Skip to content

Administration

Overview

The platform includes an Admin area for inspecting the health and usage of a deployment. It is read-only: nothing in it changes projects, rulesets, or data.

This area is intended for Jaxon support staff diagnosing a deployment, not for day-to-day platform work. If you are looking for your own activity, the workspace Activity page is the place to go.

Access

The Admin area requires membership in the support_admin group and that group being your currently selected group in the group picker. Selecting a different group hides the area again, even for a user who belongs to support_admin.

Users without that access see no Admin entry in the workspace menu. Opening an admin URL directly shows an access-denied message rather than an error page.

Monitoring

Monitoring reports usage and spend across the deployment. Every view shares a time-range control (presets or a custom range) with an optional auto-refresh; the JVU and project views add a project filter.

  • JVU Usage — consumption measured in JVUs (see below), with totals for the selected period and all time, averages per run and per project, a consumption-over-time chart, and breakdowns by project, ruleset, user, and run.
  • LLM Usage — call volume, token counts, estimated spend, and latency percentiles across the deployment, broken down by model and by the platform feature that made the call, plus per-model error rates.
  • Project Usage — the same picture narrowed to selected projects: run compute time, stored data, and attributed LLM calls, including how storage has grown over time.
  • API — request volume, latency percentiles, error rates, payload volume, and the endpoints and users driving them.
  • Utilization — deployment-wide totals for compute time, storage, tokens, and API calls, with storage broken out by type.

Jaxon Verification Units (JVUs)

A JVU is the platform's unit of metered work. Each document-and-rule pair consumes:

ceil(document characters / 1000) x ceil(questions in the rule / 5)

with a minimum of 1 for each factor, so the smallest possible pair costs 1 JVU. Totals are recorded when a run completes, which is why JVU figures are attributed to a run's completion time rather than its start.

System Diagnostics

System Diagnostics reports the state of the deployment's services.

  • Health — one card per monitored component with its current status and how that status was determined (a direct probe, telemetry, or not monitored), and a link into the logs for that component.
  • Logs — a filterable, paginated view across log sources, with a live mode that polls for new entries.
  • Audit — a trail of the diagnostic actions taken here, with the actor and timestamp for each.

Telemetry-dependent views

Some figures come from the platform's database and some from its telemetry pipeline (Elasticsearch). On a deployment without telemetry, the database-backed views work normally while the telemetry-backed ones show an empty range instead of an error:

Source Views
Database — always available JVU Usage; compute time, storage, and LLM attribution in Project Usage and Utilization; Health direct probes; the Audit trail
Telemetry — needs Elasticsearch The API view; LLM Usage; the token and API-call totals in Project Usage and Utilization; Diagnostics logs and telemetry-derived health status

A banner appears on the Health view when telemetry is degraded or unavailable, so an empty chart can be told apart from a genuine absence of activity.