Usage & dashboards

Observe answers “what happened.” Usage answers “how is my product being used,” sliced by the dimensions that matter to your business — see custom dimensions for how attributes become dimensions.

Usage explorer

Under Observe → Usage, pick a dimension and see it three ways at once:

• A breakdown ranking its values by traffic, cost and errors.
• The trend over time, and the latency percentiles, for the current slice.

The dimension picker lists the built-in dimensions (model, provider, status, tool, agent, app, user) alongside the custom ones your app has emitted. Select a value on a custom dimension to scope the trend and latency to just that slice — for example participant_role = candidate.

Dashboards

A dashboard is a saved JSON spec of widgets, and each widget binds directly to the analytics API:

• kpi / timeseries — a metric (calls, errors, cost, tokens, latency, ttft)
• breakdown — a dimension (including tag with a tagKey)
• latency — percentiles for the scope
• every widget takes an optional scope (source, model, provider, app, and filterTagKey/filterTagValue)

Because a dashboard is data rather than code, one generic renderer draws any spec. Manage them under Observe → Dashboards, or over the API: GET/POST/PUT/DELETE /dashboards.

Agent-authored dashboards

The same properties make dashboards ideal for a coding agent to author. An agent with the axonpush MCP attached can read your backend, discover which dimensions exist (analytics_dimensions), stamp any high-value ones that are missing, and save a dashboard tailored to your domain — all as tool calls, no bespoke UI code.

If your agent supports skills, run /axonpush-tailor-dashboard (from npx skills add axonpush/skills): it does exactly this end to end.

Related

• Custom dimensions, the model this is built on
• Observe, traces, logs and the core analytics
• Analytics API · Dashboards API