Skip to main content

Documentation Index

Fetch the complete documentation index at: https://datost.com/docs/llms.txt

Use this file to discover all available pages before exploring further.

A Tracked Metric is a query that Datost reruns on a schedule, storing each result as a point on a time-series. Add a Monitor and Datost will watch that series around the clock, ping Slack the moment something looks off, and kick off an AI investigation before you finish reading the alert.
Tracked Metrics turn a one-off answer into a standing dashboard. Monitors turn that dashboard into a paging system with a data analyst attached.

​
What you get

  • Any number your warehouse can produce, watched on the cadence you choose.
  • Slack alerts with context: current value, comparison window, and the chart.
  • Automatic investigations that hypothesize why the metric moved and post findings back in thread.
  • A running timeline of events, notes, and snapshots for every metric you track.

​
Create a tracked metric

1

Start from a chart or a question

Ask Datost a question in Slack or open any saved chart in the web app. When the answer is a single number or a simple time-series, you will see a Track this metric action.
2

Name it and pick a cadence

Give the metric a name your team will recognize (Signups (daily), MRR, P95 checkout latency). Choose a refresh interval: hourly, every few hours, daily, or weekly. Datost will rerun the underlying query on that schedule and append a snapshot each time.
3

Confirm

Datost backfills recent history where it can and starts capturing forward. The metric shows up on your Tracked Metrics page with a live sparkline.
Break a metric out by a dimension (plan, region, customer) and Datost will track each series under the same metric, so one monitor can watch them all.

​
Add a monitor

A Monitor is an alert condition layered on a tracked metric. Pick the shape that matches what “bad” looks like:
Fire when the value crosses a hard threshold you set, for example “alert if daily signups fall below 50” or “alert if error rate exceeds 2%”.
Fire when the value moves by more than a set percentage versus a comparison window (default: 24 hours). Good for “wake me if revenue drops 15% day-over-day”.
Fire when the value retraces a set percent from its recent peak (or rebounds from its recent trough). Useful for metrics that climb over time, where the absolute number keeps changing.
Each monitor has a cooldown so you do not get paged every refresh, a Slack channel to notify, and an optional investigation prompt that steers the AI.

​
Monitor states and what happens when one fires

A monitor sits in one of three states: OK (all series within bounds), triggered (at least one series has crossed the condition), or recovered (the series came back inside the bounds). Transitions are what generate alerts.
1

Slack alert

The moment a monitor triggers, Datost posts to the channel you picked: metric name, current value, the threshold it broke, and a chart with the breach highlighted.
2

Auto-investigation kicks off

Datost immediately starts a metric investigation in-thread. The AI pulls recent snapshots, slices by the dimensions on the metric, queries related tables, and forms a hypothesis about the root cause.
3

Findings posted in-thread

When the investigation completes, Datost replies with a summary, a likely root cause, affected entities (plans, regions, customers), the data points it found, and a recommended next step.
4

Recovery

When the series moves back inside the monitor’s bounds, Datost posts a recovery message and the monitor returns to OK.

​
The metric timeline

Every tracked metric has a timeline that captures its life: when it was proposed, when it activated, each monitor trigger, completed investigations, and any notes your team added from Slack or the web. Notes are first-class, so whenever someone explains a spike (“marketing launch on Tuesday”) that context is pinned to the event and shown the next time the metric moves.
Investigations are only as smart as the data Datost can see. If a monitor fires on a metric that lives in a warehouse Datost is not connected to, the AI will note that it could not investigate. Connect the relevant source and Datost will retry.
Pause a monitor during a planned migration, archive metrics you no longer care about, or tweak thresholds as your business changes. Everything is editable from the metric’s detail page.