TokenPolice Blog
Notes on LLM cost control, agent loops, and FinOps for AI apps.
How much is this one user costing you?
Seven cuts of one user's LLM spend, and the screen in the TokenPolice dashboard that answers each one.
Sybil attacks: how one person drains your AI credits with a thousand free accounts
Per-user caps don't stop signup abuse: a script that mints accounts gets a fresh budget every time. What the loss scales with, and what bounds it.
The dashboard's first screen, in one minute
What the Overview tab of the TokenPolice dashboard shows when you open it: six numbers, two charts, and which of your agents looped last night.
Getting started with TokenPolice in three steps
Create a key, let your coding agent add the SDK, pick a limit. Under ten minutes, and nothing your users see changes.
When to downgrade GPT-4o to GPT-4o-mini automatically
The price gap is real, but request type is the wrong thing to route on. Route on budget pressure, and check the Audit log before you enforce.
Where observability stops and enforcement starts
Tracing tools tell you what a call cost after it ran. A firewall decides whether it runs. What that changes in practice, and why you still want both.
Per-user LLM budgets in a multi-tenant SaaS
One free-tier user can spend a month of LLM budget in an afternoon. Four decisions that stop it, and the rule that implements them.
Stopping agent loops before they cost you $700
What a loop rule watches, the two detectors TokenPolice ships, and why the stop lands one call late.
Seven hidden multipliers on your LLM bill
The per-request price you planned around is the price of one clean call. Production doesn't make clean calls. Seven things that multiply in between.