DEV Community

AgentChip
AgentChip

Posted on

4 AI Providers, 4 Dashboards, 1 Silent Budget Leak — Track Them in One Command

#ai #devops #python #productivity

Your LLM bill is spread across four dashboards and you have no idea which model is eating the budget.

If you build on AI APIs, you know the drill: OpenAI has a usage page. Anthropic has a usage page. DeepSeek has one. Gemini has one. Your monthly "reconciliation" is you logging into each dashboard, squinting at charts, and trying to remember whether that $80 spike was the client project or the internal agent experiments.

Small teams can't afford a $500/mo FinOps platform. But you also can't keep flying blind — one runaway loop in a background agent can quietly burn a hundred dollars before anyone looks.

So I built a local, zero-dependency tracker that turns four messy CSVs into one answer.

What it does:

  • Auto column mapping — exports from OpenAI/Anthropic/DeepSeek/Gemini use totally different headers. It recognizes them all (Chinese headers too: 平台/模型/金额/币种/项目 — bilingual books are covered)
  • Multi-currency, unified — dollar rows, yuan rows, euro rows all converted to USD in one pass (built-in rates, overridable with --rate)
  • Five-dimension aggregation — by provider, by model, by project, by month, by day. "What did OpenAI cost this month" and "which model burns the most" are both one command
  • Budget alerts — set --budget 200 and it exits with code 1 the moment you're over. Hook it into cron and you get a daily budget watchdog for free
  • Filters — slice by provider/model/project/month with simple flags
  • Markdown + JSON reports, and it's 100% local — your usage data never leaves your machine

The exit-code contract is the real feature:

0 8 * * * python3 llm_spend_tracker.py --csv usage.csv --budget 200
Enter fullscreen mode Exit fullscreen mode

Code 0 = under budget. Code 1 = over budget (cron fires an alert). Code 2 = bad input. That's a daily FinOps check on every AI spend, forever, with no subscription and no upload.

Built with Python stdlib only — no pip install, no dependencies, Python 3.8+. One-time purchase, free updates, instant download.

If you run agents, ship AI features, or just got surprised by a provider bill last month, grab it here: agentechip.com — it's in the store alongside the rest of the ops toolkit (agent process watchdog, backup verifier, uptime monitoring).

Your models are working for you. Make sure their bills are too.

Top comments (0)