Changelog

Sort and arrange tables without changing the chart

When you sort a table chart, hide a column, or reorder columns, those changes now stay with you. They’re saved to your own view without creating a new version of the chart or affecting what anyone else sees. Under the hood, sorting is applied as a safe, validated query change rather than rewriting the chart’s saved SQL, so the chart stays exactly as its author built it.

This also means you can now sort and rearrange columns on read-only and public table charts, which previously wasn’t possible. And when a chart’s SQL changes in a way that renames or removes a column, your saved sorts and column tweaks reset cleanly instead of getting stuck on data that no longer exists.

Search your chats from the sidebar

The chat list now has a filter bar at the top, so you can type to instantly narrow your history down to the conversation you’re looking for by name. Clear it with the button or the Escape key to jump back to the full list.

We also virtualized the chat sidebar, so it stays fast and smooth to scroll even when you have thousands of chats.

A faster, more predictable public API

The public API now allows 60 requests per minute by default—much higher sustained throughput than before—and we fixed a subtle issue where steady, well-behaved clients could get locked out until they went completely idle for a full hour. It’s now enforced race-safely under concurrent requests, too.

Every response also carries standard X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset headers, so your integrations and agents can see exactly how much headroom they have and pace themselves automatically.

Fixes and improvements

• Fixed very large tables and queries that could spin forever instead of loading a preview.
• Fixed a chat crash when querying PostHog data with certain ClickHouse column types.
• Fixed exporting a chart that references a definition, which previously failed with a “database object not found” error.
• Replaced raw HTML error pages with a clear message when a chart query fails unexpectedly.
• Fixed the “Updating data from X ago” label so it only appears during an active refresh and clears right after a manual refresh.
• Fixed cases where AI-created charts and tables could go missing from a chat’s final answer.
• Improved chat reliability under heavy load by eliminating intermittent message-write failures.
• Made automations with large run histories load and update much faster.
• Fixed AI requests that could fail from over-sending context on smaller GPT-5.4 models.
• Upgraded the default fast AI model to GPT-5.6 Luna for quicker, lower-cost lightweight AI tasks.

MotherDuck is now a supported data source

You can now connect MotherDuck as a first-class data source and query it right alongside your other databases. Connect through your MotherDuck Postgres endpoint with an access token, point Basedash at the database you want to sync, and you’re ready to build charts and dashboards and run AI chats on top of it.

Because Basedash understands MotherDuck natively, the AI writes proper DuckDB SQL—including support for nested types like lists, structs, and maps—and the schema explorer and autocomplete know your tables from the moment you connect. It’s a straightforward way to bring serverless DuckDB analytics into the same workspace as the rest of your data.

Reference definitions like tables in SQL

Definitions—your saved, reusable queries—now behave like real tables. Instead of the old {{ definition("name") }} template syntax, you reference a definition with a simple schema-qualified name: select * from definitions.your_definition. It reads like normal SQL and composes naturally into joins, CTEs, and subqueries.

Definitions also show up as virtual tables throughout Basedash now. They appear in SQL editor autocomplete with their columns, and the AI sees them in its schema catalog, so it can discover and build on your definitions without any extra prompting. Existing definitions were migrated automatically, so nothing breaks.

Auto-refreshing dashboards keep up under heavy load

Dashboards that refresh on an interval—especially the always-on ones running on office TVs—now update far more reliably. Refresh work is fairly distributed so a single busy workspace can’t monopolize the queue and delay everyone else, and throughput is much higher, so charts update within seconds of their interval instead of lagging behind. When several screens show the same dashboard, they now share a single refresh per chart instead of each running its own live query, dramatically cutting the load Basedash puts on your database.

We also made individual queries more resilient. The query timeout is now 120 seconds, up from 60, to give heavier charts room to finish, and when a query does time out, Basedash now cancels it at the source across Postgres, Snowflake, Oracle, DuckDB, and ClickHouse instead of leaving it running against your warehouse.

Fixes and improvements

• Improved voice-to-text in chat by moving to a newer, more accurate transcription model.
• Fixed scheduled chart subscriptions failing to render for charts that rely on date ranges or default variables.
• Fixed chat image attachments that wouldn’t load when the file’s name contained special characters.
• Fixed the Slack channel picker returning an error when your Slack connection needed reauthorizing—it now prompts you to reconnect.
• Fixed the upgrade flow for expired trials, which could appear to do nothing, so it now shows clear loading and retry states.
• Fixed automation run history freezing or crashing when a report contained very large query results.
• Fixed ClickHouse queries that could fail on read-only connections, which now retry automatically.
• Renamed the command menu’s “New draft chart” to “New chart” and made it easier to find when searching for “SQL query.”
• Added a public API endpoint to delete an organization.

Chat has a fresh new look

Chat got a visual refresh. Your own messages now appear in clean blue bubbles that shrink to fit exactly what you typed, so conversations are easier to scan at a glance.

Following along with the AI is clearer too: every tool call now shows a plain-language description of what it’s doing—like searching your schema or running a query—instead of raw JSON. You can see each step the AI takes without decoding technical output.

Sort tables right inside chat

Tables in AI responses have been redesigned with a cleaner, card-style layout, right-aligned numbers, and a tinted header row that’s much easier to read.

Best of all, you can now sort any column with a single click. Numbers, dates, and text all sort intelligently, so you can reorder a result set to surface the biggest values or the latest records without leaving the conversation or writing more SQL.

AI database edits are validated before you approve them

When you ask Basedash to change data in your database, it now checks the query before showing you an approval prompt. Invalid or broken mutations fail fast and the AI corrects them on its own, so the change you’re asked to approve is one that’s actually ready to run.

That means less time reviewing writes that would have errored out anyway, and more confidence that an approved change will do what you expect. Edits are handled one at a time, keeping each approval clear and easy to reason about.

Fixes and improvements

• Made connecting Fivetran data sources smoother by fixing a confusing “already connected” state and error toast that could appear while a connection was still being set up, plus a stray “allow popups” warning on the desktop app.
• Fixed chats showing a stale approval request when several actions needed approval at once, so you always see the current one.
• Fixed chat contents occasionally appearing out of order, and stabilized the ordering of automations, MCP tools, and hidden schema lists.
• Scoped AI variable and filter updates to the dashboard you’re working on, preventing changes from leaking to other dashboards, including duplicated ones.
• Gave the AI richer context about your data—row counts, sample values, and null rates—so it writes more accurate queries and charts.
• Improved chat answer quality on tougher analysis with an updated prompt and default model, and let the AI consult a second model on genuinely hard questions.
• Clarified how the AI formats percentages in charts so they render on the correct scale.

Chat, dashboard, and automation suggestions

Basedash now gives every workspace personalized suggestions across chat, dashboards, and automations, so you have useful starting points before you type a prompt or build from scratch. Suggestions are generated from your connected data, past chats, dashboards, and workspace context, then appear where you’re already working: questions under the chat input, dashboard ideas with build plans, and recurring report ideas for automations.

Suggestions are rolling out to all workspaces today. Read the full launch post, Introducing Basedash Suggestions, for the deeper walkthrough of how they work and how to get started.

Drive Basedash chats from the public API

You can now start and continue AI chats programmatically through the public API. Send a question, wait for the answer synchronously or poll for it, stream responses as they generate, list past turns, and cancel or delete chats—all with the same API-key auth as the rest of the platform. Chats created through the API stay out of the app UI, so automated workflows don’t clutter your team’s chat history.

We also added a general chart image endpoint, so you can fetch a rendered PNG for any chart you have access to—whether it lives on a dashboard, in a chat, or in an insight—always rendered with your own data permissions. Chart responses now include a ready-to-use image URL, making it simple to drop live Basedash visuals into your own docs, apps, and reports.

Dashboards load instantly, even after a few days

Opening a dashboard no longer means waiting on fresh warehouse queries for every chart. Basedash now recognizes when a chart is really asking for the same data as before—even when relative date ranges like “last 7 days” shift to a new calendar day—and serves the previous result instantly while quietly refreshing in the background. Charts that felt slow to reopen after a weekend now snap into view.

Query caching is also on by default now, with a new admin toggle in Settings → General so you stay in control. Filter changes, date-range edits, and pagination still fetch fresh data, so you always see exactly what you selected.

Invite and manage members more easily

Inviting teammates is now a single dialog you can open from the sidebar, command menu, members settings, or right in the middle of connecting a data source. Pick a role and, optionally, drop new members straight into groups as you invite them—and re-inviting someone you previously removed restores their account cleanly. When invites go out to several people at once, each one is handled independently so a single bad address doesn’t block the rest.

The members list is clearer too: each member shows how many groups they belong to, and admins can adjust group membership right from a member’s menu. We also reworked the language from “deactivate/reactivate” to the more intuitive “remove/restore.”

Fixes and improvements

• Kept long chats going instead of hitting a dead end—when a conversation grows too large, Basedash now automatically compacts older context and continues, rather than asking you to start a new chat.
• Improved how the AI explores large database schemas with new catalog and table-description tools, so it finds the right tables faster and spends less time introspecting.
• Fixed opening tables in the Data tab for PostHog connections, which previously failed with a raw error, and made PostHog error messages readable.
• Fixed chat messages occasionally rendering out of order, so an assistant reply no longer appears above the question that prompted it.
• Fixed follow-up messages appearing to go nowhere when you open a chat directly from a link, so live status now updates correctly for deep-linked chats.
• Fixed the tool-approval card staying on screen after you approve an action, so chat resumes without a manual refresh.
• Fixed a jump in the chat view when the assistant’s final answer starts streaming, keeping the message height stable.
• Fixed exported chart images (PNG and PDF) and Slack chart images so fonts load correctly and tightly measured numbers are no longer clipped.
• Improved the reliability of AI chart edits when the model leaves display settings unchanged.

Manage dashboards and charts with the public API

The public API now covers dashboards and charts. You can list, create, update, and archive dashboards (including their tabs), list the charts on a dashboard, and create, update, and archive charts—all through /api/public, following the same auth and pagination conventions as the rest of the API. The full spec lives at /api/public/openapi.

Chart content like the name, SQL query, chart type, and axes is flattened right onto the chart resource, so it’s simple to read and edit. Behind the scenes, content edits create a new chart version, so your history stays intact while the API stays easy to work with. We also expanded the organization endpoints (Insights AI context and billing email settings) and added controls for setting MCP tool approval modes.

A dedicated home for connecting data sources

Connecting a data source has moved out of the command menu into full pages in the Data section. There’s now a searchable directory of everything you can connect—databases, warehouses, Fivetran connectors, MCP servers, and demo data—organized into categories so it’s easy to find what you need.

Each option gets its own connect page with a logo, description, and docs link alongside an inline connection form, whether that’s database credentials, a Fivetran connect card, an Airbyte spec form, or an MCP server URL. Once a connection succeeds, its data page opens right away. This replaces the older experimental Integrations pages.

View and edit access for data sources

Data source access is now split into view and edit levels, so you can give people exactly the access they need instead of an all-or-nothing setting. View access lets teammates query and explore a source, while edit access is required for AI-run write queries—which still require explicit approval before anything runs.

You manage these levels right from the data source’s Data Details page, which is now the single source of truth for access. Existing access defaults to view, keeping edit only for sources that already had AI database edits enabled.

Chat automatically uses all your data

The chat input no longer asks you to pick a data source for each message. Instead, Basedash automatically queries across every source you have access to and quietly skips any that are offline, so you can just ask your question and get an answer. You can even start a chat when your only connection is an MCP server.

The input itself is cleaner as a result: the model picker and a plus menu (for AI context and file uploads) now live on the action bar, with less chrome getting in the way of your question.

Fixes and improvements

• Improved how AI writes drill-down queries, so clicking into a chart element reliably shows the underlying records once the chart is on a dashboard.
• Enabled the AI to color single-series charts (a line or bar without a breakdown), matching what you could already do by hand.
• Simplified voice recording in chat down to a single primary action button, with a dedicated cancel control that appears while you’re recording.
• Improved sidebar and chat panel behavior with a smoother open and close animation, Settings as a top-level nav item, and log out and settings available from the org switcher.
• Kept the dashboard AI input focused when you click a chart to add it as context, so you can keep typing without losing your place.
• Kept dashboard and automation prompt suggestions visible while they refresh in the background, and extended how long they stay fresh.
• Fixed a custom date range filter drifting one day backward after a reload in timezones behind UTC.
• Fixed newly connected data sources not appearing on the Data page until a manual refresh.
• Added opt-in invoice emails so you can receive each paid invoice by email, with an editable billing email in settings.
• Added a sidebar notice for admins when a subscription is past due, linking straight to billing.
• Kept the required confirmation text visible while typing in the data source delete flow, so you always know exactly what to enter.
• Improved the reliability of AI chart creation and editing when the model sends minimal settings.