model.ledger

LLM spend accounting that engineering and finance can both read. A price book, per-team and per-model rollups, budget breach detection, daily series, and spike detection. Standard library only.

AI cost problems rarely announce themselves. One team ships a prompt that doubles its context, and the bill moves a month later. A ledger that attributes every call to a team and a model makes that visible the day it happens.

Use

from model_ledger import Ledger, Price, Usage

ledger = Ledger(
    prices={"small": Price(0.2, 0.6), "large": Price(3.0, 15.0)},
    budgets={"search": 50.0},
)
ledger.record(Usage("search", "large", input_tokens=2000, output_tokens=1000, day="2026-03-01"))

ledger.by_team()       # {'search': 21.0}
ledger.over_budget()   # {} until search passes 50.0
ledger.spikes()        # days far above the prior trend

What it does

• Costing. Input and output tokens priced separately per model. An event for a model with no price raises UnknownModel instead of silently costing zero.
• Rollups. Total, by team, by model, and daily, optionally scoped to one team.
• Budgets. over_budget() returns each team past its limit with spend and budget side by side.
• Spikes. A day is flagged when it exceeds the mean of all prior days by z deviations. The deviation is floored at 10 percent of the mean so flat, healthy history does not produce false alarms.

Tests

python -m unittest discover -s tests -t . -v

Part of a set of small, tested tools for running AI in production, alongside mao-framework. MIT licensed. Built by AndMaverick LLC.

---

Related reading: The Meter Is Running. Nobody Built the Architecture to Stop It.

If this is your problem. If your AI spend is outrunning your results, that is an architecture problem, and it is solvable. I am happy to look at it with you. Begin a conversation.