Changelog

dbt Integration

Connect your dbt project once and every model becomes a governed cube. The dbt integration converts dbt models into cubes — dimensions, measures, column descriptions, primary keys, and joins inferred from dbt relationship tests and foreign-key constraints — so you never redefine your transformations by hand, and dbt stays the single source of truth.

Sync however fits your workflow: pull manually from the IDE, call the REST endpoint from CI right after dbt run, or trigger on every push with a webhook. Every automated sync lands on a review branch that a human approves before it reaches production. Cube connects to your repository over HTTPS or SSH deploy keys and only ever parses your project — it never touches your data warehouse. Read more in the announcement.

Other improvements and fixes:

Workbooks

• You can now drill down on a time dimension in Workbook Table charts and the results grid
• Introduced a new-tab experience that lets you quickly start an exploration from a single launchpad: browse or search your semantic model and source tables, or duplicate any existing workbook tab

Dashboards

• Read-only viewers of a published Dashboard can now subscribe to and unsubscribe from its scheduled refresh notifications
• Scheduled refresh email notifications can now be sent to user groups

Visualization

• Added a unified, context-aware color selector shared across chart series, maps, and conditional formatting
• Fixed chart dates not matching the results table by anchoring chart time values to UTC

Administration

• Added a tenant-wide setting to restrict CSV downloads, plus a deployment-scoped Download Data permission for custom roles
• Added a deployment-scoped Manage secrets permission to grant access to one deployment's configuration and credentials without exposing others
• You can now select historical Cube versions in any release channel

AI

The Cube MCP server can now edit the semantic data model, not just read it — external agents can open a dev branch, write model files, and commit changes for review.

Embed

Creator Mode embeds can now hide the Semantic SQL and Generated SQL views, so embedded end users can't see your data source's table and schema names.

Integrations

• You can now search reports and folders by name in the Google Sheets add-on
• Users with access to a shared report can now refresh and place it in the Google Sheets and Excel add-ons, even if they didn't create it

Scheduled Tasks

Answering the same question every Monday morning shouldn't depend on someone remembering to ask it. With Scheduled Tasks, you save an agent prompt once, pick a cadence — hourly, daily, weekly, monthly, or a custom cron expression — and Cube runs it on schedule: querying your semantic layer, creating and updating reports and dashboards, and emailing results to workspace members.

Every run produces its own thread in Analytics Chat, so you can open any past run and ask follow-up questions in context, and each task's detail page shows its run history at a glance. Tasks run under the security context of the person who created them, so results respect row-level security and access policies. Create one from the new Scheduled section in the sidebar — or simply ask the agent in Analytics Chat to set it up for you.

Other improvements and fixes:

Workbooks

• Added subtotals to pivoted tables: turn on Subtotals in the totals menu and every pivot group gets its own "Total for" column, computed by the database so non-additive measures stay correct
• Fixed the fill in missing rows option resetting the query's sort order

Dashboards

Creators can now edit a published dashboard directly from its chat — ask the agent to add a chart, change filters, or adjust the layout, and it updates the draft; an Open in builder button then carries the conversation into the builder so you can review and publish.

Visualization

• You can now display a numeric column as a sparkline in Table charts, rendering the measure as a mini line, area, or bar chart over time
• Added a freshness indicator to charts in Workbooks and Dashboards: a leaf icon that shows how long ago the served data was refreshed
• Fixed Table charts ignoring date formats set in the Results tab
• Fixed a brief no-data flash while a Table chart loads

Administration

Chat History in the admin console now labels where each conversation originated: Web, Slack, MCP, API, or a scheduled task.

AI

• Added Claude Sonnet 5 support, both as a predefined model and for BYOM
• The Auto agent now uses Claude Opus 4.8 for data modeling
• The Cube MCP server now includes read-only searchDataModel and runQuery tools, so external agents can explore the data model and run governed queries directly
• Analytics Chat can now build map charts
• Custom MCP connectors now support API key, Basic Auth, and custom header authentication

Embed

Deployments can now authenticate Chat and embedding API requests with their own Cube authentication (JWT/JWKS) instead of a separate embed token, via a new deployment setting.

Modeling

• dbt pull can now generate joins between cubes, inferred from dbt relationships tests and foreign-key constraints
• You can now select which branch of your dbt repo to sync, and dbt connection settings moved from environment variables into deployment settings
• View-level default_ui_filters now accept relative date ranges (like "5 weeks ago" to "today") and use the same operator vocabulary as the workbook filter bar

Workspace

You can now rename a workbook from its context menu in the Workspace.

Content Validator

Renaming or deleting a field in the data model can silently break saved reports and dashboards that reference it — and you usually find out from the people whose charts stopped working. Content Validator, a new tab in Data Model IDE, scans your saved content and flags every report that references a measure, dimension, or view that no longer exists.

Validation is branch-aware: run it in dev mode or on a shared branch to see exactly which reports a model change would break before it merges. For each broken draft report you can jump straight to it, replace the missing member with a valid one, or delete the report; published dashboards link back to their drafts so you can fix and republish.

Other improvements and fixes:

Workbooks

Added a fill in missing rows option so time-series charts render every date in a range, even ones with no data.

Dashboards

• Report widgets in the dashboard builder now keep their title and settings menu in every state, so you can fix or replace a broken or slow-loading widget without deleting it
• Time-granularity controls now automatically apply to every chart from the same view, matching how dashboard filters work
• The filter sidebar now organizes views into view groups, matching the Data Pane
• Fixed a layout flicker when a dashboard first renders

Visualization

• Added border controls to the Table chart: horizontal, vertical, header, totals, and outer borders, each with its own color and width, plus one-click presets
• You can now set per-column widths on the Table chart, with fixed and flexible modes and fit-to-content

Administration

• You can now scope an API key to specific public API endpoints
• You can now change your password from the Preferences page

AI

• Expanded the Cube MCP server beyond the chat tool: external agents can now create, update, and publish dashboards
• Added Slack to the MCP connectors available to the Cube agent
• The Cube MCP server now supports allowlisting custom OAuth redirect URIs, so hosted AI tools like Portkey and n8n can connect

Embed

• Your app can now listen to typed iframe events from embedded surfaces (ready, view, download, drilldown, AI query) and drive them with actions like setting the color scheme or a filter
• Creator Mode embed users can now share workbooks, folders, reports, and dashboards with embed user groups
• Embedded surfaces can now be localized into 12 languages, and Creator Mode workspace titles can be set per language

Integrations

• A report can now be placed in multiple spreadsheets at once in the Google Sheets and Excel add-ons, refreshing independently in each
• Fixed report editor issues in the Google Sheets and Excel add-ons, including pivot zones overflowing when a query has many fields

Modeling

Fixed unresolved validation errors when pushing a data model to Snowflake semantic views.

Workspace

• Cube now shows a notification when a new version of the app is available
• Fixed newly saved explorations not appearing in the Workspace until a page refresh

Cube Evals

Cube Evals brings deterministic accuracy checks to your Cube agent. Pair natural-language questions with known-correct answers, run the agent against that suite, and get an objective accuracy score plus a per-case breakdown of what failed. Grading compares result sets directly — it tolerates numeric variance and ignores column aliases — so there's no LLM judge, and every failure shows the agent's SQL next to the ground-truth SQL.

Eval cases live as YAML in your data model repo, so you can run them on a branch in front of a code review, inside AI Studio. Read the full story in Introducing Cube Evals.

Dimension links

A number on its own is a dead end. Add links to a dimension in your data model and they appear as actions in the menu that opens when you left-click any results-table cell — so a reader can jump straight from that number to the context behind it. There are two kinds:

• Dashboard drill-ins navigate in-app to another dashboard, auto-filtered by the clicked row — for example, "View Customer" opens that customer's profile dashboard.
• External links open a per-row URL in a new tab, so you can deep-link into a CRM, docs, or a search.

Other improvements and fixes:

Administration

• Added a Boolean user attribute type — a True / False / Unset toggle — across user attributes, per-user values, and the security-context override editor

AI

• The agent now warns when a query's results are truncated to 100 rows
• Fixed the agent occasionally saving a dashboard report with no chart

Embed

• Added an Embed preview playground for admins to test chat, dashboard, and Creator Mode embeds
• Create embed tenants directly from the admin Embed section
• Embedded surfaces, including Creator Mode, can now be localized to Spanish
• Fixed a crash in embedded dashboard agent chat when the agent returned SQL
• Export embedded dashboard widgets as CSV

Modeling

• Review a dev branch's unsaved and uncommitted changes from anywhere in dev mode, and revert individual files
• Organize views into nested view groups

Workspace

• Share a folder directly from its header

MCP Connectors

A number tells you something moved; the reason usually lives in a ticket, a doc, or an error tracker. With MCP Connectors, the Cube agent can now reach the external tools your team already uses — and pull that surrounding context into the same conversation, while every answer stays grounded in your semantic layer. We're launching with Notion, Linear, Sentry, and Attio, with many more connectors on the way — and you can already point to any remote MCP endpoint with a custom connector today.

Ask "activation dropped last week — what changed?" and the agent can combine the metric with the features that shipped and the errors that spiked. It works both directions: results can be written back as a Notion page, filed as a Linear ticket, or posted wherever your team works. Set up connectors from the connector directory; administrators choose exactly which tools the agent can use — nothing is enabled by default — and every connection respects your existing access controls.

Each user connects their own accounts: connectors that need a personal login are authorized per user and can be reviewed or revoked from Connected accounts in your preferences. Read the full story in Introducing MCP Connectors.

Other improvements and fixes:

Workbooks

• The SQL Runner is now available in Explore, not only in Workbooks
• Semantic view hierarchies now render in Explore
• You can now delete a workbook directly from the workbook page
• Workbook widgets now show structured, readable query error messages

Dashboards

• Added a Duplicate action to the dashboard and workbook builder menus
• Dashboard builder filter and time-granularity pickers now show dimension type icons

Visualization

• Added color scale (heat mapping) for the Table chart, so cell backgrounds shade by value
• The Table chart can now render values as in-cell data bars with a Display as control

Administration

• Added a SQL Runner permission to the custom role builder so you can control who can run raw SQL

AI

• The Cube agent can now search the web and read linked pages, bringing current, real-time information into the conversation alongside your data
• The Cube agent can now email members of your workspace, for example to send analysis results or a summary

Embed

• You can now share a Creator Mode dashboard with a specific embed tenant, not only with all embed users

Integrations

• Semantic view hierarchies now render in the Google Sheets and Excel add-ons
• The Google Sheets and Excel add-ons now follow your light and dark theme

Modeling

• The Security Context tester now has an Advanced mode for overriding Cube Cloud groups and user attributes, so you can preview exactly what any user or group sees across Explore, Workbooks, Dashboards, and Chat
• Added BigQuery support for dbt pull

Usage Analytics

The new Usage Analytics page gives account administrators a set of pre-built dashboards that show how your deployments are used: query volume and latency, cache efficiency, user adoption, and AI activity with token consumption. Each dashboard answers a different question about your account — Overview, Query Activity & Performance, Users & Adoption, and AI & Token Tracking.

Usage Analytics is itself a Cube application: your account's usage telemetry is modeled as a curated set of views — API Requests, AI Usage, and Users & Adoption — and the pre-built dashboards are regular Cube dashboards on top of them. It runs the full Cube experience in Creator Mode, so you can explore the usage data, create workbooks, and assemble your own dashboards — anything you build stays private unless you share it. Available on Premium and above plans.

Other improvements and fixes:

Workbooks

• Column and row totals enabled on the Results tab now carry over when you switch to the Table chart
• Fixed the run options menu being disabled on new empty queries, which made the Auto-run toggle unreachable

Dashboards

• Scheduled dashboard emails are now rendered per recipient: each recipient gets the dashboard refreshed and captured under their own security context

Administration

• Build only selected partitions from the pre-aggregation partition list
• The Custom Roles page now remembers your sorting
• Fixed time range filter drag and navigation bugs on the monitoring pages

AI

• Added Claude Fable 5 support for the agent LLM selector and BYOM
• Fixed agent-built line charts sometimes rendering as an empty "Configure your chart" state
• New conversations now appear in the Chat History dropdown immediately
• Stopping a chat response now settles the UI right away, and follow-up messages send normally
• Fixed the MCP Server endpoint URL for deployments in dedicated and BYOC regions

Embed

• Creator Mode users can now duplicate dashboards shared with all embed users and build on top of them
• Rename or hide the embedded workspace title from the App Theme page

Integrations

• Added least-privilege mode to Semantic Layer Sync for Tableau Server and Tableau Cloud

Modeling

• Views can now declare default workbook filters via meta.default_ui_filters — selecting the view in a new workbook pre-populates them as editable, removable filters

Workspace

• Raise a collaborator's inherited access on a subfolder directly in the share dialog

Agent Skills

Agent Skills let teams capture repetitive analytical workflows as reusable, named procedures that the agent runs consistently every time — across Analytics Chat, Workbooks, and Dashboards. Instead of re-explaining the same multi-step analysis each time, you define it once and the whole team runs it the same way.

There are three ways to invoke a skill: click its button in chat, type / for a slash menu filtered by title and description, or just ask in plain language and the agent matches your request to the right skill and runs it. Once you've picked one, you can refine it inline — "for EMEA, last 6 weeks" — and the agent adjusts the run accordingly.

Skills are authored as plain Markdown files in your project's agents/skills/ directory: frontmatter plus plain-language steps, no special syntax. Running a skill is available to Explorer and Viewer roles; authoring one requires data-model edit permissions. Read the full story in Introducing Cube Agent Skills.

Other improvements and fixes:

Workbooks

• Stop and cancel running queries from Explore and Workbooks
• Disable auto-run in Explore and Workbooks
• Reworked the Table chart's Fields tab — separate visibility and labeling controls, plus a 2×2 pivot grid (columns/rows/values)
• Rename a Workbook by double-clicking its title
• Fixed: the semantic view picker in Explore re-opening the previous view instead of the one just selected

Dashboards

• Filters fall back to an unfiltered query when a filter doesn't apply to a widget
• Dashboards now remember your layout compaction setting

Administration

• CSV export on the billing Statements tab
• Refreshed the deployment Resources logs view
• Per-deployment cloud storage now supports Google Cloud Storage and Azure Blob Storage over OIDC

AI

• The Dashboard Agent can now change filters and time granularity
• Attach a dashboard chart to the Dashboard Agent as follow-up context

Embed

• Fixed a 404 when opening an admin-published dashboard shared with all embed users in Creator Mode
• Item owners are now hidden in embed Creator Mode

Integrations

• The Google Sheets and Excel add-ons now match Explore formatting for column and row totals (bold totals values)
• Added a "Loading data…" marker in the sheet while a query runs

Modeling

• The Merge button appears immediately after you commit on a branch
• Developers can override the global security context when testing queries

API

• Public REST API for notifications
• Public API endpoints for managing folder contents

Workspace

• Newly created Workbooks now sort to the top of "Last Viewed"

Dashboard Agent

Published Dashboards can now include a built-in Dashboard Agent so viewers can ask questions about what they're looking at. A chat bubble appears in the bottom-right corner — click it to open a side panel and the dashboard reflows alongside it.

The agent is read-only and grounded in what the viewer sees: it knows the dashboard's widgets, the charts they render, and the viewer's active filters and time grains, so answers always reflect the current view. Ask it to summarize the dashboard, explain a trend, or compare widgets — much like Analytics Chat, but for consumers rather than editors. Conversations persist in the URL, so a refresh restores the thread right where it left off.

Region Maps

The Map chart now supports a Region mode that plots query results as a choropleth over polygon GeoJSON. Choose a built-in source — World Countries or US States — or point to any custom HTTPS GeoJSON URL, then join it to your data on a region property. Cube auto-detects the join fields and color-scales regions by your measure, with optional shading for unmatched regions.

Other improvements and fixes:

Workbooks

• Creating a Workbook no longer prompts for a name — it's auto-generated and you can rename it later
• Drill-downs now carry over filters from filtered measures
• Charts now show the pre-aggregation indicator only when results are actually served from a pre-aggregation
• Transient query timeouts no longer flash an error while a chart is still loading

Dashboards

• Improved the Time Grain widget edit sidebar

Administration

• You can now customize the app theme to match your brand
• Deploy with CLI mode now shows the connected Git remote

AI

• The MCP server now supports multiple agents
• Added a loadQueryResults MCP tool to paginate through large query results
• Agent configs now support llm: auto to automatically select a model
• AI-generated data models now default cubes to public: false

Embed

• App theme customization extends to embedded apps, so they can match your product's branding
• Creator Mode workspaces now come seeded with starter dashboards

Chats Sharing

You can now share an Analytics Chat thread with teammates as a read-only conversation. Click Share in the chat header to grant view access to individual users, a user group, or flip General access to Organization to make the chat visible to everyone. Use Copy link next to Share to grab a direct URL once access is set up.

Recipients open the chat at the same URL as the owner and can scroll through the full conversation, expand the agent's reasoning and tool calls, and explore the charts and tables it produced — but they can't send new messages. Only the owner can continue the thread, and the header shows a "Shared by …" label so viewers always know whose conversation they're reading. If the owner asks more questions later, those messages and the agent's replies show up for viewers on the next load. Recipients still need access to the deployment and the underlying data the agent used; otherwise they'll see a "Chat not available" page.

Other improvements and fixes:

Workbooks

• Reworked the chart configuration panel with cleaner tabs and persistent panel state
• Added Table chart row banding and hover row highlighting for easier scanning
• Table charts now support conditional formatting, so you can style cells based on rules like greater than, contains, or is null
• Charts are now color-scheme aware so colors stay readable when switching between light and dark themes
• Default and maximum row limits now follow your deployment's model configuration

Dashboards

• PNG and PDF exports now include the dashboard's active filters and time grains
• Admins can now delete custom Dashboard themes directly from the dashboard builder
• HTML charts now adapt their background color to the surrounding theme

Administration

• Added a new Data Sources settings page to list, add, edit, and delete database connections without hand-editing environment variables
• You can now map IdP groups to Cube custom roles for SAML and SCIM so users automatically receive the right roles based on their Okta or Microsoft Entra ID group membership
• Added a Build pre-aggregations permission so admins can build custom roles that prevent users from triggering pre-aggregation builds
• Added an Access AI agents permission to grant deployment-scoped access to AI agents in Analytics Chat, Workbooks, the IDE, and Dashboards
• Refreshed the Custom Role Builder layout for clearer policy editing

Embed

End users can now share Workbooks they create in Creator Mode with each other, so saved explorations can be collaborated on from inside your embedded app.

Integrations

• Tableau Semantic Layer Sync no longer hangs when the personal access token lacks permission to create a project
• The Google Sheets and Excel add-ons now require an explicit Run Query click after reordering pivot rows or columns when auto-run is disabled

Modeling

Views now support default_filters so you can apply filters automatically to every query against a view.

Workspace

• Table sort order now persists across navigation in the same session
• Clicking the Star icon on a workspace row no longer triggers row dragging
• Restored the empty workspace state interactivity so primary actions are clickable
• The Deployment switcher now stretches to fill the flyout sidebar

Access Policies Viewer

The new Access Policies viewer surfaces every access policy defined in your data model — row-level filters, member-level restrictions, and member masking — broken down by the user groups they apply to. Audit who can see which cubes and views, and how each policy is composed, without grepping through cube files or running test queries.

Open it from the Model module by clicking Access Policies in the sub-sidebar. The viewer reflects whichever branch and build you are currently on, so policies you are editing in development mode appear alongside what is live in production.

New Dark Theme

We've rebuilt our dark theme from the ground up to improve its look and feel — with refined contrast, more readable typography, and polished surface styling across every Cube surface. Switch between light and dark themes from Preferences.

Other improvements and fixes:

Workbooks

• Added a query info badge that surfaces cache and pre-aggregation status for the current chart and links to Query History
• All charts now use a transparent background by default for better theme integration

Dashboards

• Published Dashboards now render charts immediately on first load instead of waiting for filter dropdown suggestions
• Dashboards now gracefully handle missing members in widgets instead of showing an error

Administration

• Added a Configuration Settings tab in Deployment Settings to manage default time zone, row limits, and query timeout
• Chat History admin page now supports filtering by user, agent, and creation date range
• Sensitive agent environment variables are now masked in admin responses

AI

• Certifying a query from chat now opens a dialog to add it as a certified query file in your data model repository, on a new dev branch
• Conversations started from MCP clients like Claude Desktop and Cursor now appear in your Cube chat list
• The AI agent now asks for confirmation before converting model files between languages
• The AI agent no longer asks for a branch name when entering dev mode

Embed

• The generate-session API now accepts a userProfile payload to set display name and avatar for external users
• Embedded dashboards now use dynamic viewport units so they fill the available height correctly on mobile browsers
• Dashboard preview in Creator Mode now stretches to the full viewport height

Integrations

You can now duplicate an exploration with Save as in the Google Sheets and Excel add-ons.

Modeling

• The branch switcher is now available in the IDE toolbar when viewing production, so you can switch or create branches without first toggling dev mode
• Dev Mode now explains why it is unavailable when it can't be entered