sailfork / t3code-github / .repos/ alchemy-effect/ website/ src/ content/ docs/ axiom/ guides/ dashboards.mdx

An Axiom dashboard is a single JSON document: charts, a grid layout, a time window, a refresh interval. In alchemy that document is the input to one Dashboard resource, so a chart tweak is a reviewable diff and every stage gets the same dashboard.

This guide builds an ops dashboard chart by chart, then adds the two resources that round it out: saved Views and deploy Annotations. You'll need the Axiom provider registered (setup) and a dataset to query (ingest).

Start with an empty dashboard

import * as Axiom from "@alchemy.run/axiom";

yield* Axiom.Dashboard("ops", {
  dashboard: {
    name: "Ops Overview",
    owner: "",                 // org-shared — required for API tokens
    description: "Top-level service health",
    charts: [],
    layout: [],
    refreshTime: 60,           // seconds: 15 | 60 | 300
    schemaVersion: 2,
    timeWindowStart: "qr-now-1h",
    timeWindowEnd: "qr-now",
  },
});

The resource takes the full dashboard document at schemaVersion: 2. Axiom assigns the uid on create (exposed as an output along with id and the timestamps); updates overwrite the whole document in place, so the deployed dashboard always matches what's in your code.

Add a chart

import type { Chart, LayoutCell } from "@alchemy.run/axiom";

const errors: Chart = {
  id: "errors-5m",
  name: "5xx errors / 5m",
  type: "TimeSeries",
  query: {
    apl: `['my-app-traces']
      | where status >= 500
      | summarize count() by bin_auto(_time)`,
  },
};

yield* Axiom.Dashboard("errors", {
  dashboard: {
    name: "Errors",
    owner: "",
    refreshTime: 60,
    schemaVersion: 2,
    timeWindowStart: "qr-now-24h",
    timeWindowEnd: "qr-now",
    charts: [errors],
    layout: [{ i: errors.id, x: 0, y: 0, w: 12, h: 6 } satisfies LayoutCell],
  },
});

chart.id is a free-form string you pick — the layout cell's i joins to it. There's no dataset field: the dataset is implicit in the APL query (['my-app-traces']).

The chart union

const p99: Chart = {
  id: "p99",
  name: "p99 latency",
  type: "Statistic",
  query: {
    apl: `['my-app-traces'] | summarize percentile(duration_ms, 99)`,
  },
};

Chart is a discriminated union over the kinds Axiom's API validates: TimeSeries, Table, Pie, Statistic, Heatmap, and LogStream for data charts, plus Note and SmartFilter (below). Every data chart carries exactly the same payload — { id, name, type, query: { apl } } — and switching kinds is just changing type.

Add a note

const readme: Chart = {
  id: "readme",
  type: "Note",
  text: "## Ops overview\n\nEscalation: #ops-oncall. Runbooks linked per chart.",
};

Note is the one chart without a query or a name — its payload is strictly { id, type, text, variant? }, with text rendered as markdown in the cell.

Add a filter bar

const filters: Chart = {
  id: "env-filter",
  type: "SmartFilter",
  filters: [
    {
      id: "environment",
      type: "select",
      name: "Environment",
      selectType: "list",
      options: [
        { key: "production", value: "production", default: true },
        { key: "staging", value: "staging" },
      ],
    },
  ],
};

SmartFilter is the wire name for what the UI calls a filter bar. Each entry is either a free-text search filter or a dropdown select filter — dropdown options come from an inline list (as here) or, with selectType: "apl", from a query returning { key, value } rows. Other charts pick the value up by declaring a query parameter named after the filter's id: declare query_parameters (environment:string = "").

Lay out the grid

layout: [
  { i: "env-filter", x: 0, y: 0, w: 12, h: 2 },
  { i: "errors-5m",  x: 0, y: 2, w: 6,  h: 6 },
  { i: "p99",        x: 6, y: 2, w: 6,  h: 6 },
],

Each LayoutCell positions one chart by its i: x/y place it on the grid, w/h size it, and optional minW/minH/maxW/maxH and static constrain how it can be resized or moved in the UI. Every chart in charts needs a matching cell.

Compare against yesterday

yield* Axiom.Dashboard("compare", {
  dashboard: {
    name: "Compare vs yesterday",
    owner: "",
    refreshTime: 300,
    schemaVersion: 2,
    timeWindowStart: "qr-now-1h",
    timeWindowEnd: "qr-now",
    against: "-1d",            // overlay the same window from 24h ago
    charts: [],
    layout: [],
  },
});

against overlays each chart with the same query shifted by the given offset — the classic "is this spike new or does it happen every day at this hour" view.

Payload gotchas

Axiom's POST /v2/dashboards is strict, and the failure modes are worth knowing up front:

  • Relative time windows use the qr-now-{duration} form ("qr-now-7d", "qr-now") — plain "now-7d" is rejected.
  • When authenticating with an API token, dashboard.owner must be "". Axiom rewrites it to the org-shared X-AXIOM-EVERYONE; per-user "private" dashboards aren't allowed for tokens.
  • Data-chart payloads accept only id, name, type, query — extra keys like dataset or description fail with Unrecognized keys: "<name>".
  • refreshTime is in seconds: 15, 60, or 300.

Save a view

yield* Axiom.View("recent-errors", {
  name: "recent-errors",
  description: "Last 100 5xx responses",
  datasets: ["my-app-traces"],
  aplQuery: `
    ['my-app-traces']
    | where status >= 500
    | order by _time desc
    | take 100
  `,
});

A View is a named, shareable APL query — the investigation you keep re-typing, pinned as a resource. Its name is the identifier, so renaming a view is a replacement (delete old, create new).

Mark deploys with annotations

yield* Axiom.Annotation("deploy-1.2.3", {
  type: "deploy",
  title: "Release 1.2.3",
  datasets: ["my-app-traces", "my-app-logs"],
  time: new Date().toISOString(),
  url: "https://github.com/acme/app/releases/tag/v1.2.3",
});

An Annotation draws a vertical marker on every chart over the listed datasets — pass a single time for a point event or time + endTime for a range (an incident window), with type grouping markers visually. Deploy markers next to your error charts answer "did the release cause this" at a glance.

Where next

  • Alerting — turn the same APL queries into monitors and notifiers.
  • Ingest — datasets, tokens, and wiring OTEL output into Axiom.
  • Axiom overview — all resources at a glance.

Reference: