This release adds skills to AI Mind, which you create, import, share, test and promote to a company skill, a company Skills page in AI Configuration, an AI Report Builder that plans and builds report changes, report bookmarks that save a view, hide visuals and build a story, custom visuals you publish to a shared library, a Dependencies tab that names the reports a BI table feeds and a warning before you break them, workflow schedules that run in the time zone you pick, email subscriptions that send several times a day in their own time zone and can be tested on every recipient, layers and backgrounds on the DataPrism report canvas, AI functions you call in SQL, data safety guards that stop a load before it publishes something wrong, a Locks view that shows what waits behind a busy table, sign-in with your DataMind Platform account, the documentation itself opening in a drawer and answering questions inside the Minds and DataMind Oda, public report links and embeds, a rebuilt sign-in page that shows the DataMind products, and the DataMind Oda tools for Microsoft Excel, which now also search the web, run your company's skills and reach your connectors.
AI Mind: Skills
Skills tab: Every Mind now has a Skills tab beside Training Hub, with the skills it can load in four groups: This mind, Shared with this mind, Organization and Built-in.
Create and import: Write a skill in the editor, have the Mind write one with Create with AI, or import an existing SKILL.md file or a .zip. A skill the Mind writes starts turned off until someone reviews it.
Share and copy: Share a skill with other Minds, copy it to them as an independent skill, or convert a shared, organization or built-in skill into one the Mind owns.
History: Every skill keeps its revisions, with a diff between two revisions and a restore.
Skills in chat: The / menu lists the skills a chat can run, and creating or editing a skill from a chat shows a preview you approve before it is saved.
Source on every skill: A badge says whether a skill is built in, comes from your organization, was written by the Mind, or runs only when someone types its name.
Test a skill: Every skill can be tested. A benchmark compares it with running without it or with its previous revision, a trigger test shows which requests make the Mind load it, and a tuning run looks for a better description.
Tests panel and task bank: A skill's Tests panel keeps its tests, the practice tasks with the checks a good run must pass, the trigger tests and the reviewers' notes. Every test starts only after you approve its estimate.
Usage: A skill's row shows how often it loaded in the last 30 days and how many of those loads were followed by no tool call.
Skill tests in chat: Ask a Mind for a test in a chat. A test run adds its own steps to the conversation and to Tool History, and the strip above the message box reads Skill test with Open the results.
Administration: Skills
Skills page: AI Configuration now has a Connection & models tab and a Skills tab. The Skills tab lists your organization's skills together with the built-in and Mind skills.
Organization skills: Write a skill once and choose the agents it reaches, including Oda, the Minds, the builders and the AI tools. A skill can be limited to roles and given its own slash command.
Forced skills: Marking a skill Forced makes every chosen agent list it, and nobody can turn it off.
Delivery: A skill is Managed, so later edits reach everyone, or Install once, so each person keeps the copy they installed.
Promote a skill: A skill a Mind wrote can be turned into an organization skill after a review, and the Mind keeps its own copy. The revision needs a passing benchmark first, and the person who wrote the skill cannot be the one to promote it.
Oda skills moved: The skills you send to DataMind Oda users are now managed in AI Configuration > Skills. The Oda admin page keeps General, MCP and Feedback.
Administration: A rebuilt admin and a new look
Tabbed AI Configuration: The AI settings page and the Oda page were rebuilt on the shared administration layout, with inline field errors and an unsaved-changes bar that offers to discard before you leave.
MCP servers as a list: The MCP servers you send to Oda users are a table with their address, access and audience, and each one opens in a side panel.
Update replaces Platform: Admin > Platform is now Admin > Update, a services sidebar beside the tabs Updates, Map and Job logs. The Installer updates first, then DataMind OS, and old Platform links still open.
Uptime per service: Each row in the Update sidebar now ends with how long the service has been up, such as 4h 12m, and a service that has finished its setup reads Completed.
Container logs moved: Container logs now open under Observability > Logs > Platform, with match modes, level and exclude filters, a histogram, a live view and CSV export.
New wordmark: The sidebar and the loading screen carry the DataMind wordmark, and the loading screen now shows a progress bar and offers a reload if the app is slow to start.
Universe cards: Each product card on Universe has a shorter description and one clear action, such as Open Docs, Open Platform, Open FM and Download.
Administration: Roles
Role names: A new role's Role Name carries its format hint, Use English letters and underscores only, and is checked before the role is created. A name that starts with the platform's reserved prefix is refused, and the message names the prefix.
Table access that does not stick: Saving a role's tables warns Some changes were not applied when some tables kept their previous access, and says how many.
Role members: The members list leaves out only the platform's own seeded accounts, so a person is no longer hidden by what their name contains.
Permission names: Granular permissions follow one naming pattern, such as Create Users, Create/Edit Lake Objects and Execute/Stop Warehouse Objects, and a few Standard sets were renamed to match, such as User Permissions Editor and Chat Sharer. Roles and users keep the same access under the new names. Every name is listed in the Permission reference.
Editor: AI in SQL
Ask Kartlos about every row: A SQL query can call six functions, aiGenerate, aiClassify, aiExtract, aiTranslate, aiFilter and aiRedact, to write text, pick one of the labels you list, pull a field out of free text, translate, keep the rows where a plain-language condition holds or redact personal data. Every call uses Kartlos Standard and is billed to the company's standard bucket. See AI in SQL.
Off until an administrator turns it on: Switch AI in SQL on in Admin > System Configuration > AI in SQL. While it is off, every query that calls an AI function is refused, in the SQL editor, Minds, DataMind Oda and Warehouse transformations. A person also needs Use AI in SQL on their role, which Data Layer Administrator does not include.
Where they run, and where they do not: the SQL editor, Minds, DataMind Oda, the MCP query tools and Warehouse transformations, whose runs write the AI results into a table. They never run inside a view, a materialized view, a dictionary, a function or a column expression.
Limits: A query makes 1,000 calls by default, raised in the query with SETTINGS ai_function_max_api_calls_per_query from 1 to 20,000, and up to 5,000 tokens per allowed call in each direction. An AI query is not counted first and returns its first page only, so it cannot be paged, charted or downloaded: narrow it with WHERE or LIMIT, or write the results into a table with a Warehouse transformation. An answer comes back as plain text, without Markdown, unless your prompt asks for a format.
In the AI log: A call that comes from AI in SQL appears in Observability > Logs > AI with its model, status, duration and token usage, and always without its request or response content.
Workflows: Schedules in your own time zone
Timezone: A workflow's schedule now carries a Timezone, chosen by searching a city or a country.
Read as a sentence: The schedule reads as a sentence, such as "Runs in Tbilisi time (GMT+4), the same as your clock.", with a Change link.
Daylight saving: The start time, end time and schedule day are read in that zone, so a daylight-saving change no longer moves the run.
Default: A new workflow starts in your own zone, and a workflow with no zone keeps running in UTC.
Workflows screen: The list gains a filter pane beside the search box, a New button, counts on the type chips, a schedule state on every card, and a toggle that reveals test and single-node runs in the runs table.
Workflows: the run logs
Run Scope: The Lake Sync Log and a table's Transformation Runs gain a Run Scope column, so a log that mixes a workflow run, a Single Node run and a Test run is readable at a glance.
Processed Rows: The column counts the rows a production Fact run staged for publishing, instead of a target output a Fact run no longer leaves behind.
DataPrism: The report canvas
Layers: A Layers panel lists the page's visuals from front to back and reorders them by dragging, with Lock on each.
Order: Bring to front, Bring forward, Send backward and Send to back arrange a selected tile.
Text and image tiles: A new Text & media group places text boxes and images on a page.
Backgrounds: A tile and a page each take a background, a fill or one of the report's images.
Report assets: A Report assets drawer keeps the report's files in one place, shows where each file is used and refuses to delete a file that is still in use.
DataPrism: The Report Builder
A builder in the AI panel: An editable report's AI panel now holds the Report Builder beside Model AI. Describe what to build or change, and it plans the job, applies it and shows what it did.
Plan and approve: A big job shows a plan card first, listing the pages, visuals, slicers and measures it will make and anything it will delete. Edit the plan, then approve it. Auto-accept, Manual and Unrestricted set how much of the job is checked with you.
New report with AI: Start a report from a description in the report list, or hand the builder a screenshot or a sketch to rebuild a report from.
Review my report: Have the finished pages checked for overlaps, query errors, empty visuals, slow visuals and numbers that disagree, then fix what it finds, page by page or for the whole report.
Follow a job: The job view keeps a Timeline of who did what, with the model, tokens, cost and time of each step, and a Graph of the agents and the hand-offs between them. Mention an agent with @ to send it a message.
Hand-offs: The builder asks Model AI for a measure it needs and Custom Visual AI for a visual, and a large job runs one agent per page and closes with a check of the finished pages.
Activity: On desktop the running builder sits in the sidebar dock with the job's state and a Stop; on a phone it is a floating pill, which turns amber when a question parks the job.
Saves and versions: A job saves the report as a version before it starts and writes a version named Report AI: after each finished page, so you can always go back.
DataPrism: Custom visuals
Built-in visuals: Ten built-in visuals are ready to place, and a report or a page can be built around them.
One studio: A visual's code, options and a live preview now sit in one studio, with the code beside its preview twin and a Docs tab.
Publish and pin: A visual can be published to the library with an icon and a description, pinned for a team, and picked from the library when a tile is added.
Dock: Collapse all and Restore panels fold and reopen the right dock.
Slides: The DataPrism dock
Data in the deck: The Slides editor gains the DataPrism right dock, so a deck's data work stays in one panel, with Collapse all and Restore panels.
DataPrism: Public links and embeds
Public links: The Public links card in the share dialog shows a badge with how many links are active, and is not offered at all for a report built on a model with row-level security, though existing links stay revocable.
Embed header: A public embed now carries a slim header with the report name, the page title and question, Refresh, Present and Documentation. Add ?header=0 for a chromeless iframe.
Leave without saving: Leave without saving in the report editor now discards the report's unsaved edits instead of leaving the draft behind.
DataPrism: Bookmarks
Report and personal bookmarks: Save the view you are looking at as a bookmark on the report, in groups, from the Bookmarks menu while you read or the Bookmarks panel while you edit. Any bookmark can be copied to your own with Save to mine.
What a bookmark saves: the filters and slicer selections, the field-parameter option on screen, the marks, the drill state, the page, and the visuals you see.
Hidden visuals: Hide a visual from Layers or from the visual's own menu. A hidden visual is not drawn and runs no query, and a hidden slicer keeps filtering.
Bookmark navigator: A visual that steps a reader through the report's bookmarks in order.
Present with bookmarks: Present and the wall display step through the bookmarks, with a caption bar.
Export and subscribe: An export can follow your own view, the author's view, or put one slide per bookmark, and a subscription takes a view of its own.
Bookmark health: The panel summarises what each bookmark sets and warns when it points at something that is no longer there.
Switching bookmarks: Switching bookmarks updates the visuals in place, without a blank frame.
The navigator's size: A bookmark navigator can be made as small as 6 grid columns wide and 2 rows tall, and text boxes and images shrink to their own minimum instead of a chart's.
DataPrism: Calculated columns
A column from an expression: A BI table can hold calculated columns, whose value is a ClickHouse SQL expression over that table's own warehouse columns, worked out on every query and never stored.
Where you create one: From the BI tables panel of the model editor, or of a report you are editing, with New calculated column. The name is fixed; the expression changes later with Edit expression.
Preview and type: Preview runs the expression on 20 rows and shows the BI type and the database type before you save.
Where they work: Axes, legends, values, table columns, slicers and drill, in filters, cross-filtering, compare mode, Sort by and field parameters, and in measures by name, as in sum("Margin").
When one stops working: A column whose warehouse column was dropped, renamed or retyped turns red but is never deleted, and Health offers Fix. A warehouse column that arrives with the same name is held back instead.
In a model file: A .semanticModel file carries calculated columns and records format version 3.
AI reads and changes them: Model AI reads calculated columns, their expressions, whether they are broken and which measures use them, and it creates, changes, re-checks and deletes them under the same rules as the editor. The Report Builder asks Model AI for a column when a report needs a value per row, and Model AI can open one for editing from a report too.
DataPrism: Semantic models
Dependencies on a BI table: A BI table's popup gains a Dependencies tab, beside Bi Table and Preview, that lists every report using the table in order of report name. Under each name, chips name the columns the report reads in a chart or in a report filter, and clicking a name opens the report in a new tab. Reports you cannot open are counted, not named. The tab shows only to people who can edit the model, and it waits for the table's first save.
A warning before you break a report: Deleting a BI table or a measure, or turning a column off for reports, first checks which reports use it. It names the first five, counts the rest and the ones you cannot open, and the button reads Delete anyway or Hide anyway. A report counts when a chart or a report filter uses the item; bookmarks and uses inside the model are not counted. The check warns and never blocks the change.
Every path runs the same check: A BI table can be deleted from the model's BI tables tree, from its popup's Bi Table tab or with Remove from semantic model on the Diagram tab, and a column can be turned off for reports from the popup's IR column, the BI tables panel or the column toolbar. The report filters that target a BI table go with it, and the reports stay, without those filters.
DataPrism: The report editor
A sort for each option: A chart driven by a field parameter keeps the sort you chose for each option. Your own choice stays with you, the author's is saved in the report.
Currencies: The currency picker in a value format is now searchable and covers the whole ISO list of 61 currencies, with the ones people use most at the top.
KPI titles: A KPI card's title follows the measure under it when you rename or replace it, and returns to its default name when the card holds no measure.
Where a field lands: A field dropped in a well lands where you dropped it, with a line showing whether it goes above or below the field under the pointer.
Visualisation guides: A tile in Visualizations opens a short guide with a sample, what to use it for and when to avoid it.
Test a subscription: The subscription editor gains Send test, which stays off until you save the settings you changed.
DataPrism and Slides: Subscriptions
Several times a day: A report or deck subscription can send at up to 24 times a day, at least 1 hour apart. Add time adds one, and the frequency and day settings apply to every time.
Time zone: Each subscription has a Time zone, which starts on your own. Its schedule reads like Daily at 09:00, 13:00 (Asia/Almaty), and Next run follows the same zone. Existing subscriptions keep the time zone they already ran on, so no send time moved. The field reads as a sentence with a Change link, as a workflow schedule does.
Test to all recipients: Beside Send test email to me, Send test email to all recipients sends a test to everyone on the subscription the way a scheduled send would, marked [Test], without moving the schedule. It asks first when more than 10 people would receive it, and runs once every 5 minutes per subscription.
Where to find it: In DataPrism, in a subscription's menu and under Send test in the edit dialog, which replaces Send test to me. In Slides, under the Send a test email button on a subscription's row.
Warehouse and Lake: Data safety guards
A load that stops itself: A Dimension with the Full Reload or Diff Delete pattern, a Fact with a Truncate Insert pattern, and a Lake table with the Full Reload (Truncate Insert) sync type now stop a run before it publishes when its source returns no rows or when the run would remove more than a set share of the table. Nothing is published, and the table keeps its data. Every transformation and table with guards starts with both guards on and a limit of 50%. See Data safety guards.
Skip them for one run: Tick Skip safety guards for this run when you execute a transformation or synchronize a Lake table, and that run records what would have failed without changing your saved settings. A run that failed on a guard offers the same thing as Run once without guards.
Who changed them: History beside the guards lists every change and every skipped run, with the person, the time and the new values.
The builder Minds: A Warehouse Builder or Lake Builder Mind can set the guards and start a run without them with your permissions, and asks for your approval unless the chat runs in Unrestricted mode. See The Warehouse Builder Mind.
Lake and Warehouse: How a load publishes
Nothing half loaded: A Fact run publishes one day at a time, and a Lake load publishes the whole table or one day at a time, so a day is never left half loaded. A guard failure or a failed step leaves the table exactly as it was.
Upserts rebuild aside: A Lake upsert load builds the new version of the table beside the old one and swaps it in, so a load that fails never leaves a half-updated table.
Switch sync type without reloading: Changing a Lake table between Full Reload (Truncate Insert) and an upsert type keeps the table and its data, and the next run loads it as the new type.
Temporary ETL tables: Observability > ClickHouse > Storage lists the temporary tables a load keeps while it runs, flags them ETL temporary (read-only), and shows the run that owns each one with its rows, size and age. See ClickHouse storage and maintenance.
Warehouse and exports
Export what you see: A table export writes the columns you can see, in the order on screen, with the filters and search you applied, so a hidden column stays out of the file.
Large exports: Export all above 100,000 rows asks first, and an Excel export above 1,045,000 rows says so before it starts.
Static tables: Deleting rows in a Mappings tab saves them straight away once you confirm, so the separate Apply step is gone, and the deleted rows stay visible under the Deleted toggle.
AI primary key: A Refreshing or Appending view offers the AI primary key button on its Columns toolbar, and says why it cannot run: save the view, save the query, a personal-data column, or a key already set.
AI keys and personal data: Suggest primary key with AI now also runs on a table or on a view whose data includes personal-data columns, working from stand-ins in place of the real values. It still answers when it can read no sample rows, using the column names and types alone, and the comment helper is still refused on such a table.
Popups: A popup stays on screen after you zoom the browser or resize the window, and is drawn to fit while its saved position and size wait for the room to come back.
Observability
Trace a request: A record in Logs > Audit opens on Overview with a Trace tab, which draws the request's spans as a waterfall from the edge. Spans are kept for 30 days.
Request columns: The Audit table adds State, Errors, Service and Upstream Status, so a request still running, or one refused upstream, is visible in the list.
Live Metrics on your clock: Health-check times and outage windows now follow your own time zone, and each day of service health runs from your own midnight.
Alerts in their own time zone: An alert's schedule now carries a Timezone, which starts on your own and reads as a sentence with a Change link, as a workflow schedule does. The list and the details panel name the zone, such as "Daily at 09:00 Tokyo time", {{run_at}} follows it, and alerts saved earlier keep running at the same moment.
Alert recipients by address: An alert can also email addresses you type under Recipients, so a mail group or someone without a DataMind account receives it. Enter or a comma adds one, up to 50, each address receives the alert once, and the field needs Notify Others with Alerts, as a role does.
Observability: ETL locks
See what a table is waiting for: Observability > ClickHouse gains a Locks view beside Queries. It lists every holder of a lock, the person who started it and the server process behind it, and under each holder what waits and in which order. It needs View Active Queries or Manage ETL Locks, and the ETL Locks Manager preset bundles both. See ETL locks.
What a lock covers: A Dimension run and a periodic Fact run lock their own source system, so two source systems of one table load side by side; a Fact with any other pattern, a Lake sync and a Writeback load take the whole table; a transformation test run locks only its own work tables; an OPTIMIZE shares the table; a schema change takes the whole table and waits for the loads to end.
A busy table and a queued change: Changing a busy table offers Run when free, which queues the change behind the loads using it and runs it once. A queued change goes ahead of the loads that arrive after it, and the dialog tells you how it ended. See When a table is busy.
Abort a holder, or release a stale lock: Abort holder & release stops a holder and its ClickHouse statements and ends its run Aborted, with an optional reason kept in the audit log. A Stale lock whose process stopped heartbeating can be released instead, which queues its step again. See ETL locks.
The builder Minds: A Warehouse Builder or Lake Builder Mind can read the same list and abort, release, queue or cancel with your permissions. See Warehouse tool cards in chat.
DataMind Installer
Time zone: The installer's TZ value is picked from the list of time zone names instead of typed. An update keeps the value you set, a new install starts at Etc/UTC, and a zone the database cannot load is refused before it stops the deployment.
The Installer goes first: An action that needs a newer Installer opens its update dialog instead of only failing, and after a failed Installer update it offers Continue anyway. Update is offered also when only the published deployment files changed, and the dialog names them.
One command prepares the server: The published install command checks the host, prepares Docker and installs the DataMind Installer, asking before it changes anything. Run it again later and it keeps the host, channel and install directory you chose. See Install on a fresh host.
HTTPS or HTTP, chosen at setup: Configure Platform asks how people reach DataMind OS. HTTPS takes the domain and the certificate with its chain and private key, uploaded or as paths on the server, and checks the domain and the certificate before Finish turns on; HTTP takes the server's address and needs no certificate. See TLS and certificates.
Release Notes read the change log: The Installer's Release Notes tab shows the DataMind OS change log from datamind.guide, newest release first, and datamind.guide joins the domains the server needs outbound access to.
Continue with DataMind Platform: The Installer's first-run and sign-in screens take a DataMind Platform account, and its account menu connects or disconnects one.
DataMind Platform
Install the console as an app: The console installs from the browser and opens in its own window, as an app you can keep on your desktop. Its install prompt carries a DataMind Platform card.
Sign in to your servers: An Administrator can reach a company's DataMind OS and DataMind Installer servers with their DataMind Platform account, after an approval page that names the server and the product. See Sign in to your servers with DataMind Platform.
Sign In To Instances: A new permission, held only by the Administrator role, is what lets a company's administrators sign in to its servers this way. See Roles.
The app: sign-in and updates
A sign-in page that shows the products: The sign-in page is rebuilt. It opens like the loading screen, then shows the DataMind products beside the form with visible Email and Password labels. A failed sign-in no longer moves the card, and Reset your password uses the same full page.
Update ready: When a new version is ready, the sidebar shows Update ready with Reload and the account menu adds Reload to update. DataMind OS now also updates itself when you move to another section, and waits while an answer streams or an upload runs.
Status lines: The start screen reads Starting up…, and a sign-in that succeeds shows Opening your workspace… before your workspace opens.
The app: Documentation inside DataMind OS
Documentation left the sidebar: Documentation and BI are no longer sidebar items. The docs open from the Documentation card on the Playground, and an older Documentation link opens them too. See Read the documentation in DataMind OS.
The documentation opens in a drawer: datamind.guide replaces the in-app documentation. A link to a page in a Mind answer, an Ask AI answer, an Integrations tile or a connector form opens that page in a drawer on the right, with its title, its one-line description and its topic chips. Open on datamind.guide opens it in a new tab, and the drawer offers Back when you follow a link inside it.
Ask AI: Ask AI on the DataMind Docs card in Universe answers from the documentation, keeps no copy of your question, and can be pinned to the sidebar dock so it stays open while you work. See Ask AI about DataMind.
The Minds and DataMind Oda ask the documentation themselves: Every Mind and every signed-in DataMind Oda can look up the DataMind documentation and answer with links to its pages, in the desktop app, in VS Code and in Excel, with no Minds permission needed. Only the question and the area of DataMind it was asked from reach the documentation service. See Questions about DataMind itself.
The app: sign in with DataMind Platform
Sign in with DataMind Platform: Where the server offers it, Sign in with DataMind Platform is on the sign-in page: an Administrator of your company on the DataMind Platform checks the page DataMind Platform shows and continues, and two-factor is still asked.
The first administrator: A new installation no longer ships a bundled administrator. It asks you to create the first administrator, with a password or with your DataMind Platform account, and the page goes away once an administrator exists. See First sign-in to DataMind OS.
Connect from your profile: Profile carries Connect DataMind Platform and Disconnect DataMind Platform for an administrator, and neither is offered while you are impersonating someone, so an account that signs in this way can always find its server again.
Minds: Sharing and files
Share a chat read-only: A chat can be shared with several named people, for a set time, with a Live or Snapshot view. A guest reads the chat and cannot send, and either side can end the share.
Sharing moves into the chat's title menu: Share now sits in the menu next to the chat title, between Bookmark and Delete, and carries how many people can open the chat. A people icon next to the title marks a chat that is shared, before you write in it.
A note with a shared chat: When you share a chat you can write why and what to look at. The person reads the note above the message box and under Shared with me, and you can edit it later from People with access.
Allowed file types: Each Mind now chooses which file types people may attach, images, PDFs, text and Excel or CSV, in place of the single Excel and CSV switch.
Waiting for uploads: A message waits while a file is still uploading, and a file can be pasted while a message is being edited.
Editing a message: A file you removed while editing is not attached again, and an Excel or CSV table stays usable by the Mind after the edit.
Minds: Reading answers
Maximize on a table: The Maximize control on a table in an answer is always visible, on a desktop as well as a touch screen, and the last heading cell keeps its text clear of it.
Copying code from an answer: A selection that crosses a code block takes the code itself, with its line breaks and indentation, instead of the line numbers, and leaves Open in Query Editor out. A selection that covers only part of a block takes just that part, and Attach to message carries the same text.
A finished plan ends with the written answer: A plan that finishes ends with the Mind's own summary of what it did and what it found, and a plan whose only open step is the last one folds as N/N · all steps done instead of reporting a step it did not mark.
Minds: Scheduled tasks
Auto-pause: Every task can pause itself when nobody opens its last runs, from 1 to 100 or off. A new task starts at 5, or 15 on a Warehouse Builder or Lake Builder Mind, and Run now is never blocked.
Why a task paused: A task that paused itself says why, on the task and in the list: nobody opened its last runs, five runs failed in a row, the API key is invalid, or its creator can no longer use its Mind.
Tasks follow Mind access: A task can be created only on a Mind you can use, and Run now needs the same. A task whose creator loses the Mind, or whose creator's account is removed, pauses before its next automatic run.
Edit Others Tasks: A new permission, in its own Others' Tasks Editor preset and held by the Admin role, lets a user pause, resume and reschedule other users' tasks, change their triggers and set Auto-pause. The instructions, Mind and model stay with the creator.
History: Pausing, resuming and changing Auto-pause now create versions, and a pause the task made by itself is listed by System.
Time zone: A task's schedule now runs in its own time zone, chosen under Time with Change. A new task starts on your own, a task a Mind creates in a chat runs in UTC, next runs read like "Next run · Tomorrow · 09:00 Tokyo time", and restoring a version brings back its zone.
Tasks in chat: A Mind can now list your tasks, change, pause or resume one, run one now and delete one, as well as create one. A deletion waits for your approval on a card.
Minds: The Training Hub
Leaving unsaved edits: One Unsaved Changes dialog now covers the wizard, the details panel and the chat, and names what has unsaved edits. Discard & Continue drops them and Cancel stays, and Save & Continue appears while the Training Hub is open and saves the cards you changed.
Closing the upload dialog: Closing Upload files with files still pending asks Discard selected files? first, and Keep files returns to them.
One batch at a time: Files you add while a batch is still converting join that batch, and more than 50 files in one batch is refused with Upload blocked.
An interrupted upload: Reloading or closing the tab while files upload asks first, and a reload that cuts the upload short warns Upload interrupted and names the files that may not have been sent.
Audit columns in Sources: A table's audit and history columns are listed with an Audit badge, Select all leaves them out, and you can tick one by hand to grant it.
Selecting several cards: Batch select follows the Explorer's search and card filters, so a search narrows what Select all covers.
DataMind Oda
Oda in Excel: The DataMind Oda tools for Microsoft Excel arrive: workbook, cell and structure tools, data tools, charts and pivots, and formula tools, with permission modes, approvals before a write, change tracking, undo and a sign-out.
Oda in Excel, rebuilt: The pane is rebuilt on the DataMind look, follows Excel's own light and dark theme, and is English only.
Oda in Excel: the web: Web search and page reading sit behind one switch in the composer's + menu and in Settings. A page is read without asking only from a link you gave, a search result or a page Oda has already read, and a figure taken from the web carries its source.
Oda in Excel: skills: Seven built-in finance skills ship with the pane and run from the / menu, beside the skills your company sends to Excel.
Oda in Excel: connectors: The connectors your administrator makes available arrive with their own sign-in, the values you enter in Settings, and an approval card before a connector is asked to change anything.
Oda in Excel: files and code: Attach a file, run Office.js code in a sandbox, and run Python with pandas, numpy, openpyxl and scipy, each within the limits your administrator sets.
Oda in Excel: feedback: Rate an answer with Good answer or Not helpful, or report a harmful or wrong one with Report this answer, and choose whether the answer and your message go with it. Administrators read what arrives on Admin > Oda > Feedback.
Oda in Excel: memory and instructions: Memory keeps short notes Oda saves when you ask, and Instructions holds your standing rules. Both stay on your computer, off until you turn them on, and every change shows with Undo.
Oda in Excel: the conversation: Pick the model and its effort, keep typing while Oda works with up to five queued messages, choose the text size, and rename, delete, copy or download a chat. Each workbook keeps up to 50 chats.
Connection recovery: When a DataMind AI stream drops, Oda retries on its own, and a turn that cannot reconnect offers Continue, so the steps already finished are kept.
Organization skills in Oda: A skill your organization requires appears with a padlock and cannot be turned off; every other skill has its own switch on the computer.
A local server asks first: When your organization sends a server that runs on your computer, Oda asks before it starts and asks again if the command changes.
Oda from Microsoft Marketplace: Admin > Oda gains an Oda from Microsoft Marketplace switch, which decides whether an Oda in Excel added from Microsoft Marketplace may connect to this DataMind. With the switch off, that pane cannot sign in and its requests are refused, and the add-in downloaded from your DataMind is not affected. See DataMind Oda settings.
Publish to datamind.exchange: Text from an Oda answer, or the text you select in an answer or a Markdown preview, becomes a public link on datamind.exchange that anyone can open until it expires, with its headings, lists and tables. The datamind.exchange card in Settings > Connectors keeps your Published links, with Copy, Open and Unpublish. Administrators turn it on for the whole company with Ability to use datamind.exchange to share code snippets in Admin > Oda > General, which is off until they do, and nothing in the text is masked. See Publish to datamind.exchange.
Public links: Admin > Oda gains a Public links tab, where people with Manage DataMind Oda Shares and View DataMind Oda Settings see every link published from Oda in the company, search them by title, name or email, and unpublish any that still works. See Manage DataMind Oda for your company.
Browsing in scheduled tasks: A task can let its runs use Oda's browser on the websites you list and nowhere else, signed out or with your Windows work account, so unattended work reads a page without a person approving each site. Administrators allow it, and the sign-in method, on Admin > Oda > General; a run whose browser is refused ends Failed with the reason. See Let a run use the browser.
DataMind Platform: License Keys
API keys are for direct AI use: The API key card now reads "For direct AI API use. DataMind OS runs on License Keys, under Instances.", and a License Key never appears on it.
License Keys: Each DataMind OS install runs on its own License Key, which starts with dm-lic-. A server installed with the company API key keeps working on it until its DataMind Installer takes over a License Key at its next update.
Two new refusals: The API answers 402license_suspended when an invoice is past its AI cut-off date, and 402payg_not_available for a pay-as-you-go model during a trial. See Errors and retries.
Administration: License
Admin > License: A new License page shows the installation's license status, with its stage, its dates, its key and its last check-in, runs Check now and takes the installation's License Key. It needs the new View License and Edit License permissions, included in Admin Viewer and Admin Configurator. See License.
One key per server: Each installation runs on its own License Key, checked with the DataMind Platform about once an hour. AI Configuration shows the key read-only as Platform key, and Test Connection checks the saved key against the saved API host.
Banners, emails and the lock screen: Administrators see a banner and receive an email when a payment is overdue or when the installation cannot reach the platform. When AI and maps are paused, everyone sees why in chat, on maps and in a banner, and a locked installation shows a lock screen. Nothing is deleted, and scheduled workflows keep running.
The DataMind Installer: The install script and the setup form ask for this server's License Key. A server installed with the company API key exchanges it for its own License Key at its next update. See Secrets and registry access.
Updates while locked: A locked installation refuses DataMind OS updates and image pulls with Updates are paused: DataMind OS is locked for an overdue invoice. Installer updates still work.
No key yet: While no key is stored, the Installer shows No DataMind platform key: paste your API key or License Key., and Paste the key takes an admin straight to the key field in System secrets. The Installer's own update toasts now say Installer.
Lake and Minds: The Lake Builder
One import limit: An import takes up to 1,000 tables, whether you tick them in the source panel, upload an Excel file, ask a Lake Builder Mind or import into Writeback. A larger selection is refused before anything is created, with a message that lists the selected schemas, and the Excel rule of at most 50 tables per schema is gone. See Import tables.
Tables chosen by their columns: The Lake Builder judges a table by its columns and size, never by its name, names every table it leaves out, and asks before leaving out one that holds business columns.
Personal data before deploy: With Force Mind to Assign PII Domains to lake tables on, the Lake Builder assigns PII domains before it deploys or loads anything.
A final report: Every Lake Builder turn that changed the Lake ends with a report of the tables that loaded and ran empty, every table not loaded with its reason, the primary keys it chose, the personal data it tagged, and whether the workflow and its schedule are on. See The Lake Builder Mind.
Connections proven by a read: The Lake Builder calls a connector broken only after a read of its rows fails.
Sources in chat: Source paths in Lake tool cards read as breadcrumbs, such as Sales › sales_db › dbo › Orders, and connectors show their kind: Database, File, Meltano or ANY API. The source catalogue opens with counts, lists at most 1,000 rows and leaves out system databases unless you name one. See Manage connectors from a chat.
Source previews: A column whose name looks personal stays hidden in a source preview after its table is imported too, and the preview says why. An empty table still shows its columns.
Workflow and run cards: A Lake workflow card counts tables and deployed tables, has a Deployed column and shows a large workflow page by page. A run card counts every failed step, sorts skips under Skipped steps by reason, adds a Note column, and lists only the unfinished steps of a run with more than 50. See Workflows in chat.
Swarm reports: A swarm report shows as the Filed by swarm row in Tool History and as its own node in the Message Hierarchy, never as your message, and the Mind's answer to it uses the settings of your last message.
Lake: the Columns step
AI data types: The Columns step's toolbar gains Detect data types with AI (the AI Data Types popover) beside the comments and key helpers. It reads a sample of the source rows and each column's source type, proposes a Data Type for every unlocked column whose values call for a different one, and moves each of those columns to the Custom data domain as an unsaved change. Tell it what you know in the prompt, such as a column that must stay text, and your instruction wins. See AI helper agents.
Minds: Who can use builder Minds
Permission groups on builder Minds: An admin can now give a Warehouse Builder or Lake Builder Mind Permission Groups in its Settings, as on any other Mind. A published builder Mind opens to a member of those roles who also holds the kind's core permissions: View Warehouse Objects and Create/Edit Warehouse Objects for a Warehouse Builder, View Lake Objects and Create/Edit Lake Objects for a Lake Builder. Until now a Warehouse Builder was for admins only, and a Lake Builder opened to anyone with View Lake Objects. See Let other roles use a builder Mind.
A warning for a role that falls short: A role in a builder Mind's groups that lacks a core permission carries a warning that names what is missing. Saving still works, and the role's members see the Mind once it gains the permissions.
Admins manage builder Minds: Creating, changing, duplicating and deleting a builder Mind, and its permission groups, training cards and skills, is for admins only, and the builder kinds are offered under Mind Kind to admins only. Anyone else is refused with Only admins can manage Warehouse Builder Minds. or Only admins can manage Lake Builder Minds.
Still your own permissions: A builder keeps working with the permissions of the person who messages it, and a semantic model design, turning an existing view into a Refreshing view outside an approved design and a new host for a connector with a data access grant still wait for an admin's approval.
Access checked on every message: A member whose role is taken off a builder Mind's permission groups can still read the chats they own there, but their next message is refused with You can no longer use this Mind, so you cannot send messages in this conversation.
DataMind Platform: License invoices
License invoices: Billing gains a License invoices table for your DataMind OS instances: in your company's license currency, USD or GEL, billed in advance monthly or yearly, a Top-up for an instance added during a window, and VAT by your company's country. See Billing.
DataMind Platform: Instances
Instances: A new Instances screen lists the company's DataMind OS instances, adds one with its License Key and install command, and regenerates a key. See Instances.
Invitations: An invitee who already has an account signs in as the invited address before accepting, a new account is signed in straight away, and a company that sends too many invitations in an hour is asked to wait an hour.
DataMind Platform: Company setup and the product tour
Setup without an API key: Finish setting up your company no longer asks for an API key. It covers the billing requisite and the plan and, when your console offers Instances, then offers to install DataMind OS, a step you can skip. See Finish company setup.
A starting plan: A new company can start on a plan before anyone selects one. The Pricing plan step shows it under Selected plan, and an Administrator switches from it on Pricing like from any other plan.
Send the install to your IT team: The install step of the setup wizard lists what the server needs, and Copy instructions for your IT team copies the requirements, the install command and the License Key as one message. See Send the install to your IT team.
The product tour: The first time you open the console after setup is finished, a short tour of its menu starts. The question-mark button in the header, Take the tour, runs it again. See The product tour.
Fixes
A subscribed PDF or PPTX export leaves hidden report pages out.
A weekly or monthly subscription that is created, edited or resumed sends on its first matching day, today included when a send time is still ahead, instead of skipping a week or a month.
The Validate button on a transformation runs with the parameter values you set.
The Streams tab of a recipe no longer fails when it is opened outside the Lake page.
A batch user upload refuses a file over 5 MB, with a message naming the limit.
The default notification channel can no longer be switched off until another channel becomes the default.
Interact on a custom visual now works in Prime and in every report editor.
Folding a panel in the right dock no longer closes the rest of the dock.
A busy AI service is reported with its own message instead of a failed answer, and a turn that waits for a busy service can be stopped.
A swarm agent whose question nobody answers carries on by itself after 24 hours.
Web search now reaches the Warehouse Builder, the Lake Builder and the Meltano Mind, whose Web toggle showed but did nothing.
A task that watches workflows says which trigger it watches instead of showing a schedule it does not have, and a failure trigger no longer fires on failures from before it was switched on, or on a failure the workflow has recovered from.
Users, the system settings, the password policy and the theme list offer a retry when they fail to load.
Manage Tables is no longer offered in an Administrator's row menu, where it opened nothing.
A single sign-on configuration that fails on its last step names the field to fix.
A View query that uses WITH RECURSIVE now passes Validate, instead of being refused for a database the check cannot read.
Opening a cell in a Data Mapping grid and leaving it no longer counts as an edit, and Escape discards what you typed instead of saving it.
The run timeline in a workflow's header stays in place when there is nothing to plot, reading No runs in this scope yet, offers Retry when it cannot load, and follows the filters and search set on the runs table.
A run duration that came back negative now reads as 0 in the runs table, on the timeline and in exports.
A collapsed table in a semantic model's Diagram grows to show every column used in a relationship and every key column instead of stopping at 10, and right-clicking a table's card starts a new relationship from it.
A custom visual whose drawing sandbox cannot start reads The visual sandbox couldn't load. A browser extension or security policy may be blocking it. rather than blaming the visual.
A map visual narrowed by a cross-filter redraws its places instead of failing, and reads Placing {n} rows on the map while it places them.
Deleting a BI table that a report filter targets no longer fails, and the report filters go with the table.
An Oracle DATE column now arrives as DateTime64 and keeps the time it carries, instead of arriving as Date32.
A custom table's Source preview runs the table's own query and shows the rows it returns, instead of coming up empty.
A server the install command has just prepared offers Start instead of Update.
A column's Statement now applies on a table whose source is a file, such as an Excel workbook, instead of being left out of the load's query.
The recovery dialog in the DataMind Oda app says how its background service ended, such as stopped with exit code 0 or stopped by <signal>, instead of only stopped.
Ctrl+Shift+U toggles the text you selected between upper and lower case in every SQL editor, and the Keyboard shortcuts window lists it as Toggle upper/lower case in Editing, in place of the two case rows that never did anything.
A prompt typed to the assistant inside a connector's code editor or recipe editor is no longer lost when you minimise the popup or close and reopen the assistant panel.
A tree-row shortcut no longer takes the key while you type in a SQL editor: Enter in a Query step inside a Warehouse popup inserts a new line again instead of opening the highlighted row.
A new AI connection starts switched on, so a fresh setup reaches the models without a second step.
A Finished workflow run no longer offers Restart; start a new run with Run instead.
Unpublishing a brand theme removes it from everyone who is not its owner, instead of leaving the colours on their screens.
A role's table access can be changed only with Edit Role Table Permissions, and a refusal names the permission.
A workbook saved as Strict Open XML Spreadsheet is refused with its own message and the Save As fix, in the Lake and in a Mind chat.
A History run on a table with calculated columns no longer fails.
A custom-script table's Source preview reads its query the way a load does, so a script the datasource refuses reports its error when you preview it instead of passing and failing later at load time.
A workflow starts even while a schema change is queued for its table: the run begins and its steps of that table wait behind the change.
The Warehouse card on the Playground offers every Warehouse object it can create, a semantic source, a logical model, a transformation, a dimension, a fact, a dictionary, a static table and a view.