Changelog
User-visible changes to Dashies, newest first, dated by the day each one shipped to production.
Every entry below is dated by the day the change merged and deployed. Dashies deploys on merge, so those are the same moment. That is deliberately not the same as the day the change became visible to you: some of this work lands behind the scenes first and is switched on later, so treat a date as when the code shipped, not as when the feature appeared.
Read this before you use the list as evidence
What is here. Only changes a user can see: a new warehouse engine, a new cadence, a new tool argument, a behaviour that used to work differently. Internal refactors, infrastructure work and test coverage are not listed. The list starts at 2026-07-11; nothing before that date is recorded.
How it is kept. By hand. Someone adds an entry when a user-visible change ships, and nothing checks that they did.
So an empty stretch means nobody wrote anything down, not that nothing shipped. Do not read a gap as a quiet period, and do not conclude a behaviour is unchanged because no entry mentions it. If you need to know how something behaves today, the page describing it is the answer; this list only tells you what someone thought to record.
Where the entries come from. The entries below were each read against the code, the schema or the design document that defines the behaviour, rather than against the summary attached to the change that shipped it. Those two disagree more often than you would expect, and two of the entries here were wrong on their first draft for exactly that reason. That is a statement about the entries that exist today, not a guarantee about any added later: nothing enforces it, for the same reason nothing enforces the rest of this page.
2026-08-03
The documentation moved to docs.dashies.xyz. This site replaced the single
/docs page on dashies.xyz, which listed five of the eighteen MCP tools and
nothing about warehouses, specs, dataset modes, schedules, versions or teams.
dashies.xyz/docs now redirects here permanently.
A filterable cube costs far fewer raw bytes. The data island can now store a dimension column as a dictionary of its distinct values plus one index per row, rather than repeating every label on every row, and the encoder keeps whichever form is smaller. Adding a filter dimension no longer multiplies the island the way it used to.
Three things this is not. It is not a transfer win: the edge already serves dashboards compressed, and compression collapses repeated strings on its own. It is not a cap raise: the publish and island byte limits are unchanged, and the win is that the bytes are genuinely smaller, which matters because those limits are enforced on raw bytes and browser parse cost is raw-byte bound too. And it does not migrate a dashboard you already published - a refresh keeps the island in the shape it already has, so an existing dashboard takes the new encoding when it is next published, not on its next run.
Nothing changes in how you author a cube. See The data island.
2026-08-01
Fixed: the Versions page showed no workspace dashboards. It listed personal dashboards only, and a workspace with versions to show rendered the "No dashboards yet" empty state instead. See Version history.
2026-07-31
Publish a spec by hash, or by edit. publish_dashboard accepts spec_hash
to republish a spec the server already holds, and spec_edits to change part of
one, instead of resending the whole document every time. See
publish_dashboard and
Author a dashboard.
Signing in returns you to the workspace you were last in, rather than always landing on your personal space.
Fixed: History and Activity showed the wrong space. Both feeds now scope to the workspace on screen instead of mixing in personal activity.
2026-07-29
Warehouse connections can belong to a workspace. A connection created inside a workspace is usable by that workspace's members and by its dashboards. The scope is fixed when the connection is created and cannot be changed afterwards, in either direction. Read Connections and scope before you bind one: a workspace dashboard bound to a personal connection publishes successfully and then can never refresh.
A data source now reports what refreshes actually observed, alongside the result of the last connection test. The two are separate facts and can disagree without either being wrong: a credential that tested fine in the morning and has been failing every refresh since is exactly the state this surfaces.
2026-07-26
Teams. A team is a group of people inside a workspace with a team screen: an ordered set of sections holding references to dashboards. A team screen points at dashboards, it does not contain them, so a dashboard can appear on any number of screens and removing a tile never deletes the dashboard. See Work as a team.
2026-07-18
Refresh every 15 or 30 minutes. Paid plans can set a sub-hourly cadence on an hourly dashboard. See Schedules for what the cadence guarantees and what it does not.
Dashboards store their spec. The YAML spec a dashboard was published from is kept server-side, so an AI tool can read it back with get_dashboard_spec rather than reconstructing it. For a dashboard published before specs existed, derive_dashboard_spec reconstructs a draft from the stored manifest and the published body; it stores nothing, and republishing that draft to the same slug is what converts the dashboard.
2026-07-17
SQL Server. The sixth warehouse engine. Read Connect SQL Server before you plan a cube against it: its per-query caps and the absence of a Parquet offload are stated there, and they are the two things that decide whether a dataset fits.
2026-07-15
Databricks. See Connect Databricks.
Amazon Redshift gained its connection form in the app, completing the engine. See Connect Redshift.
2026-07-11
Snowflake. See Connect Snowflake.