ai-hist stats is your dashboard for the corpus you have built — and the entry point to the Burn lens. At its simplest it tells you how much history you have and where it came from. As your record grows, it becomes the place to see where token spend actually goes and which sessions are burning money for nothing.
Overview
Run stats for a snapshot of the whole database:
ai-hist statsExample output:
Total entries: 47,665
By source:
claude: 37,406
codex: 10,259
Date range:
2025-10-05 to 2026-03-08
Top 10 projects:
8,701 /Users/you/Projects/my-app
4,586 /Users/you/Projects/api-server
...
You get total entries, a breakdown by source, the date range covered, and your busiest projects — a fast read on where your AI work is concentrated.
The Burn lens
Recall tells you how your team prompts. Burn tells you what it costs.
Local ai-hist stats reports entry counts, sources, date range, and top projects. Cost and token attribution — the Burn lens below — is built on the per-event token and cost fields (inputTokens, outputTokens, costUsdMicros, model, provider) captured in Relayloop Cloud, not in the local overview.
Once your history is in the cloud, the same record that makes sessions searchable also makes spend attributable: each event carries its source (tool), project, model, and — across a team — the engineer behind it.
That attribution turns an opaque monthly token bill into answers:
- By model and tool — which agents and models account for the spend.
- By project — which codebases are the expensive ones to work in.
- By engineer — where leverage is high and where sessions churn.
Finding waste and runaway loops
The patterns Burn surfaces are the ones worth acting on:
- Repeated dead ends — the same prompt retried over and over in a session is a signal that an approach is not landing. Drill in with
ai-hist context <id>to see the loop. - Runaway loops — a session that keeps regenerating without converging burns tokens fast. Catching these early, before they bill you twice, is direct savings.
- Expensive habits — a project that consistently costs more per task is a candidate for better repo rules, which the Insights lens turns into proactive guidance.
The goal of Burn is lower, more predictable token spend — not by rationing, but by seeing the waste clearly enough to cut it.