access.
• Draft workflows and scripts by describing them (opt-in) — the workflow designer and the script editor gain an assistant panel: describe the change you want and it proposes a complete definition or script, checked before you see it and shown beside your own work. Nothing is saved until you save it, and a per-administrator setting decides what the assistant may do on its own and what it must put to you first. See the workflow designer.

​

🧠 AI Search: choose what gets indexed

• Automatic indexing, by document type (opt-in) — a new page under Admin settings decides which new documents join the AI search index as they are stored: none, every document type, or the types you pick. Previously this was an all-or-nothing choice made when the environment was set up. A change applies from the next document with no restart, and the page tells you if anything it needs is missing. See index administration.
• What is indexed stays current — whatever the automatic setting is, documents already in the index keep receiving their content, keyword and redaction updates, so nothing outdated is left searchable.
• Large documents stay findable — a document too large for its contents to be indexed is now indexed by its metadata and keywords instead of being retried until it is dropped, the size limit for indexing contents has been raised, and content from an earlier revision is cleared when a new one is too large.
• A tidier index — a document deleted before its indexing event was handled is removed from the index, and an indexing failure is reported once with the reason instead of being retried over and over.

​

⏳ Workflows

• The task you opened is the task you complete — if a workflow re-arms a task while you have its form open, completing it now acts on the round you opened rather than a newer one — in the client, in the Office add-in and through the API — and you are told the task has moved on instead of being told someone else completed it. See tasks.
• Ask the person who started it — a task can now be aimed at the user whose action started the instance, alone or alongside the usual group audiences: the natural way to ask whoever uploaded a document to confirm something. Ownership carries into the workflows that one starts, including every item in a batch, and the setting is now reliably kept when a definition is saved or published.
• See why a run stopped — a failed run shows the step it stopped at, the failing action and the recorded reason, and the run journal records which user or service account performed each step. For a run that executed a script you can open its log directly from the history, and each definition has an Instances action that opens its own runs. See monitoring.
• Mistakes caught before you publish — a trigger whose event type is missing, misspelt or unrelated to the workflow’s anchor, and a filter that reads something the event cannot provide, are now rejected when you publish instead of publishing quietly and never starting. Caseflow trigger filters can also read the object’s attributes, including an attribute’s previous value, so a workflow can start only when a value changes to something specific.
• Repeated events no longer start duplicate runs — a document or Caseflow event delivered more than once is recognised as a repeat, so it can no longer start a second instance. This applies to your existing triggers, with nothing to publish or switch on.
• Caseflow wording throughout — the designer, its validation messages and the action catalogue now say Caseflow everywhere, and your saved drafts, published versions and running batches are converted for you without any change in behaviour. If you have an integration that reads or writes workflow definitions, check it against the new naming: the retired names are no longer accepted.

​

🧾 Scripts

• Delete a script you no longer need — administrators can now delete a script together with its versions and its stored secrets. If a binding, Caseflow template or workflow still references it, the deletion is blocked and you are shown exactly what to remove first; workflows already running or scheduled keep the version they were built with. See authoring and publishing.
• More caught before you publish — validating a version now also reports imports that are not allowed, names that do not exist and calls a script cannot make, each with the line it is on, alongside the existing syntax and contract checks.
• Errors that say what happened — a failed script shows its own error message in the editor’s test run and in results returned to a caller, service errors report the call and its status without exposing credentials or content, and a result that does not match the script’s contract names the fields at fault.

​

🎨 Caseflow

• Link a case to a document — a new Document attribute links a Caseflow object to a single document. Templates can let users pick one from a search, open and preview it, and clear the link; a newly uploaded document can be written straight back to the attribute; and tables show the document’s name rather than an identifier. See applications and classes.
• Access rules follow relations (opt-in) — an attribute-security rule defined on a class now applies wherever that class is reached through a relation, not only when its records are opened directly. Filtered lists, saved searches and combined search leave out rows whose related record falls outside a group’s rules, so a record can no longer appear in a list and then be refused when opened.
• Related data respects your rights — a search, filter or sort that follows a relation into a class you may not view is refused instead of returning its values, including for searches an AI assistant runs, and reaching a class by name requires the rights on that class.

​

📄 Documents & upload

• PDF renditions on demand (opt-in) — a document revision can now be converted to a PDF stored alongside the original: images including multi-page scans, Word, Excel and PowerPoint files, and .eml or .msg emails. Bundling uses a stored rendition where there is one and converts those formats too, so a bundle can gather a mixed set of documents into a single PDF, and you can convert a doubtful document first to see whether it will work. Scripts can create renditions as well, and a conversion that is refused says why.
• Scanned pages open in the redaction viewer — pages in certain scanned image and compression formats could open blank, leaving nothing to review or mark; they now show their content. Documents that rely on common non-embedded fonts, or contain Chinese, Japanese or Korean text, also render properly.
• Masked values follow their mask — a value that must fill its mask now has to fill every position, including the fixed characters in the pattern, and bulk edit and reindex apply each keyword’s own mask settings instead of the strictest reading. Typing or pasting into a field with a maximum length now stops at a whole character. Trying a sample value against an input mask also shows the whole sample, instead of dropping its final character when the mask is mostly fixed characters.
• Related keywords shared as one set — when you share keywords across the documents in a multi-document upload, keywords belonging to the same autofill set are carried together, the confirmation shows which related fields will be cleared, and an incomplete set is caught before the files are sent. See autofill keyword sets.

​

🔎 Search

• Links that keep every filter — a link to a saved search now opens exactly the search it was copied from, however many filters it holds; a long search could previously open a narrower set of results. Searches handed over to AI chat keep all their filters too, and a link that cannot be read in full is refused rather than run with a different scope.
• Combined searches run with the values you configured — a Caseflow filter that takes its value from a document keyword now reliably uses that value and asks you for it when there is none, the fixed criteria appear beside your inputs so you see the whole query before running it, optional inputs are marked, and dates are shown in your local format.
• System keywords in every picker — keyword types tagged as system-use now also appear under their own System keywords heading in the search filter, sort and output-column pickers, listed last so fields filled by automation are easy to tell apart from the ones you maintain. Nothing is hidden or removed: a system keyword already among your columns stays exactly where it is. See keyword behaviour.
• Excel exports read the way the screen does — exporting results from a search or a Caseflow list to Excel now writes date and date-time columns in the same layout the grid shows them in, rather than a fixed technical format, while keeping them as real dates that still sort and filter. Languages using a non-Gregorian calendar keep the unambiguous standard layout, so no date can be misread.
• A tamper-evident search audit (opt-in) — where search execution auditing is switched on, each record is now linked to the one before it, so an entry edited or removed can be spotted when the trail is reviewed. Records keep the local time offset the search ran with, carry less detail that was never searched on, and a configured retention window now also covers searches recorded before this change. See history and auditing.

​

🛠 Administration & access

• Machine integrations without a claim from your identity provider (opt-in) — an API client that signs in with client credentials can now be mapped to the service account it should act as, instead of being refused unless your identity provider adds a user claim. Only the clients you list are mapped, a token that already identifies a person is never re-pointed, and a mapping can be pinned to one account. See API clients.
• Administrator groups created later get their permissions — a built-in administrator group created after your environment was first set up now receives the standard application permissions automatically, instead of being left without administration access. Anything an administrator has already granted or revoked is left exactly as it is. See application permissions.
• The audit trail records the right revision — viewing a document, adding or deleting a revision and deleting a rendition now record the document’s actual revision number, where entries could previously be one or two higher than the document really was. Existing audit records are unchanged.
• Cascading defaults accepted as you enter them — a default value on a keyword governed by a cascading dataset is now accepted in the form the cascade holds it, including a value typed with surrounding spaces or a masked value written with or without its fixed characters, and re-saving a configuration no longer fails over a default you did not touch. See cascading datasets.

​

📊 Dashboards

• Designing charts without losing them — changing a chart’s calculation from count to sum or average no longer replaces the element with an “invalid element” placeholder while you are still choosing the value column; a column that is missing when you save is reported on the field instead. A value card with no column chosen now says so rather than showing 0, and one whose column has been renamed or removed on its data source says that instead of quietly scoring zero. See building dashboards.

​

📥 ContentFlow

• Connect to partner systems that require a sign-in — the HTTP request and Webhook callback steps, and workflow scripts, can now sign in to a partner’s API for you: give them the partner’s sign-in address and the global variables holding the credentials, and the token is fetched, reused and refreshed automatically. Partners that issue no secret, and instead register a key or certificate with you, are supported as well.
• A States page that shows the whole picture — States now opens on run counts for completed, failed, running, scheduled and pending work, with separate Runs and Batches views, filters by time, workflow, batch and trace, and auto-refresh. Open a run beside the list to follow its status, error, properties, payload and logs while it is still processing — resize that panel, give the run the whole page, or open it in its own tab. See states and recovery.
• Rerun with a choice — when you rerun failed work you now choose whether it continues from the step that failed, keeping what it had already done, or starts again from the first step — for a single run, for whole batches, or for a pasted list of trace IDs.
• Run logs you can actually read — each entry is one line, errors and entries carrying an exception are opened for you, matches are highlighted, and you can expand or collapse everything at once, copy or download what you see, and open any entry at reading size with its full message and recorded fields, stepping through the run from there.
• Document type rules — a new Document type rules menu holds ordered rule tables that decide which document type a document receives, applied by a new Derive document type step. Several workflows can share one table, and a change reaches the next document processed without republishing anything.
• Prepare metadata without scripting — a new Modify property step sets a fixed value, copies from another property, translates values through a table you can paste in from Excel, removes a property or evaluates an expression. Try it on sample values before you publish, and save a translation table under a name to reuse it — and maintain it — in one place.
• Tests that check the result — the testing tab leads you through naming a sample, entering the incoming metadata, setting the expected outcome and running it against the saved draft. The result shows whether it passed, which property checks matched or differed, and the metadata before and after, so you can see exactly where a rule behaved differently from what you expected. See testing workflows.
• Edit a flow as JSON — the editor can show the open flow’s steps as JSON in the same shape as a downloaded workflow file. Paste in an array of steps, a whole file or a single step, check it, and apply it to the canvas as one undoable edit; nothing is saved until you save the draft. Settings the editor does not recognise are now kept as written instead of being dropped when you apply or save.

​

⚡ Reliability & platform

• Search and browsing keep getting faster — the notes indicator on search results is now checked once for the whole page instead of once per document, working out which document types you may see is quicker (which speeds up the lists and views that depend on it), and browsing a large result set across several document types with a date range is noticeably faster. A failed notes check also no longer sets off a burst of repeated requests.
• Keyword lists that keep working — if the cache serving cascading datasets and keyword metadata is briefly unavailable, the data is read straight from the database, so keyword dropdowns and entry forms carry on. Background database connections are managed more conservatively under heavy load as well.
• A cleaner API reference — the published API documentation now uses short, readable model names instead of long internal prefixes.

​

Nobly Insight — early-September update

​

🤖 AI assistants: connect an assistant to your own documents

• AI assistant access is here (opt-in) — you can connect an AI assistant such as Claude to Nobly Insight over the Model Context Protocol (MCP). The assistant signs in as you and sees exactly what you can see — never more. See AI assistant access for what is exposed and what is recorded.
• Capabilities you grant per group (opt-in) — new MCP Access permissions on the Permissions screen decide what each user group’s assistants may do: View Catalog (list document types and keyword types), Run Searches, View Documents (metadata, revisions and content), and View Keywords. An assistant whose group holds no MCP permission connects but is offered nothing. See granting access.
• Assistants search your documents (opt-in) — a connected assistant can find documents by type, keyword, date and name using the same search rules and document rights as the person using it; results are capped and clearly labelled. Assistant searches are always recorded in the search audit trail, marked as coming from an assistant.
• Assistants read documents and keywords (opt-in) — with the right permissions an assistant can open a document’s metadata, revisions and content, and read a document’s keyword values and the value list a keyword accepts. Reads are recorded in the document’s history under your account, and restricted keywords stay hidden.
• Manage the assistant’s sign-in yourself — with the Manage API clients permission you can view and edit, from Settings → Access → API clients, the callback addresses the assistant may return to after sign-in — including adding the hosted Claude connector’s addresses directly instead of raising a request. Addresses are validated as you save. See connecting an assistant.

​

⏳ Workflows: schedule runs and fan work out

• Run workflows on a schedule (opt-in) — a workflow with no document or case attached (a system workflow) can now run on a recurring schedule in your own time zone, using a simple preset or a custom cron expression. The new Schedules page under Workflow lists every scheduled workflow — pause or resume a schedule, start a run immediately, and review the run history. See building workflows.
• Fan work out across many items (opt-in) — a workflow can start one child workflow for each item from a list, a workflow-selector script, or a document or Caseflow search, and wait until all children have started or finished before continuing. Follow it from the parent instance: live counters as children start and finish, the child instances it spawned, and when a join step releases. See monitoring & administration.
• Start a workflow from a document action (opt-in) — custom document actions can now do more than open a URL: an action can start a workflow on the document it is used on. Administrators pick a published workflow when creating the action, and end users who click it don’t need workflow permissions of their own — the action’s own scope decides who can start it, and a clear message appears if the document is already running that workflow. See workflows.

​

🧰 Office add-in: tasks for the document you have open

• Workflow tasks in the task pane — the Office add-in now shows the workflow tasks waiting on the document you have open, so you can act on them without leaving Word, Excel or PowerPoint. One-click tasks finish with a single click; tasks with fields open a small validated form in the pane. If someone else has already completed a task, you’re told so and the list refreshes. See tasks.
• Safer archiving and clearer errors — a document reference the add-in cannot handle is now rejected before anything is uploaded, a refused revision cancels its temporary upload instead of leaving it to expire, and failed uploads and revisions now show the reason the server actually returned instead of a generic error. The announcement strip also stays responsive when the announcements service is slow.
• Keep working after you archive, always on — the flow that reconnects the document you just archived so you can keep working and save revisions straight away is no longer opt-in: it is now a permanent part of the product with no setup needed. See document management.
• Long editing sessions stay alive — a Word, Excel or PowerPoint document you have open in Nobly Insight no longer drops its connection after an hour: as long as you keep editing, the session keeps renewing, so long working sessions no longer end in failed saves. A safety limit still applies, so an open document can never stay connected indefinitely.

​

🗂️ Documents & search

• Browse documents in keyword folder views — a new Folders entry in the sidebar opens virtual folders built from the keyword values on your documents: a folder exists only while at least one document you can see matches it. Expand folders, filter by value or name, and open a folder to see its documents in the familiar results grid. See smart search.
• Folder views you administer — a new Folder views page under Admin settings lets you create, edit and delete folder views: choose the keyword that drives the folders, split them into subfolders by document type, pick which user groups see each view, and preview before saving. A sync action imports your existing folder configuration so the new views match it.
• Folder views, more reliable (opt-in) — administrators with the user-account administration right automatically get folder-view management, folder view names must now be unique (duplicates are renamed for you), you can save a view as disabled and enable it later, and a view can no longer be built on an encrypted keyword value. Editing is safer too: reordering levels saves correctly, invalid definitions are rejected with a clear message, and a view changed by someone else asks you to reload instead of silently overwriting. Browsing a large view is noticeably quicker.
• Faster document search — searches that filter by keyword, date or document type now return faster on large archives, especially when a filter narrows the results to a small set of documents; paging and oldest-first sorting use the same fast path. Searching with a specific keyword such as a person number or a case ID is now noticeably faster when that keyword alone matches few documents.
• Stable result ordering — document search results now come back in a guaranteed, consistent order, so paging through results and exporting always return the same documents in the same order. See smart search.

​

📊 Reporting: approve what queries may read, and colour your charts

• Allowed sources, managed in Settings — administrators can now approve, from Settings → Reporting → Allowed sources, the database tables and views that reporting queries may read — no developer needed. Adding a source checks that reporting can actually read it; a query referencing an object that isn’t approved fails with a message naming the object, so authors can ask an administrator to approve it. Revoking a source also removes the snapshot data of any data source built on it, so withdrawn data stops being readable immediately. See data sources.
• Per-category chart colours — in the dashboard designer you can now pin a colour to a specific category on a pie or bar chart, so the values that matter most stand out; categories without a pinned colour keep the theme palette, which now holds eight colours tuned to stay readable in light and dark mode. When a dashboard can’t be saved, the designer now tells you which element is blocking the save. See building dashboards.

​

🎨 Caseflow & administration

• Caseflow access by user group works again — granting a group the Caseflow feature (Admin settings → Access → User group feature access) now reliably opens Caseflow for its members, and the feature appears under its proper name rather than an internal one.
• Caseflow configuration permissions — who administers Caseflow templates and attribute settings is now controlled by the standard application permissions on the Permissions screen: Manage Caseflow Configuration allows creating, editing and deleting templates, generating templates with AI assistance, and configuring filter attributes, while View Caseflow Configuration provides read-only access. Existing grants are carried over automatically, so no one loses access. See Caseflow administration.
• Caseflow administration access restored — administrators whose groups hold the Caseflow configuration permissions can once again open Settings → Caseflow administration, with existing grants moved automatically onto the renamed permissions.
• Caseflow values stay exactly as stored — text and numeric attribute values that happen to look like a date are now passed through exactly as stored, instead of being silently converted (with a date such as a birth day in dd.MM.yyyy sometimes coming back day/month-swapped). Autofill and script-button actions in Caseflow also execute reliably once more.
• Caseflow filter prompts and sorting — a filter prompt whose required setting was never explicitly stated is treated as required again, so a search can’t run with it left unfilled, and filter bars and saved searches with no sort direction now sort ascending instead of unexpectedly descending.
• Column options, usable by everyone — the panel that opens from a column header (sort, filter, pin, hide) now works with screen readers, announced as a dialog for the column, and keyboard focus moves to a neighbouring column header when you hide or pin one instead of being lost. See lists & search.

​

📥 ContentFlow

• Dynamic keyword mapping — the Upload to Nobly Insight step now has a Dynamic keyword mapping option that maps each item property straight to the Nobly Insight keyword it is named after, with no mapper configuration needed. With dynamic mapping, the step stops and reports it if a property does not resolve to a keyword on the target document type rather than quietly dropping the value — you can switch back to the lenient behaviour if you prefer.
• Webhook callback step — ContentFlow workflows can now notify a third-party system over HTTP that a document has been archived. Place the step right after an upload step and it sends a templated JSON payload — document id, trace id, file name and more — to a partner endpoint, with OAuth 2.0 client-credentials or API-key authentication. A non-success response, timeout or token failure fails the step, so your workflow’s retry or exception path decides what a failed notification means.

​

🔎 More improvements

• Related-fields confirmation, only when it matters — adding or removing an instance of a keyword group bound to an autofill set no longer warns that related fields will be overwritten; the confirmation appears only when an edit would actually re-apply values from the set to an existing instance. See autofill on edit.
• Copying keyword assignments, faithfully — when you create a document type by copying from an existing one, the copied keyword assignments now keep the source type’s exact order.
• Cascading keyword defaults validated properly — a default value on a keyword governed by a cascading dataset is now validated against the cascade’s value list, so valid cascade values (such as a child value of the chosen parent) are accepted instead of being rejected. See cascading datasets.
• System-use keywords, easy to spot — administrators can mark keyword types as system-use, and those keywords now appear under their own System keywords heading in the keyword pickers instead of being mixed in with the regular groups.
• More accurate API errors — the REST API now returns the right error status — “not found” for a missing saved search, folder or conversation, “bad request” for a genuine bad request, and a server error for an unexpected fault — and error responses no longer repeat internal details, giving you a plain description and a trace id to quote when reporting a problem.

​

Nobly Insight — end-of-August update

​

🤖 AI Chat: a completely new interface

• Enhanced AI Chat, with a completely new interface (opt-in) — start from a search or a single document and ask follow-up questions without re-typing your context. Each conversation stays with the documents it started from, and answers keep the documents they draw on visible, bringing in related Caseflow items for context. See the search documentation for how AI Chat works.
• See the assistant work — a live status line shows each step as AI Chat prepares an answer, such as searching documents, reading a document or looking up a Caseflow object, and you can expand any answer to see the full list of steps it took.
• You stay in control — when the assistant suggests updating document metadata, nothing is changed until you review and approve the proposal.
• Quick prompts, aimed where they matter — administrators can define ready-made question starters and limit each one to the user groups and agents it is meant for, so the right conversation begins with a single click.

​

🧠 AI document classification: document type and keywords in one step

• One analysis, both answers (opt-in) — on the upload page, the robot button on a file row (or AI Index for every file) now reads the document once and suggests both its document type and its keyword values. The separate “guess the type” and “guess the keywords” actions are gone. See AI document classification.
• Driven by a classification tree you author — suggestions come from a tree of branches your administrator defines: what a document in each branch looks like, which keywords to extract, and guidance for matching values against datasets. The tree defines the complete set of types the AI may suggest, and without a configured tree the analysis makes no suggestions at all — author it first under Admin settings → AI → Document classification; see Configuring the tree.
• Runs on Nobly’s own AI — this replaces the previous type suggestion and keyword extraction built on Azure Document Intelligence. Classification and extraction now run on the Nobly AI platform (open-weight models on Nobly infrastructure), and scanned documents are read by Nobly OCR — no document content leaves Nobly-operated services.
• Confidence you can inspect — every suggestion carries a confidence score; hover it to see the path the document took through the tree and the evidence quoted from the document at each step. The score is deliberately conservative, and a document that matches no branch is left for you to index by hand. Users only ever see suggestions for document types they are allowed to pick.
• For administrators: what changed in settings — the AI Suggestion Settings and Document Intelligence pages have been removed. The tree page now holds the one OCR page cap (Max pages, default 20 pages) and the dataset-matching guidance. Previously configured page caps and Document Intelligence model mappings are not carried over, so set Max pages on the tree, and note that the AI context descriptions entered under the old metadata-suggestion settings are dropped the next time those settings are saved.

​

🔎 Search

• Unified result grids — Caseflow lists and document search results now share the same grid: a Find in results box to narrow the rows on screen, one Columns menu to show, hide and pin columns, and the same Export card everywhere.
• Export exactly what you need — the Export card shows your row count up front and lets you choose All results or Selected rows, as Excel or CSV; on document search you can also download the selected documents themselves as a .zip.
• Selection you can rely on — each grid opens the way you left it: row checkboxes stay on or off per surface, leaving selection mode clears the ticked rows, and closing a half-finished bulk edit asks before discarding your changes.
• See what you searched for — a Search filters button lists the criteria the current results were searched with.
• Shared links keep their meaning — a shared search link now always stands for the same search you were viewing. If a keyword filter cannot be represented in the link, the link is not created and a message explains why — and the same guarantee holds when you carry a search over to AI Chat, so a conversation never runs against a wider set of documents than the results you were looking at.
• Guided searches, more predictable — prompts marked required stay required when a saved search is reopened, leaving an optional field blank no longer stops a guided search or adds conditions you never asked for, and document search prompts are always required, so a search can’t run half-filled.
• Status lists stay reliable — the status markers on search results (redactions, annotations, email attachments) no longer fail for the whole list when a single document has since been removed.

​

🎨 Caseflow

• Relations between classes — administrators can add a relation attribute to a Caseflow class that links its objects to one object of another class in the same application, or to the class itself. The related class is chosen from a searchable list, and the attributes list shows where each relation points.
• Relations in daily use — relation attributes are validated when created and work with relation pickers and related-object navigation. See Caseflow administration for configuring attributes.
• Template editor, opened the way you like — the editor now opens with the preview as a wide row beneath it (you can still switch the preview to the side or above), and the AI assistant panel starts closed, opened from the toolbar whenever you want it.
• Keyword assignments, reordered safely — reordering a document type’s keyword assignments when the list includes keyword groups no longer fails with a conflict error, and moving a group keeps its members’ settings intact.

​

📄 Documents

• Redaction viewer in narrow panes — when the viewer is shown in a narrow space, such as the Caseflow document preview or a half-width window, the toolbars and action buttons now wrap onto extra rows instead of running off the edge, so Save and Discard are always within reach.
• Side panels that adapt — the thumbnail and redaction side panels now start closed when the pane is too narrow for them, and text-selection actions such as Mark for redaction stay inside the viewing area. See the document management documentation.
• Document not available — a link to a document that has been deleted or that you no longer have access to now shows a clear “Document not available” message instead of an error or empty screen. Deleted documents no longer open from any direct link — preview, revisions, history, activity, keywords and permissions all answer with not found.
• Redaction viewer, roomier on narrow screens — the toolbars wrap onto extra rows instead of running off the edge, side panels start closed when the pane is too narrow, and text-selection actions stay inside the viewing area. Typing an out-of-range page number now shows the page you land on.

​

📤 Capture: encrypted documents, handled safely

• Automatic decryption of received documents (opt-in) — documents delivered already encrypted, for example from an archive or migration feed, are now decrypted just before they are stored, so the readable content only ever exists inside the product.
• Nothing questionable gets stored — a document that cannot be authenticated with the configured key is rejected and nothing is stored. Enabling the feature requires configuring the decryption key and choosing which document types it applies to.

​

🧰 Office add-in: archiving that keeps your work

• Keep working after archiving, on desktop Office too (opt-in) — archiving the Word or Excel document you have open now reliably reconnects the open file to the archived version, so you can keep working and save revisions right away instead of ending up with a separate local file.
• No more “save changes?” for a document you just archived (opt-in) — after you archive from the add-in, the document is reopened from Nobly Insight in its own window and the copy you archived from is dealt with for you: closed without a prompt when the add-in can prove it was untouched, or — for a document that was never saved, such as a macro-generated letter, or one edited after the archive — a question in the task pane asking whether to close that copy without saving, so nothing is ever discarded silently. Word and Excel on Windows; PowerPoint and Office on the web keep a notice telling you which window is the live one.
• Save a revision straight from the reopened document — Save revision in the task pane now saves the open document to Nobly Insight as a new revision, carrying the comment you typed in the pane, and the confirmation says “sent” until the save has landed.

​

🗂️ Administration: keyword types and clearer lockouts

• Deactivate keyword types — you can now mark a keyword type as deactivated from the Keyword types screen. A deactivated type is no longer offered when you build a document auto-name, while existing auto-names and documents keep working as before. A Status column and filter help you keep track, and you can reactivate a type at any time.
• Auto-name picker, tuned — the token picker for auto-names now leads with the keyword types assigned to the document type you are naming, and leaves deactivated types out entirely, so the tokens that can resolve to a value are the easiest to reach.
• Clearer lockout reasons — when an account is locked for an unrecognised reason, the text shown in user administration is now formatted consistently whatever the language, instead of occasionally containing garbled characters.

​

📥 ContentFlow

• Submit without a file — items can now be submitted metadata-only when the file will be attached later by a workflow, recording a file name and MIME type as hints. A new step fetches the item’s file from an external Azure Blob container using credentials from global variables and attaches it as the item’s content, copying the blob’s metadata onto the item.
• Work on already-archived documents — new steps update the keywords of a document already archived in Nobly Insight (a value sets the keyword, an empty value clears it, and keywords the metadata no longer carries are removed), or delete it (a soft delete to the trashcan).
• More reliable connections and reruns — workflows no longer fail to authenticate when the token endpoint is left unset (the discovered endpoint is used instead), runs finish cleanly with temporary files removed exactly once, and a failed step can be rerun without the uploaded file being lost.

​

🛠 Configuration & auditing

• Search audit trail (opt-in) — when enabled, Nobly Insight keeps an append-only record of every search run on the search page: who searched, the terms and filters used, and the outcome, kept for a configurable period after which older entries are removed automatically.
• Configuration audit, easier to read — the audit log now shows exactly which fields a change altered with values before and after, and system-search changes are described in everyday wording — filter column, operator and value as you set them, names instead of internal ids — so you can see what an edit did at a glance.
• Accurate lock status — an account is reported as locked only when it actually is blocked from signing in. Unlocking or enabling a user resets the failed-login counter and clears stale lockout records, so the account can be used straight away, and the shown lockout reason no longer contains garbled characters.

​

👤 Sign-in & identity

• Full names from single sign-on — sign-ins through Microsoft Entra ID or another OpenID Connect provider now reliably carry the user’s full name, including middle names, into Nobly Insight instead of leaving it empty or incomplete.

​

✨ Nobly Insight — Late August Update

​

🎨 New: Caseflow template editor

• Write with guidance — autocomplete for every template component and its options, inline documentation on hover, and live checks that flag mistakes as you type, with click-to-jump problem navigation.
• See it before you save — a live preview pane renders your template through the real engine, against either generated sample data or a real case you pick. Preview is completely safe: nothing is saved or triggered.
• Start quickly — insert components from a palette, scaffold a working template directly from a case class, or begin from a library of ready-made starter templates (update form, create form, tabbed case view, and more).
• Version history — every save is kept. Compare any earlier version with your current draft and restore it with one click.
• AI-assisted templates (opt-in) — describe the screen you want in your own words and let the AI generate or refine the template. Changes are always presented as a proposal you review and accept — never applied automatically.

​

📧 Redact & annotate email files

• Emails in the redaction viewer — .msg and .eml files can now be opened for redaction. The email is converted to a searchable PDF (including headers and inline images), so text search works during redaction, and the redacted result is stored as a normal PDF revision.
• Clear feedback — when a mail file cannot be converted (for example encrypted S/MIME messages), you now get a message explaining exactly why.

​

🔎 Search: history, new columns & quicker filtering

• Recent searches — your latest searches are now remembered per user and available both on the landing page and in the search sidebar, labelled by what you actually searched for. Re-run one with a click; delete individual entries or clear all, and control how far back history is shown in your personal settings.
• File size column — file size is available as a display, filter, sort and grouping column in document search — handy for spotting oversized or duplicate attachments.
• Filter as you type — text columns in Caseflow grids now filter live as you type, the same way document search does, instead of picking values from a checklist.
• Keyboard-friendly grids — column sort/filter menus in search results and Caseflow are now fully operable from the keyboard.
• Trashcan for system searches — deleted system searches are no longer gone forever: administrators can find and restore them from a new Trashcan tab, and deletion prompts now tell you honestly what is recoverable.

​

📤 Smarter document upload

• Progressive keyword forms — keyword fields can now be shown or hidden based on whether another field has been filled in (new is empty / has value conditions), so upload forms reveal themselves step by step instead of showing everything at once.
• Default document type — administrators can choose which document type the upload page starts with, and users can set their own personal preference, so uploading starts with the right keywords already loaded.
• You decide on ambiguous AI matches — when AI metadata suggestions find several near-identical candidates for a field, the field is no longer silently auto-filled. A clear “multiple matches” indicator lets you pick the right value from a list — or explicitly leave it empty.

​

🧰 Office add-in: a fresh look & seamless archiving

• A look that matches Nobly Insight — the add-in has been redesigned to match the Nobly Insight web client, with full light and dark mode that follows your Office theme.
• Archive and keep working (opt-in) — archiving the document you have open in Word or Excel now reconnects it to the archived version: you continue working in the archived document, a normal save becomes a new revision, and Save Revision is available right away — no more “do you want to save changes?” for a file that is already safely archived.
• One-click revisions — saving a revision of an open document is now a single click, with the optional comment field available from the start.
• Steadier and clearer — the add-in no longer gets stuck with document actions disabled, the notice that someone else has the document open works reliably again, and upload errors show clearer messages.

​

🧷 Calmer editing of linked keywords

• One clear question — a single dialog names the field you changed, the keyword set it belongs to, and exactly which related fields will be updated — replacing two overlapping pop-ups.
• Asked once, not twice — confirming covers the whole editing session, and the question correctly returns the next time you start editing.
• Type in peace — text no longer disappears mid-typing while a background lookup runs; any adjustment waits until you leave the field.
• Clear as a unit — clearing one linked field clears its companion fields with it, so a half-cleared set can no longer be saved.

​

🗂️ Keyword administration

• Full options for keyword set members — keywords reaching a document type through a keyword set can now individually be made required, hidden, read-only or unique, and given default values — just like standalone keywords.
• Saves that just work — saving keyword assignments on document types with keyword sets no longer fails with a technical “sequence conflict” error.
• Honest duplicate warnings — the duplicate-assignment warning now only appears when there genuinely is a duplicate, and names the keyword types involved.
• Dependent value lists, polished — the cascading-dataset editor now uses the full window for long value lists, and deleting a dataset shows exactly what is affected — and reassures you that documents and their stored values are not.

​

⏳ Retention policies, more flexible

• Count from the date that matters — retention can now be calculated from any date keyword (for example an end-of-cover date), not just document or stored date.
• Match documents with missing values — retention criteria gain is empty / is not empty conditions, so a rule can select documents where a keyword carries no value at all.

​

🤖 Faster, more accurate AI

• Tune how much the AI “thinks” — administrators can now set a reasoning level per AI task (extraction, summarization, search…) directly in AI settings, with immediate effect. Turning reasoning off for extraction has shown the same or better accuracy at a fraction of the response time. Settings are validated against what the selected model actually supports.
• Forms extract better — document extraction now preserves the row-and-column layout of PDFs, significantly improving accuracy on invoices and other form-shaped documents.
• Mixed documents fully read — PDFs combining scanned and digital pages are now read in full: scanned pages are OCR’d individually instead of being silently skipped.
• Finer-grained AI permissions — AI administration is now governed by dedicated permissions (AI configuration, extraction profiles, search index) manageable in permission administration, instead of a legacy configuration right.

​

📊 Dashboards

• Scheduled refresh restored for parameterized sources — dashboard data sources using query parameters are refreshed on schedule again, can be refreshed manually, and show their true data freshness — while loading dashboards faster.

​

⚡ Performance & reliability

• The email attachment indicator in search results has been restored on affected environments, background processing and script execution have been hardened, and a range of stability and monitoring improvements land across the platform.

​

🔮 On the horizon