Your first dashboard

This walkthrough starts where the quick start stops: the demo stack is running at http://localhost:8080, you set SLASHOBSERVE_LLM_PROVIDER and SLASHOBSERVE_LLM_API_KEY before make demo, and you are signed in as admin@example.com. Without a model the chat shows “No model is connected to this server.” and prompts get 503 agent_unavailable; see LLM providers.

Ask for a dashboard

Open Observe in the sidebar. The page asks “What do you want to see?” and has a box labelled “Message the agent”. Type a prompt and press Enter, or choose Send. Shift+Enter starts a new line, and pressing / anywhere on the page moves focus to the box.

checkout health by pod, last 6h

Under the box, a chip lists each healthy datasource the agent may use, followed by the name of the model. In the demo the datasources are prom (Prometheus) and ch (ClickHouse). The buttons under “Example prompts” send a ready-made prompt at once, such as “Service health by pod, last 6h” or “Error logs by service from ClickHouse”.

Starting a dashboard from the chat needs Edit on a folder. Without it the chat home shows “Starting a dashboard needs Edit on a folder. Ask an Admin if you need it.” See Roles.

A prompt is 1 to 4000 characters.

Watch the agent work

The page switches to a conversation that shows your prompt and, below it, “Building the dashboard” with a progress list. Each line is one step the server recorded, grouped by phase:

Repeated steps of one phase fold into a single line such as “3 times, latest: Listed 12 metrics on prom”. A step that failed gets a warning icon and is read out as “needed a fix”. The agent only works with names it discovered, so a wrong metric or label usually shows up here as a failed check followed by a repair.

A save is refused when a panel’s query is invalid or returns no data. The agent then repairs the draft and saves again, a few repair rounds. If it still cannot save, the conversation ends with “The dashboard could not be built” and a Try again button.

When the dashboard is saved, the progress list collapses into a line such as “Ran 14 steps, repaired 1 query”, which you can open to see every step. Below it a result card shows the dashboard title, how many panels it has and which datasources they use, and the footer “Saved as version 1”.

Open the dashboard

Choose Open dashboard on the result card. The dashboard opens in the same tab with the version button reading “Live · v1”.

A new dashboard goes into the folder you last created a dashboard in. If you have not created one, or can no longer create there, it goes to General, and if you cannot create in General either, to the first folder where you can. In the Dashboards page’s Made by column it shows your name, then your email and “Agent”. See Folders.

Choose Activity at the top right to open the activity card. It lists the saved version, with the agent’s one-sentence summary, next to any comments and requests.

Ask for a change

Go back to the conversation, or use the activity card, and ask for another change. In the conversation the box now reads “Ask for another change to <title>”, with the note “Each change is saved as a new version of the dashboard.”

add a panel with memory usage by pod, and show the last 12h

The agent revises the latest version of the same dashboard instead of starting a new one. It keeps every panel you did not ask it to change. When the run finishes the result card says “Saved as version 2”.

If you have the dashboard open, a toast says “Updated to version 2” and the page switches to it. Version 1 is kept. You can compare the two versions or restore the older one; see Comments, agent requests and versions.

Change the time range and variables

The button at the top of the dashboard shows the current range, for example “Last 6 hours”. Open it to pick a quick range such as “Last 1 hour” or “Yesterday”, or type values into From and To. Both accept now-6h, now/d and 2026-10-17 09:00, and Apply sets them. On a time series panel you can also drag across the chart to zoom to that range.

The Auto-refresh menu next to Refresh now reloads the panels on an interval. A dashboard can set its own refresh choices and quick ranges, start on a refresh interval, or hide the time picker except for Refresh now.

If the dashboard has variables, a bar of selectors appears under the top row: a menu for a query or custom variable, with an All choice when the dashboard allows it, a text box for a textbox variable, and so on. The agent adds variables only when the prompt asks for a breakdown people will filter, so a dashboard from the prompt above may have none.

The range, refresh interval and variable selections live in the page address (from, to, refresh and var-<name>), so you can share a link that carries them, and nothing is saved to the dashboard. The same holds for Chart hover in the More menu, which Cmd/Ctrl+O cycles through: “This chart only”, “Shared crosshair” and “Shared tooltip” choose how hovering one time chart shows on the others. A choice that differs from the dashboard’s saved cursorSync goes into the address as cursor. To change the saved default range for everyone, ask the agent in a follow-up, such as “make the default range the last 24 hours”.

Next steps

• Chat and follow-ups explains what the agent does at each step and what it cannot do.
• Comments, agent requests and versions covers review, comparing versions and restoring.
• Edit a panel covers changing a query or a panel by hand.
• MCP and CLI build and edit dashboards from your own tools.
• Import Grafana dashboards brings in dashboards you already have.