Dear Past Me: Never Start a Free AI Stack Without a Token Audit

Описание и применение

В статье описывается, как отсутствие учёта токенов в бесплатных AI‑стэках приводит к потере ресурсов и неэффективности. Предлагается добавить логирование запросов в CSV и контроль бюджета, чтобы избежать переплат и неудачных повторов.

Задача

Using free AI tiers without token accounting leads to wasted tokens, silent failures, and runaway costs.

Результат

Token usage became transparent, silent failures were detected, and budget overruns were prevented, saving productive time and avoiding unexpected costs.

Рабочий процесс

• Wrap every LLM call in a logger that records prompt/completion tokens, errors, elapsed time, and timestamp.
• Store each log entry as a CSV row in usage.csv.
• Before starting expensive work, read usage.csv to calculate spent tokens.
• If spent tokens exceed the allowance minus a buffer, abort or reduce workload to stay within budget.
• Use the budget guard in retry loops to avoid infinite retries.
• Monitor usage.csv regularly to maintain visibility into token consumption.

Доказательства

Wrap every API call in a usage logger before any loop starts.

Источник

Add a budget guard that reads the CSV before starting expensive work.

Источник