PROMPT LEDGERLocal agent observability
AGGREGATED · NO PROMPTS

30-DAY OPERATING REPORT / 01

What did the
agents consume?

A local snapshot of Claude Code and Codex usage—translated from billions of tokens into rhythm, concentration and cost.

API-EQUIVALENT COST $12,831.86
↑Peak: $1,942.73 on Aug 09
TOTAL TOKEN TRAFFIC 14.06B
OUTPUT TOKENS 71.01M
5.0‰ of traffic
ACTIVE DAYS 25/30
83% operating cadence
DAILY ACTIVITY

Usage velocity

JUL 11JUL 20JUL 30AUG 09
09 AUG Highest-output day also carried the highest estimate: 13.50M output tokens and $1.94K.
CADENCE

Operating calendar

UTC+03
MTWTFSS
QUIETINTENSE
5

rest days created two clean breaks in an otherwise dense run.

TOKEN ECONOMICS / 02

Cache did the
heavy lifting.

Almost all token traffic reused existing context. The tiny visible slices are where new context and model output entered the system.

TOTAL TRAFFIC14,063.70M

Cache reads dominate. 95 of every 100 tokens in this traffic view are repeated context—not newly supplied text.

MODEL MIX / 03

Cost followed
the frontier.

Two Opus generations account for nearly three quarters of the API-equivalent estimate.

01
Claude Opus 551.52% of estimated cost
$6,610.44
02
Claude Opus 4.822.75% of estimated cost
$2,919.22
03
Claude Fable 516.95% of estimated cost
$2,174.80
04
Claude Sonnet 57.46% of estimated cost
$957.61
05
GPT-5.6 Sol0.93% of estimated cost
$118.90
06
Terra + Haiku0.40% combined
$50.90