Get AX Code · FreeDocs (EN)
AX Code.
en
English
Get started

English documentation · runtime 7.24.4 · SDK 2.6.7. Content is maintained with runtime development; see each guide's scope and review date.

Run Report

Status: Active
Scope: current-state
Last reviewed: 2026-08-18
Owner: ax-code runtime

The run report is a local web page served by the built-in HTTP server: a workspace overview plus a per-session drill-down. Open it from the TUI sidebar (Analysis → run report, opens the current session’s report), from the command palette with session.report.web, or from the CLI with ax-code run-report [--index] [--open]. AX Code Desktop embeds the same pages.

Workspace Overview (/run-report)

The workspace overview for the current project, covering the last 30 days:

  • Usage — sessions, messages, total tokens, and cache share (cache-read tokens as a share of all input reads; a higher share means less re-reading).
  • Activity — tokens per day for the last 14 days; hover a bar for the exact day, token, and session counts.
  • Model usage — top models by tokens with message counts.
  • Tool usage — top tools by call count.
  • Session health — average risk, readiness breakdown, and validation pass rate across recent sessions.
  • Sessions — recent sessions with token, risk, and readiness chips; View → opens the per-session report.

Session Report (/run-report/session/<id>)

A per-session drill-down, ordered so the sections every session has data for come first:

  1. Usage — input/output/reasoning/cache tokens, cache share, duration, and per-model breakdown.
  2. Summary — what the agent decided and did, with step/tool/error counts.
  3. Changes — per-file change table with risk and diff stats.
  4. Timeline — execution Gantt chart (expand to render).
  5. Activity — slowest steps, tool timings, agents involved, and rollback points.
  6. Trust sections (Verdict, Risk, Validation, Branches) render only when the session has substance for them — a chat-only or trivially small session does not render an empty 0/100 gauge or scorecard. Validation appears only when validation commands were recorded; Branches only when the session has sibling branches to compare.

Data sources

All numbers come from the local SQLite store: sessions and message token counts, the per-session event log (steps, tools, durations, errors), and the risk assessment. No pricing data is bundled with model catalogs, so the dashboard reports tokens and cache efficiency rather than estimated cost. Pages live-update via server-sent events and reload on relevant session changes.

  • ax-code stats — terminal token usage report with --days, --models, and --tools filters.
  • Supported Providers and Models — connecting providers whose models appear in the usage breakdowns.