Changelog

MCP servers

You can now connect MCP servers to Count, giving the agent a growing toolbox of external capabilities to draw on as it works. Plug in the tools your team already uses and let the agent reach beyond your warehouse to get the full picture — pull issues from Linear, enrich a customer from HubSpot or Stripe, search Slack for the context behind a number, or fetch a doc from Google Drive, all without leaving your analysis.

Connection-level agent context

You can now set agent context at the connection level, alongside the workspace and project levels you already had. Capture the conventions of a specific data source once: how its tables relate, which columns to ignore, what those acronyms mean, and every agent working against that connection inherits it automatically. You can also now update workspace, project and connection context through the public API, so your guidance stays in sync as your data does.

Fixes

• Report cells that depend on hidden Jinja cells are now scheduled correctly
• Images are now included when exporting frames to PDF
• Fixed Slack community URLs

Dashed lines

You can now use dash styles on lines — both as a property you set directly on a line and as a new visual mapping that encodes a field as dashes in your charts. Highlight forecasts and projections, or split a series by category, so your charts read clearly at a glance.

Built-in boundary data

Count now ships with built-in country, US, EU, and UK boundary data, so you can map your data the moment you need to — no hunting down shapefiles or wiring up external geographies first. Go from table to map in seconds.

Resume agent over MCP

A new resume-agent tool over MCP lets you continue an existing Count agent conversation from your connected MCP client, picking up exactly where it left off. Carry analysis on across tools without losing the thread or starting over.

Fixes

• Prevent clipping of arrowheads when a line changes direction sharply
• PDF export now skips hidden and malformed layers, and no longer culls off-screen cells
• Catalog repo setting toggles now update optimistically in the UI
• Fixed the position of the feature hint card

New workspace home

We've redesigned the workspace home so the work that matters is the first thing you see. Find your work faster, jump straight back into what you were doing, and get oriented the moment you land.

Table totals

Tables now support totals and sub-totals, so the summary numbers your audience looks for are built right into the table instead of stitched together by hand.

DuckDB query profiles

You can now inspect query profiles for in-browser DuckDB queries to see exactly how they execute, making it easier to understand what's slow and tighten up heavy queries.

Fixes

• Disable PR comments and checks for GitHub-connected catalogs
• Fixed amending the row limit on Snowflake's UNPIVOT (... FOR ...)
• CTE detection in Databricks
• View keys are now preserved when navigating between frames
• Nicer error messages for model network request failures
• Show all possible OAuth scopes
• Fixed the delete subscription button in the Slack UI
• Handle executing multiple DuckDB cells that depend on the same parent
• Added Redis TLS and tenant key prefix support

ClickHouse support

Count now connects natively to ClickHouse, including over SSH tunnels. Point Count at your ClickHouse cluster and build catalogs, canvases, and agent-driven analysis on top of it just like any other warehouse, bringing your fastest data into the same workflow as everything else.

Public API for canvases

New API routes let you read canvas contents and execute cells programmatically, so canvases become building blocks you can drive from your own apps, scripts, and pipelines; not just the Count UI.

Fixes

• Improved SQL syntax highlighting
• Place labels on maps now wrap, and their opacity is preserved in PNG exports
• Right-click drag now works on maps regardless of user preference
• Expressions in low-code cells with trailing comments now parse correctly
• Better handling of catalogs connected to GitHub repos with more than 100 branches
• Fixed an auth bug when uploading workspace fonts
• Entra ID tokens now refresh correctly
• Fixed fetching queries with expired URLs

Agent updates

We've shipped a major update to the Count agent that makes working with it faster to follow, easier to return to, and far more capable.

As the agent makes changes, the canvas follows along, panning to keep its edits in view and highlighting the objects it's touching — and stepping aside the moment you pan or zoom yourself. Anything it does can be reverted and reapplied straight from the chat, and whatever you've selected is automatically added to its context. Past conversations are now a searchable dropdown grouped by when you worked on them, and a new home screen gets you started fast: upload a CSV, connect a database, or load demo data, then pick from analysis questions tailored to your source.

Under the hood, the agent is more powerful too. It can now read the canvas on its own and edit any cell, text, or frame — not just the ones it created — with new support for maps, sorting, legends, formatting, labels, and tooltips. It can search the Count docs to answer product questions, and it checks that the cells it edits actually run, catching errors as it goes.

Agent in Slack

The Count agent now lives where your team already works. Ask a question in a channel or a DM and get an answer back without opening Count — turning everyday Slack threads into a front door for your analytics.

Faster, smaller data transfer

We've improved how data moves between your browser and Count, shrinking the amount sent over the wire. The result is noticeably quicker loads and smoother canvases, especially on large results and slower connections.

Fixes

• Control the title style of multiple cells at once
• Default mark color now depends on layer index
• The sources pane can now be resized
• Improved weekly spacing of axis ticks
• Fixed flexible joins for aggregate-only catalog queries
• dbt metadata is now fetched for alerts
• AI agent features are correctly hidden when disabled for a workspace

Public API and MCP server

Count now has a full public API and a hosted MCP server, so everything you do in the app is available to your own tools, scripts, and agents.

The REST API is documented by a public OpenAPI spec, so you can generate clients and explore every route. It's authenticated with service accounts, created and managed right from the workspace UI, and authorised exactly like a user, so they can only reach what they've been granted access to.

The MCP server is the governed front door for agents into your workspace. Connecting goes through OAuth tied to a real user account, so an agent acts as that person and inherits their permissions, with optional scopes to lock it down even further.

There's also a new home in the app for managing service accounts and OAuth clients (and sharing resources with them), an auto-generated reference for every API route and MCP tool, and interactive request builders so you can try routes without leaving Count.

Fixes

• Service accounts can now refresh schema metadata
• Existing count.co links now navigate in-app on eu.count.co where appropriate
• Join paths are now deterministic regardless of the order in which views are selected
• Code text is now included in markdown output
• Storage API errors are now surfaced to users
• More consistent hiding of UI during product tours
• Various dependency and security updates

Snowflake semantic models

Count now reads Snowflake semantic models directly. The metrics, dimensions, and relationships your team has already defined in Snowflake flow straight into your catalogs: no re-modelling, no drift between the two definitions, and a single source of truth that both your warehouse and your agents can rely on.

Open Semantic Interchange

We've added support for the Open Semantic Interchange (OSI) format, so your semantic definitions are no longer locked to one tool. Move models in and out of Count freely and keep your metric layer portable across the rest of your stack.

Fixes

• Improved validation and autocomplete performance in catalogs with large numbers of tables
• Project group admin roles are now respected
• Fixed stale text shaping when falling back to a substitute font
• Canvas tooltips now dismiss correctly when the cursor leaves during a right-click pan
• Relaxed the conditions under which a default connection is set for new catalogs

SQL improvements

Calculated fields in visual/low code cells and expressions in field definitions in catalogs now support several additional SQL patterns:

• :: cast notation (e.g. column::varchar)
• extract(... from ...) syntax
• count() without arguments
• Window calculations in filters

Fixes

• Fixed canvas socket reconnection when switching tabs
• Fixed DuckDB cells incorrectly showing row limits
• Fixed agent query submission and execution timeouts
• Improved agent error messages
• Fixed agent history handling for interrupted conversations

Automatic schema refresh

On paid plans, database schemas now refresh daily at 4am UTC so that new tables and columns are available to your team and agents without manual intervention. Admins can adjust the schedule in connection settings.

Agent database tools

Agents can now explore your database schema and catalog files to find relevant data as they work. When an agent needs to run a query against your database, it will ask for your permission — you can approve it once, or grant access for the rest of the conversation.

Magic links

Magic links now must be opened in the same browser that requested them. This prevents automatic virus scanners in email clients from invalidating links before you get a chance to click them.

Fixes

• Non-admins can now invite users to projects
• Reverted dynamic text format controls
• Improved end of trial UX
• Fixed Snowflake OAuth code challenge method
• Fixed Redshift handling of single-selects

Workspace and project context

You can now set workspace and project-level context to guide agents. This helps agents understand your team's conventions, data sources, and preferences without repeating instructions in every conversation. (Project-level context takes precedence over workspace-level context.)

Flexible catalog joins

Catalogs now support two modes for defining joins. In the first, datasets define how their views are joined together (similar to Looker). In the second, views themselves define joins to other views, and at query time Count automatically works out the optimal join path between those views that are selected.

Cell metadata

A new cell metadata modal lets you inspect metadata for individual cells directly from the canvas, including in-browser DuckDB memory usage and cache expiry status.

Fixes

• Fixed various page crashes
• Improved handling of errors when calculating memory usage
• Fixed color palette saving
• ClickHouse connections via MySQL protocol now work correctly
• Schema filters now available for all database types
• Fixed an issue where filters weren't updating in certain cases

Count Agent

We're excited to introduce the Count Agent: a new way to collaborate with AI on your analytics. The agent works alongside you in the canvas, where everything it produces is inspectable, adjustable, and reusable. It can run over 100 queries from a single prompt, distributing work across your data warehouse and Count's compute layer for accurate, comprehensive analysis. Combined with Count Metrics for governance and guardrails, the agent turns complex analytical questions into answers in minutes, not days.

You can also set context at the workspace and project level to help agents understand your organization's conventions, terminology, and preferred approaches when working with your data.

Explorer role removal

The explorer role has been removed. Existing explorers have been migrated to analyst permissions with appropriate access levels.

Fixes

• Fixed an issue where OAuth redirects could occur too early, preventing client parameters from being updated
• Comments with leading spaces are now allowed
• Improved error handling for BigQuery schema refresh when service accounts have reduced dataset access
• Fixed circular dependency issues in query execution
• Ensure queries are never executed with the wrong DAG
• Catalog socket is now properly closed when navigating away from view pages
• Independent scales now work correctly with null values
• Fixed issue with catalog file directory rendering when searching with collapsed folders

Multi-region support

Count is now multi-tenant with dedicated regional instances. The original count.co has moved to eu.count.co for European users, and a new US tenant is available at app.count.co (the default for new workspaces). All existing count.co links automatically redirect to eu.count.co.

Maps

We're delighted to release maps in Count. You can plot points, shapes, and create choropleths using the same drag-and-drop low-code functionality you use for more traditional visualisations - no more learning Python just because you need to plot bars not bars! Maps support circles, text, and geoshapes rendered from GeoJSON or WKT fields in your data.

Nested frames

Frames can now be nested, allowing you to create more complex layouts and organize your canvas with greater flexibility. Nested frames are visible within their parent in the presentation view, are hidden/locked when their parent is hidden/locked, and if automatic filters are scoped to a frame, they also apply to all of its children.

Double width stickies

Sticky notes can now be set to double width, giving you more space for longer notes and annotations.

Fixes

• Fixed styling of loading state in the onboarding checklist
• Relative timestamps now update every minute
• Style options that don't apply to the current visual are now disabled

Share connections

Connections can now be shared with your workspace or specific groups, making it easier for teams to collaborate with shared data sources.

Fixes

• Better links to documentation in the catalog editor sidebar
• Results with runtime errors are no longer mistaken for expired results
• Cells referencing catalogs now correctly bypass view caches that are still being generated

Google Sheets

You can now connect to Google Sheets as a data source. Import your spreadsheet data directly into Count and combine it with your other data sources for analysis.

Independent scales per facet

Faceted charts can now have independent scales for each facet, allowing you to better compare patterns across groups with different value ranges.

Catalog-wide caching options

You can now configure caching settings at the catalog level, with the ability to reset canvas-level caching settings when needed.

Fixes

• Bar labels now display correctly when the independent axis is continuous
• Fixed an issue saving canvas explore cells to a new canvas
• Copy/paste events now work correctly when visual tooltips are active
• Fixed domain adjustment for reversed scales
• Cached results are now correctly ignored when caching is disabled at the workspace level
• Improved styling of unfocused Python code
• Fixed an issue where users copy community canvases into their workspaces

Improved view compilation

We've improved how views are compiled and scheduled. Related cells are now compiled together when constructing views, and view queries are scheduled more efficiently.

Fixes

• Colour overrides now work correctly for boolean fields
• Improved session timeout error messages
• Jinja queries now use more generic result fetching logic
• An issue with the frame viewport when navigating from a project has been fixed
• Empty frame names are no longer shown unless the frame is hidden
• The error "table ... must be qualified with a dataset" has been fixed in the explore UI
• Copying links to objects when viewing the report now directs to the report rather than the canvas
• Fixed an issue where text disappeared when switching focus from one object to the next

Automatic cell layout

The automatic layout feature, which arranges cells according to their DAG, is now available in the right sidebar. Previously only accessible via keyboard shortcut (Option-Command-L on macOS, Alt-Ctrl-L on Windows), you can now easily organize your cells with a single click.

DuckDB on the server

You can now choose whether each DuckDB cell runs on the client, on the server, or let the system decide automatically (which remains the default behavior).

Improved cell connections

We've enhanced how cell connectors are laid out to better represent your data flow. When a connector exits from the bottom of a parent cell, it will now preferentially connect to the top of the child cell, creating a clearer and more intuitive visualization of your DAG.

dbt 1.10.13 support

We've added support for dbt version 1.10.13.

Fixes

• Copy link button now works correctly when viewing reports
• Empty frame names are now hidden unless the frame itself is hidden
• Catalog dataset selector now includes helpful labels when sharing catalogs
• Improved error detection for DuckDB queries with multiple parents

GitHub as a catalog source

You can now use GitHub repositories as external sources for your catalogs. This enables you to manage your data catalog definitions alongside your code, with full version control and collaboration features. Changes are validated automatically through GitHub status checks, ensuring your catalog remains consistent and error-free.

Partial catalog imports

When adding a catalog to a project, you can now choose to import only specific datasets rather than the entire catalog. This gives you more granular control over what data is available in each project, making it easier to maintain clean, focused data environments.

Faster project data loading

We've added a new optimized route that significantly speeds up loading dataset lists in the manage project data modal. Large catalogs now load much faster, making project configuration more efficient.

Fixes

• Improved empty state messaging when filters are enabled
• Fixed errors that occurred with empty repositories
• Line number clamping now works correctly in the catalog editor
• Catalog files are no longer truncated when saving
• Added validation button to check catalog status for any branch at any time
• Minor UI improvements throughout the GitHub catalog integration
• Fixed tooltip display for long catalog names in the project data modal
• Improved handling of GitHub committers with no display name
• Dropdowns with a non-DuckDB source are now run on the database by default
• Fixed an issue with alerts for cells with hidden inputs

Rich output previews

Cells now display rich output previews, giving you better visibility into your results without needing to view rich output directly. This makes it faster to iterate and understand what your code is producing.

Key-pair authentication for Snowflake

We've added support for key-pair authentication when connecting to Snowflake. This provides a more secure alternative to password-based authentication and aligns with enterprise security best practices.

Fixes

• Updated Snowflake account identifier instructions for clarity
• Fixed view cache table display when no datasets are present
• Corrected text truncation issues in various UI elements
• Improved integration parameter passing in job submission
• Updated documentation links from docs.count.co to learn.count.co throughout the application
• Fixed dropdown behavior to be consistent between viewer and analyst roles
• Character offset calculations in code input layout now work correctly

Catalog file management

You can now upload and download catalog files directly through the interface, making it easier to manage your data catalog definitions. This streamlines the process of maintaining and sharing catalog configurations across your team.

Support for reverting to earlier commits

Need to go back to a previous version? You can now revert to earlier commits in your canvas history, giving you more control over version management and making it easier to recover from unwanted changes.

Fixes

• Find and replace now works correctly for Python cells
• Changes are only submitted when users have actively edited the input, preventing accidental submissions
• Color overrides now work properly for non-string columns
• Slack avatars now display correctly in the chat widget
• Help widget no longer appears when logged out
• Fixed query handling when downloading local visual results
• Improved query optimization for identifying joined views
• Applied aliases correctly to selects from deduplicated aggregated fields
• Added memory protection when loading repository buffers to prevent out-of-memory errors

Improved magic link authentication

We've replaced our Firebase-based magic link implementation with a new, more reliable system. This provides better control over the authentication flow and improves the overall login experience when using email magic links.

Default local cells workspace setting

You can now set a workspace-level preference for whether new cells should default to running locally or on your data warehouse. This helps teams standardize their workflow and reduces the need to manually change cell execution settings.

Visual layer downloads

When working with multi-layer visuals, you can now download data from all layers individually. This makes it easier to export and analyze different aspects of your visualizations separately.

Fixes

• Improved font fallback
• Format controls now appear consistently across different contexts
• Boolean filter token text displays correctly
• Tooltips now appear more reliably in appropriate contexts
• Optimized query generation to prune unnecessary joins in inner queries for catalog cells
• Fixed handling of nanosecond-precision timestamps
• Prevented connection updates when errors occur
• Added ability to copy debug info from error messages for easier troubleshooting

CSV parsing options

We've added comprehensive CSV parsing options to give you more control over how your data is imported. You can now customize delimiters, quote characters, escape characters, and handle edge cases with confidence. This makes it easier to work with CSV files from different sources and formats.

Labels for BigQuery jobs

Count now adds labels to BigQuery jobs, making it easier to track and monitor your query execution costs in BigQuery's cost monitoring tools. This helps you understand which queries are driving compute costs in your data warehouse.

Fixes

• Improved subscription creation flow to properly handle 3DS verification for payment methods
• Improved floating point number display to prevent truncation
• Fixed CSV download options for local queries to show the correct available formats
• Fixed invoice display to show more than just the first 10 invoices

Rank/percentile table calculations

We've added Rank and Percentile calculations to help you quickly contextualise your data. Now you can easily answer questions like "Which products are in the top 10%?" or "How does this region rank against others?". You can configure each calculation to sort in ascending or descending order, allowing you to easily identify the top or bottom values in your data.

Support for BigQuery structs in autocomplete

No more manually typing out nested field paths. Our autocomplete engine now intelligently suggests nested fields for BigQuery, reducing errors and speeding up your workflow. Support for other databases is on the way!

Fixes

• Improved logic to account for labels in default axes domains, allow line/area labels to overlap series, improved nice temporal domain logic
• Fixed ordering of visual selects
• Six weeks are now always shown in the date picker such that the height doesn't change when it's anchored to the bottom of the screen
• When connecting a filter to control cell, the existing operator is now retained
• Marks legends can now be styled
• Table calculations can now be applied to calculated fields
• Table filter cells can now be edited in all appropriate situations as expected
• A couple of broken links in the connection setup guides have been fixed
• Shift + scroll always horizontally scrolls now
• It's now possible to bin within a measure

Catalog branches

You can now create multiple branches in your catalogs, enabling teams to collaborate on changes in a safe, isolated environment. Preview any branch within your canvases to ensure everything works as expected before merging your changes into production.

Automated catalog creation

Save time by automatically generating entire catalogs, or adding to existing ones, directly from your database tables. We intelligently infer relationships, identify the correct fields for views, and configure the necessary joins to build your datasets.

Enhanced catalog visualizations

We've improved the ways catalog queries can be visualized. This update includes support for data binning and better normalization handling.

Window functions in aggregated queries

Perform more complex analysis directly within your visual and table cells. Calculated fields now support aggregated window functions, allowing you to create calculations like running totals and rankings on the fly.

Fixes

• Fixed rendering issues with large decimals
• Fixed database connection setup guide links
• Fixed an issue where tooltips weren't disappearing as expected
• Fixed a performance regression that caused certain canvases to hang when loading
• Snowflake schemas are now always retrieved, regardless of case sensitivity settings
• Fixed an issue whereby custom fonts that didn't have a variant with font weight 400 couldn't be selected

Automatic filter references

Control Cells can now be attached to tables/cells, affecting all downstream no-code and visual cells as a global filter. This behaviour can also be scoped to specific frames. This makes it quicker and easier to add filtering behaviour to entire canvases without having to manually connect controls to each cell.

Visual and table usability enhancements

Tooltips on visual cells can be pinned with a right-click, allowing for values to be copied.

Hyperlinks within visualisations are clickable by holding Command (Mac) or Control (Windows).

Clicking rows within tables now opens up a vertical preview pane that lets you easily see and copy wide data at a glance and navigate up and down the table with keyboard arrows. Nested structs are also presented in an expandable format.

Large CSV export optimization

Previously downloading CSVs would often run a new query. Now you have the option to convert any cached results to CSV, then download that.

You can now download up to the query result size limit configured for your workspace.

Creating CSV files is quicker for large results - expect around 40 seconds per GB.

Sidebar improvements

It is now easier to navigate left-hand sidebar when you have many tables visible. Hovering over a schema or table name reveals a "hide others" button that filters the view just to that object, and also narrows the scope of the search bar.

Analysts can optionally see the column types next to their names.

BigQuery nested column types can be unnested using the disclosure arrow.

Fixes

• Resolved 'data missing' issues for visualizations sourced from catalog views, particularly for large cached datasets
• Optimized table pane processing for very large BigQuery database schemas (>1M items) that caused browser lag
• Fixed inaccurate file size reporting that affected Parquet conversion decisions
• Resolved issues with multi-select control cells using non-string column types (integers, dates, etc.)
• Fixed "waiting for upstream cells" errors in Count Metrics explore mode

Fixes

• Prevented visual breakage when clicking on chart axes, which previously caused DuckDB binding errors
• Fixed Snowflake result set mismatches where queries in Count returned different values than direct Snowflake execution
• Resolved ILIKE ANY query issues in Snowflake cells that were working in Snowflake but failing in Count
• Fixed missing labels and tooltips for single data point series in line plots
• Resolved canvas merge issues where duplicate overview content would be deleted incorrectly
• Improved error messaging for DuckDB catalog operations involving control cells and CASE statements
• Fixed invalid index generation errors that occurred during canvas object ordering

Enhanced SQL autocomplete

Major improvements to the SQL editor including partial token completion, better table name matching when schemas aren't explicitly defined, column disambiguation showing source tables, and improved function completion.

Fixes

• The sort dialog for visualizations no longer closes after each selection, allowing you to configure sort type, column, and direction without repeatedly reopening the popup.
• Resolved issues where CSV exports would fail silently preventing downloads
• Fixed inability to change data update schedules on alerts through the UI
• Resolved color copying issues when duplicating canvases between workspaces with different color schemes
• Fixed missing column listings for joined cells in the low-code table sidebar until queries were executed

Fixes

• Fixed SQL cells that suddenly stopped working in canvases, particularly those referencing catalog datasets
• Improved connection retry logic for scheduled canvas runs to reduce "connection terminated unexpectedly" and "failed to fetch" errors
• Enhanced null value filtering capabilities in table cells
• Resolved memory issues in catalog cells when working with large datasets
• Fixed catalog view dependency conflicts that occurred when views and dependencies had identical names
• Improved query evaluation and state management to reduce evaluation failures during canvas refresh

Fixes

• Fixed workspace search modal accessibility from the Explore UI (Ctrl+K shortcut now works)
• Improved catalog schema updating and synchronization across workspaces
• Enhanced performance for large canvas operations and object management

Aggregated window functions

You can now aggregate window functions in-line within visuals and field definitions, e.g. sum(sum(Sales)) over (partition by region order by month(date) asc).

DuckDB cells are now materialised

Each DuckDB cell being run in the browser is now materialized and used in the calculation of any downstream DuckDB cells, as opposed to previously where DuckDB cells compiled all upstream cells as CTEs.

This fixes various performance issues caused by aggregating/filtering with large upstream cells.

Canvas redraw performance improvement

We've improved logic that calculates z-indices of objects, accelerating the canvas load time for large canvases with long edit history. This also speeds up other actions involving redraws such as switching between presentation/report views and the canvas.

Fixes

• Fixed various issues using the clipboard on Safari/Firefox
• Fixed catalog cells with downstream non-catalog visible cells, in which the views for those catalog cells are never executed so the visible cells were waiting forever
• Fixed an issue where Count temporarily failed to connect to Snowflake on AWS due to certificate errors
• Fixed OOM errors when reading parts of large result sets from the server
• Fixed an issue where links shifted on defocusing when placed after a newline
• Mentions are now correctly shown for workspace guests and autocomplete ordering of users has been improved

Fixes

• Drag/drop is now disabled for folders on mobile
• Resolved an issue with moving canvases to new projects and not clearing the folder
• Fixed an issue where the new folder button was not always shown
• Local queries are now consistently re-executed after cell renames
• Local control cells can now select from columns in any other cell
• Parent cells are now executed when selecting filter / adding new filter to table filter cell
• Various issues around referencing measure name / values columns in downstream cells have been fixed

Project folders

Project folders allow you to organize your canvases in a familiar folder system with drag-and-drop, full tree view, and integration throughout Count for saving and moving canvases.

This new feature increases the available flexibility in how you can structure your workspace. From using multiple workspaces (on Enterprise plans), multiple projects with their own access and data source restrictions, and now folders.

Creating folders requires an analyst or higher user role.

Fixes

• Fixed an issue in which CSVs couldn't be downloaded in the explore UI
• Fixed a case where viewers couldn't download data from cells attached to a catalog
• Local cells with complete results now download directly from the client
• Fixed a bug with the week start logic when using a truncated date picker
• Fixed an issue where the validation on saving a catalog was weaker than that in opening a catalog
• The workspace sharing setting is no longer disabled when the canvas is shared with the project
• The alert sidebar now overflows as expected
• The overview modal / sidebar now overflows as expected
• Boolean columns are now supported in Sankey diagrams

More flexibility in presenting canvases

User feedback told us that it was challenging to transition existing reporting patterns to Count. So, we have introduced more flexibility in how you share and present canvases.

Firstly, the sidebar Report button has been changed to "Present". This more accurately reflects its function from the canvas view.

Once in the present mode, there is a new default "Report" setting. This allows you to present scrollable views of taller frames and move between them with a new navigation bar, in addition to our canvas overviews. It's great for reproducing dashboards, reports and other detailed content.

We have maintained the existing behaviour as "Slides", adding the ability to more easily present these with a fullscreen button.

We believe this makes it easier to replicate existing workflows within Count, and to encourage stakeholders more familiar with scrollable reports to engage with data while maintaining the ability to jump into the underlying canvas to explore.

Canvases are shared or presented. Presentations should be explorable and questions answerable in a canvas.

Existing canvases maintain the current/previous behaviour to avoid breaking changes to the experience. /report URLs will continue to function as expected.

Field grouping in Count Metrics catalogs

As we continue to evolve Count Metrics we've made it easier to create highly usable catalogs by introducing grouping to fields. A group key can be added to view fields which will place them within an expandable list similar to that currently used for aggregations.

Fixes

• Flickering of object bounds when selected by others
• Race condition on moving a canvas to trash when within the canvas (often the canvas would still appear in the project after being redirected there)
• A superfluous Databricks test that used to be performed on connection has been removed
• Cells that aren't attached to a valid database connection are now ignored on execution
• It's now possible to open search results in a new tab by holding Ctrl (Command on macOS) and clicking
• Fixed an issue where the AI input wasn't shown in some cases
• Project explore URL can now load in a new tab
• The correct domain whitelist is now used when SSO is enabled
• The correct dependency name is now used for catalog control cells
• We now initialise views correctly when the canvas only contains dropdowns
• Explore cells are no longer considered to be hidden in reports
• Undo/redo when the canvas is read-only
• We now incorporate upstream view signatures in remote cells so cached results are invalidated when the view definition changes

Count Metrics

Count Metrics is our vision of a semantic layer that delivers all the traditional benefits, while allowing us to consider the relationships between metrics, and make the canvas a true jumping off point for exploration.

Catalogs are defined in YAML files, version-controlled in Git, with the usual affordances of rolling back and atomic, documented commits. Count will run a pre-commit check on every dependant cell/visualization, and highlight breaking changes with associated errors, and subsequently show when these are resolved in further commits.

View and dataset have an expressive and powerful syntax allowing for fine-grain control over caching, formatting, and aggregations.

Every visualization or table built upon catalogs can be “Explored” with a single click. Bringing you into our familiar no-code Explore environment where you edit the visualization without impacting the original canvas, or save your changes into a new canvas.

More information is available in our launch announcement and product documentation.

Deep Workspace Search

Searching in Count now looks further than just the titles of canvases, and will query all text contents across SQL/Python code cells, comments, stickies, and text objects.

You can filter by author and project, and sort by created and last edited date. Clicking on a result will jump right to the specific object.

It is also now accessible anywhere with the shortcut Command/Control+K

Fixes

• We've reduced the number of calls made to the Slack API for workspaces with many channels
• Count can now import more metadata about the tables from very large databases
• Fixed a regression where the frame navigator wasn't shown in reports on mobile
• Various small fixes for embedded links in the canvas
• Count now always opens a new tab/window when you click on a link in the canvas
• Fixed an issue where wheel events were ignored while hovering on the onboarding popup
• When duplicating a canvas, all individual user permissions are now stripped other than those for the user performing the action
• When multiple tables/cells are joined in a low-code/visual cell, the correct columns are now shown in the right sidebar

Overviews

Overviews help you create canvases that are easier to understand and explore, by bringing text, lists, TODOs, and clickable links to frames or cells into the left-hand panel.

It's a small area where you can introduce users to the canvas, and help them get started using it. It's visible in reports too.

Now you'll never have to pan around a canvas again to find filter controls again... they can now be placed into the overview so they are usable wherever you are looking.

We've made it super simple to get started with existing canvases. Count AI can create intro text, and we can pull all those controls in with a single click. To learn more, check out the docs.

Preview canvases as another role

Analysts can now preview how canvases/reports appear to explorers, editors and viewers, to make sure their work appears as expected to others.

dbt jinja support for viewers

dbt jinja was previously limited to canvas editors; now all users see correct and consistent output from cells that use this.

Fixes

• Notifications are now debounced and batched into emails and Slack messages, to reduce noise for recipients
• The merge button is now hidden for users that aren't able to edit code in the parent canvas
• The buttons to refresh/cancel queries are now always shown
• An issue in the rendering of sliders has been fixed
• DuckDB cells now ignore the canvas manual execution setting
• Fixed an issue in the formatting of months of the year
• The navbar in embedded reports is now the correct height
• Hidden cells are no longer executed after the canvas is edited, unless required by any downstream cells

DuckDB on the server

Approximately 80% of queries in Count are (or could be) running in-browser using DuckDB. These "local" cells make canvases highly performant, and equally importantly drive down the cost on your data warehouse.

To tackle the remaining 20%, we're now introducing DuckDB on the server. This involves spinning up DuckDB VMs in our infrastructure to run queries that touch larger amounts of data (beyond the 128MB per cell available in-browser), before streaming results back to you.

Quicker, more explorable canvases, made cheaper. For more info, check out the blog post.

Concurrency

Query concurrency can now be configured for all database types from the connection settings.

Fixes

• Fixed an issue where images weren't displayed in Slack messages in iOS
• Click events on resolved threads are now ignored
• NaN is now accounted for in sorting tables/visuals
• Control cell states are no longer reverted when synchronising locked canvases
• The "Share to web" setting is now clearer in canvases/reports
• Hidden frames are now unhidden if navigating by link to an object within the frame
• Fixed an issue in rounding numbers before constructing big integers
• Fixed an issue in Safari not matching formatting
• An error message is shown if exploding cells fail
• QUALIFY is now correctly parsed for Redshift cells
• Data points with missing data are now excluded from default domain calculations
• Shared styles are now supported for toggles/sliders
• Fixed an issue of scrolling of cell inputs
• Fixed ignore/respect nulls parsing in window functions
• Bullet colors now change with that of the first character of the list
• Connectors are now hidden in reports if hidden in frame
• Links are now omitted from alert emails if the recipient doesn't have access to the canvas
• Fixed an issue in formatting of dates in legends
• Fixed an issue in updating sliders

Fixes

• Cell titles can now be copied/pasted along with other styles
• Improved error message when DuckDB references empty cells
• Locally-executed dropdowns are now sorted correctly
• Invoice links in the billing page are now clickable

Fixes

• dbt column descriptions are now shown in the visual controls
• Local control cells can now reference upstream cells attached to databases
• Fixed an issue with the display of rich Python embeds

Pending queries table

Workspace admins/owners can now see a list of all pending queries for a given database connection, at https://count.co/connection/<key>/queries. From here you can see which cells/canvases are contributing particularly to database load, and quickly cancel any number of queries.

Fixes

• The QUALIFY clause is now supported for Redshift queries
• Fixed an issue where IGNORE NULLS and RESPECT NULLS weren't parsed correctly in window functions
• Exponentiation in Athena is now supported in calculated fields
• "Data source order" has been renamed "Natural" to avoid confusion as to whether it relates to the order of the parent cell (it doesn't)
• Tooltips are now shown for truncated alert names in the subscriptions modal
• Fonts are now loaded immediately when added to a focused text object
• Copying/pasting styles between objects now correctly copies the font and text color
• Ligatures with multiple glyphs per character are now rendered correctly
• Inconsistencies of syntax highlighting between focused/unfocused states have been fixed

Improvements to control cells

• The settings for these now live in the sidebar, rather than in the canvas itself, so you don't need to keep hiding/showing the input to change them. Show/hide input is now only shown for SQL/Python cells
• You can now order single/multiple selects either alphabetically or by frequency (the default). Head to the sidebar to change this
• Filter cells have had a facelift to make the active filters clearer
• You can now customise the color of toggle controls and the color/font of sliders
• You can now change the numeric format in slider controls
• We've fixed an issue in which clicking on options in multiple selects sometimes didn't select them

Hold modifier key to select objects within others

You can now hold Cmd (macOS) or Ctrl (Windows) and drag to select objects that are within another.

Fixes

• Fixed an issue in scrolling cell inputs
• Fixed an edge case where text object changes weren't persisted
• Data points that can't be projected (null or negative for logarithmic scales) are now ignored in determining the default domain on the orthogonal axis
• Dataframe columns with array names are now handled correctly
• Fixed an issue with uploading of CSVs
• It's now possible to normalize big integers in visuals

Improvements to default domains in visuals

The default domains of Cartesian axes now extend to ensure that all the data can be seen. Previously, marks could be clipped if they extended beyond the point they described, particularly in the case of text.

Custom cron schedules for queries

It's now possible to use custom cron schedules in query schedules (for example, every weekday, or at 5 minutes past the hour), so long as successive invocations are at least an hour apart.

Edit and delete comments

You can now edit and delete individual comments in threads. To do so, click on the ellipses to the right of the comment.

Fixes

• Newlines are now captured in comments
• Fixed an issue when project member names didn't overflow correctly
• Errors in PDF export are now handled correctly
• For BigQuery, the date_add function in visuals/low-code cells now only casts to the datetime type if the interval is less than a day

Move and delete multiple canvases at once

It's now possible to select multiple canvases when viewing them as a list in the project view, by clicking on the checkboxes at the left of each row. These can then be moved between projects and deleted in bulk. Canvas previews are now also shown in this view so you can see the contents of each canvas without opening it.

Fixes

• Text within comments can now be selected properly
• Null groups are now accounted for correctly in visuals with normalization
• Text layout issues have been fixed in objects containing block quotes and code blocks
• Double-clicking on the bottom edge of a text object now consistently changes its sizing to auto-height
• An issue in which scaling some objects caused the canvas to crash has been fixed
• Positions of comments and stamps are no longer unnecessarily rounded when duplicating the objects they're attached to
• The disable cell execution button is now hidden in reports
• Fixed an issue where the Athena work group couldn't be specified when creating the connection
• Boolean columns are now supported in dropdowns
• Timestamp columns are now formatted correctly in dropdowns

Loading states

The loading state of cells are now much more informative, showing the time elapsed and the current stage of the query:

1. Connecting — The query is either waiting for a worker to pick up the job or — in the case of local cells — for upstream cells to complete executing. For databases with limited concurrency (such as Postgres and Redshift), the query may also be queued by Count
2. Running query — The database is executing the query or the results are being extracted from the database to Count's servers. In some cases (such as Snowflake), the query may also be queued by the database itself
3. Compressing results (optional) — Count compresses large result sets to make downloading them to your browser more efficient
4. Downloading results — The results are being downloaded from Count's servers to your browser

DuckDB v1.1.2

DuckDB has been upgraded from v0.9.1 to v1.1.2. To see what's new, check out the release notes for v1.0.0 and v1.1.0.

Big integer support in visuals

Big integers now enjoy first-class support in visuals. Previously, big integers were either converted to doubles if possible to do so without a loss of precision, or treated as strings otherwise. Now they are left unmodified and can be visualized as any other numeric type.

Metadata in SQL

SQL queries generated by Count are now suffixed with machine-readable metadata, to help database admins analyze those that originate from the platform.

Fixes

• Visuals containing multiple selects that differ only by whitespace no longer produce errors
• It's no longer possible to downgrade the permissions of a higher-ranked users in a canvas
• Visuals that reference columns measure_names, measure_values are now rendered correctly

Week start setting

Workspace admins can now change the day that weeks start on. This affects:

• The ordering of days in date pickers
• The ordering of days in visuals/low-code cells
• Truncation by week in visuals/low-code cells
• The behaviour of the visual/low-code functions: date_trunc('week', <date>), dayofweek and week

The default value is Sunday (unchanged from before). To change this head to the bottom of the workspace settings.

Fixes

• Text in comments can now be selected
• Embed blobs are copied properly on duplicating
• CSVs with more generic MIME types are now supported (including those created from Excel on Windows)
• Fixed an issue with the loading/error state of visuals with certain sorts applied
• The font weight dropdown now works properly
• The dbt Core loading state is now shown on refreshing
• Disabling public canvases in a workspace prevents all canvases from being accessed by non-authenticated users, even if they have previously been shared to the web
• Fixed rendering issues in Firefox@132

Rendering performance improvements

We've made various optimisations to accelerate the rendering of large canvases containing many cells.

Fixes

• 307 redirects are now followed when making HTTP requests from Python (fixes issues downloading Google Sheets)
• Fixes an infinite loop that certain canvases could enter when reconciling groups

Visual updates

A few of the most requested improvements to visuals are now live:

• There's now a funnel chart template and accompanying center stack option
• You can now drag colors in categorical palettes to reorder them
• You can now drag selects in visuals/low-code cells to reorder them
• You can now disable the rounding of domains in continuous axes
• Axes labels now wrap in visuals

Truncations in date pickers

Arbitrary truncations are now supported in custom date pickers (e.g., you can now select 3 months ago, and truncate the date to the start of the week).

Fixes

• Issues in displayed column formats for certain low-code cells/visuals have been fixed
• Fixes an issue where the lock canvas control was shown to viewers
• Canvases with manual cell execution enabled now behave properly on page load