# Dashboards

> KPIs and charts over your own data, built once by AI as saved read-only queries that anyone can open and refresh for free.

Source: https://triagic.com/docs/desktop/dashboards

A dashboard is a page of KPIs and charts over your connected data: ticket volume by
channel, daily orders, refund rate, the slowest endpoints. An admin describes it in a
sentence, Triagic saves one read-only query per chart (a SQL query, an API call,
a search), and from then on the dashboard is those saved queries. Opening it, changing the date range and pressing
**Refresh** run the queries directly against your sources. No model is involved and
nothing is charged.

Dashboards live on the desktop app's **Dashboards** page.

## Building one [#building-one]

Only admins build dashboards. Members see *"Ask an admin to build a dashboard."*
in place of the button.

### Describe it [#describe-it]

Open **Dashboards → New dashboard** and say what you want to watch, for example
*"Support load by channel and status"*. Pick the data sources it may use under
**Use these data sources**. Every source that is running for you is offered, and
only sources you can open.

### Review the proposal [#review-the-proposal]

**Propose a layout** looks at what the chosen sources hold (tables, collections,
projects, a few sample records) and comes back with a list of KPIs and charts. Each item says what it shows, which source it
uses and what it costs to build. Untick anything you don't need, or use **Change
prompt** to start again. You can leave the page while it works; a notification
tells you when the proposal is ready.

### Build [#build]

**Build N items** picks one read-only query for each item you kept, runs it once to
check the result has the right shape, and saves the query together with its chart
settings. An item whose query can't be made to work is marked failed and left out;
the rest still make a dashboard.

"Built" means the dashboard is a list of saved queries plus how to draw each result.
Dates in a query are placeholders such as `{{range_start}}`, filled in from the date
range control, so switching between **7d**, **30d** and **90d** never needs the model.
A KPI can also run once for the previous period, to show the change.

## What it costs [#what-it-costs]

Building spends AI from the provider your desktop uses, and so do these admin
actions on a built dashboard:

* **Propose a layout**: one run that looks at your sources.
* **Build**: priced per item; the button shows the total for what you kept.
* **Add a chart or KPI**, **Change** and **Repair with Triagic**: one run each, for
  one chart.

Every one of those buttons shows its price before you press it. Each run also stops
at your organization's per-run cap ($1.00 unless an admin changed it), and keeps
what it finished. See [Spending and budget](/docs/admin/spending#per-run-cap-for-generation).

Opening a dashboard, changing the range, pressing **Refresh** and editing a query by
hand cost nothing, and never appear on the [Usage](/docs/desktop/history-and-usage#usage)
page.

## Who sees what [#who-sees-what]

* **Admins** build, add, change, repair, rename and delete.
* **Everyone** in the organization can open a dashboard, change its range, refresh
  it, and open **Query** to see the saved query behind a chart.
* On a cloud-managed install, dashboards are shared with the whole organization and
  a chart may only use an organization data source or Triagic tickets, so every
  member's desktop can run it. On an install that isn't centrally managed, any of your local sources
  works and the dashboard stays on that desktop.
* A chart over a source you can't open shows a locked tile. When a source is limited
  to certain teams, members outside those teams never receive that chart's query at
  all; they get its title and a locked tile.

Queries run from your own desktop over your own connections, so you only ever see
data you can already open. Results never leave your machine. They are cached on it,
per date range, for 15 minutes; **Refresh** runs the queries again straight away.

## Supported sources [#supported-sources]

A chart can use any data source that is running for you, through its read-only
tools: the same tools the agent may call during triage. Write tools are never
used, and a chart saved with one shows an error instead of running.

Charts work best with sources that return tables or lists: databases and
warehouses (Postgres, MySQL, Snowflake, ClickHouse, MongoDB), and APIs that list
records such as issues, errors or events. A chart reads its rows from the list in
the result, and Triagic tests each chart once while building it, so an item a
source can't answer is marked failed and left out.

**Triagic tickets** counts the tickets on your own desktop, by day, week, status,
source or playbook. Two people opening the same dashboard can see different
ticket numbers.

## When a query breaks [#when-a-query-breaks]

A chart whose query fails says *"This query stopped working"* with the error. Admins
get two ways to fix it:

* **Repair with Triagic** sends the broken query and the error to the AI, which
  writes and tests a new one. You see the new chart before you **Keep** or
  **Discard** it. This is a priced run.
* **See query**, then **Edit by hand**, lets you change the query text yourself.
  It's free. Triagic test-runs your edit before saving it and refuses it if the test
  run fails.

Members see *"Not available right now"* on a broken chart until an admin fixes it.
A source that is down shows *"Source unavailable"* with the reason, rather than an
error; nothing needs repairing.

## Limits [#limits]

* 12 charts and KPIs per dashboard.
* 3 series per chart.
* 500 rows per query.
* Dashboard names up to 80 characters.