Prompt history: every run records its transcript (task, each model step with
tokens and cost, tool calls and results, engine follow-ups) - bounded in size,
scrubbed of recognisable secrets, system prompts stored once per text.
Retention in days per instance (default 14) and per agent (0 = never, -1 =
forever); an hourly janitor deletes what is past it, "never" also purges.
Usage analysis over run receipts: filter by period, agent, model, event, status,
cost and free text; sort by every column; totals over the whole filter; hourly
or daily series grouped by agent, model or event; markers for budget denials and
model changes; run detail with receipt, transcript, tool calls and approvals.
WebUI: Analysis > Token usage > Runs / Timeline, run detail, Prompt History with
retention settings. New code, comments and texts in English; texts come from
language packs (i18n/en.json, de.json) via modules/i18n.js. CLAUDE.md records
English as the main language.
Version 0.1.12.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
