Skip to content

AI-powered virtual executive team — a single coherent executive persona backed by 8 specialist agents (FastAPI + Next.js).

2.4 KB
Raw Blame
{
  "section_id": "token_usage",
  "title": "Token Usage",
  "markdown": "**What it is:** What the AI has cost across every conversation, added up from the audit log: tokens, model calls, web searches and dollars.\n\n**What it does:**\n\n- Shows totals for all time or for a date range.\n- Breaks usage down by source, by model and by day.\n- Splits input into fresh input, cache reads and cache writes, so you can see how much the cache saves.\n- Shows the work the Executive did in one go with a tool it built, and the jobs it ran on files, when there are any.\n\n**How to use:**\n\n1. Open **Settings → Advanced → Token usage**, or click **Token usage** at the top of the Audit log.\n2. Set **From / Until** to look at a date range, or leave them blank for all time. **Clear filters** resets them.\n3. Read the cards at the top: **Cost (USD)**, **Calls**, **Cache hit** (the share of prompt input served from cache), **Output tokens**, **Fresh input**, **Cache read**, **Cache write**, and **Searches** (web searches the model ran).\n4. **Done in one go** appears when the Executive built a tool to repeat work over many items. It shows **Jobs done in one go**, **Actions taken**, **Model turns avoided** and **Time taken**.\n5. **Work on files** appears when the Executive ran Python jobs on files, such as splitting PDFs or building spreadsheets and charts. It shows **File jobs**, **Charged for those turns**, **CPU time** and **Largest job** (peak memory).\n6. Use the **By source**, **By model** and **By day (UTC)** tables to find what drives spend. Each row shows Calls, Input, Cache read, Cache write, Output, Searches, Cached and Cost. **By source** names the part of the system that made the calls, such as chat turns, research specialists, triage or memory extraction.\n7. For one conversation's tokens, open the Audit log with **← Audit log** and open that session's flow chart.\n\n**Good to know:**\n\n- Days are in UTC.\n- Cost is the charge recorded on each call. Calls logged before cost tracking started show tokens with a cost of $0.\n- The owner of the Executive sees everything. Other people see totals that leave out the owner's private activity.\n\n**Example:** spend jumped last week? Set **From / Until** to that week, check **By day (UTC)** for the spike, then **By source** and **By model** to see what drove it.",
  "mermaid": null,
  "generated_at": "2026-10-15T21:00:00Z"
}