Skip to content

Explore

Explore is where you ask questions of your data without building a dashboard first. Ask in plain English; the answer comes back in words, and when a picture answers better it comes with a panel: a chart, a table, a stat or any other panel kind, drawn from rows the server just queried. Follow up and the next answer builds on the last one. Open it from Explore in the main navigation, or the command palette. A /chat link from before it was named Explore lands here.

Pick the sources the conversation may use from the chip beside the message box, or under Sources in the side panel: up to three, all in one workspace. Type a question and press Enter (Shift+Enter starts a new line). Stop, or Esc in the box, cancels an answer that is still being written. An empty conversation offers a few questions built from the first source’s catalog. The model chip beside the sources picks which model answers; the conversation keeps it for every later turn.

The answer is read-only. Explore can query the sources you picked and draw panels from them; it cannot change a source or a dashboard. Everything it runs goes through the same guard a dashboard panel does, over the conversation’s time range, and only over the rows you are allowed to see. Each source is queried in its own language: SQL for a TimescaleDB source, PromQL for a Prometheus one, LogQL for a Loki one, where a question about log lines is answered with a logs panel.

A panel in an answer is a real panel: the same spec a dashboard holds, run on the server. Under it:

  • Show as draws the same rows as another kind (a line, a bar, a table…) without asking the model again.
  • Re-run queries again now.
  • Add to dashboard puts the panel, as shown, at the bottom of a dashboard you can edit in the source’s workspace, or on a new one. It is the ordinary save: a new version, with every query checked again. Shown only when you can edit or create a dashboard there.
  • Copy spec copies the panel as JSON, for the editor or the Holotable skill.
  • Show query opens the statement, its source, and the window the server narrowed it to.
  • A long table gets a filter box; the panel’s own menu downloads CSV or PNG.

A query the model wrote in words (“Ran this query”) opens the same way.

The side panel holds Sources, History and Settings. Press [ to show or hide it, or use the button at its top (when it is hidden, the button stays in a narrow rail on the left). On a phone it opens from the button above the conversation, which names the sources and the range.

History lists your conversations under Today, Yesterday and Earlier, each named after its first question. Rename one, delete one, or Delete all conversations. The filter searches the titles that are loaded.

Settings:

Setting What it does Where it is kept
Time range The window every query in this conversation runs over. Changing it runs the panels on screen again The conversation
Live refresh Runs the panels on screen again every 30 seconds, minute or 5 minutes, skipping a hidden tab Your preferences
Show queries Opens every “Show query” and “Ran this query” by default Your preferences
Keep my conversations Keeps conversations on the server so they survive a reload and follow you to another device Your preferences

A conversation you keep is stored on the server, readable only by you; nobody else can open it, a platform admin included. It holds your questions, the answers and each panel’s spec, never its rows: opening a conversation runs its panels again. A conversation has its own address (/explore/<id>), so it can be bookmarked or opened in a new tab.

Conversations are kept for CHAT_HISTORY_RETENTION_DAYS (30 by default) after their last message, up to CHAT_HISTORY_MAX_MESSAGES messages each and CHAT_CONVERSATIONS_MAX conversations per person (the least recently used go first).

Turn Keep my conversations off and every kept conversation is deleted at once; from then on a conversation lives in the page and is gone when you leave. Its panels cannot be re-run, since nothing was stored to run.

If you lose access to a source, a conversation that used it stays in History; with none of its sources left it opens read-only.

Every dashboard has a chat. Ask it what a panel means, why a number moved, or what the data says about something the panels do not show directly, and it answers from the panel definitions first and from fresh data when it needs it. It is read-only: it cannot change the dashboard, and the only thing it can run is the same kind of guarded query a panel runs, against the same sources.

The widget’s Open in Explore button continues the same conversation on this page: the dashboard’s panels stay in context, its sources are the conversation’s, and the picks it was opened with show as read-only chips beside a From link back to the dashboard (change them there). A panel’s menu has Ask in Explore as well as Ask about this panel: it opens that conversation here with the question started about the panel. A dashboard’s conversation is listed in History like any other, and Clear chat in the widget deletes it.

Press C anywhere on the dashboard, or use the launcher in the header. Esc closes it and returns focus to where you were. The expand button docks the chat to the right edge at full height, and this browser remembers the choice.

An empty chat offers three or four suggested questions built from the panels on the dashboard. They are a starting point, and they come back after Clear chat.

A panel’s menu has Ask about this panel. It opens the chat with an “About” chip naming the panel, so your question is read in that panel’s context, and the suggestions are about that panel. Remove the chip to ask about the dashboard as a whole.

The chat sees the dashboard the way you do:

  • Your time window — the one you picked in the header, or the dashboard’s own range if you have not changed it. Shift the window and the next question is about the shifted window.
  • Your variable picks — the values you have selected. A value you are not allowed to use is replaced by the default, and the chat is told so.
  • The panel asked about, if any.

Every answer that fetched data says which range and which values the data was narrowed to, so an answer and the chart beside it cannot be about different things.

  • An answer that ran a query shows a “ran this query” footnote: the statement, the source, and which panels on the dashboard use that same query, with the range and values it was narrowed to.
  • Answers are formatted — lists, emphasis, code and small tables — and each has Copy. The latest also has Try again.
  • While it works, the status line says “Thinking…”, then “Querying data…”. Stop stops it: the model is told to stop rather than left running, and whatever was answered so far stays.

A dashboard’s conversation is yours: two people on the same dashboard have separate ones and cannot see each other’s. It is kept as described in What is kept, and Clear chat forgets only your own.

Anyone with a viewer role in a workspace can chat over that workspace’s sources, the same access the dashboard chat has. A row-filtered source answers with your rows only.