Usage Quotas — Module 13.4

How This Lesson Fits the Module & Volume

Usage quotas are longer-horizon ceilings—daily/monthly tokens, dollars, or requests—beyond per-minute rate limits. They are how orgs enforce budgets as product policy.

Pairs with spend alerts next; quotas are the hard stop, alerts are the early warning.

Related foundations: recount tokens with tiktoken (Vol. 12) and keep payloads inside the context window (Vol. 11).

Learning Objectives

By the end of this lesson, students should be able to:

Distinguish quotas from rate limits.
Set quotas in tokens and/or USD.
Implement soft vs hard quota behaviors.
Scope quotas by org, project, and API key.
Reset and carry-over policies intentionally.
Expose remaining quota to developers/users.

Definition

Usage quotas are maximum allowed consumption over a billing or calendar period (tokens, requests, or dollars), enforced independently of short-term rate limits.

Scope	Example	Owner
Account	Provider monthly cap	Cloud admin
Project	$5k LLM COGS / mo	Eng + finance
API key / env	Staging 2M tokens/day	Platform
End customer	Plan tier allowance	Product

Code: Soft then Hard Quota

class TokenQuota:
    def __init__(self, limit, soft_ratio=0.8):
        self.limit = limit
        self.used = 0
        self.soft = int(limit * soft_ratio)

    def authorize(self, tokens: int) -> str:
        if self.used + tokens > self.limit:
            return "hard_block"
        if self.used + tokens > self.soft:
            self.used += tokens
            return "soft_warn"
        self.used += tokens
        return "ok"

q = TokenQuota(1_000_000)
print(q.authorize(1000), q.used)

Rate limit

Per minute/second
Smooths bursts
Resets quickly

Quota

Per day/month
Enforces budget
Resets on schedule

Both

Needed together
Different UX errors
Different dashboards

Strengths

Predictable maximum spend
Multi-tenant fairness
Clear upgrade paths

Tradeoffs

Hard blocks anger users if opaque
Token vs $ quotas diverge with price changes
Reset cliffs cause rushes

Common Misconception

“RPM limits already protect our monthly budget.” A low RPM sustained 24/7 can still exhaust monthly dollars. Quotas close that gap.

Knowledge Check

Short Answer: How do quotas differ from rate limits? Answer: Quotas are period budgets; rate limits are short-term throughput caps.
True/False: Quotas can be denominated in USD. Answer: True.
Multiple Choice: Soft quota typically: (a) warns, (b) formats disks, (c) changes embeddings. Answer: (a).
Short Answer: Name a quota scope. Answer: Account, project, API key, or customer plan.
True/False: Sustained low RPM cannot blow a monthly budget. Answer: False.
Multiple Choice: Staging keys often need: (a) infinite quota, (b) tight quotas, (c) no logging. Answer: (b).
Short Answer: Why expose remaining quota? Answer: So clients can shed load or upgrade before hard block.
Short Answer: What meter from 13.3 feeds token quotas? Answer: Input/output token usage counts.
Multiple Choice: Hard block should return: (a) silent success, (b) clear error/upgrade path, (c) random 200. Answer: (b).
True/False: Price changes can desync token quotas from dollar budgets. Answer: True.

Key Takeaways

Quotas enforce period budgets; RPM does not.
Prefer soft warn then hard block.
Scope by project and customer tier.
Show remaining quota in APIs/UI.
Next: Spend Alerts & Budgets.

Trainer’s Guide

Lab: Design quota tiers for free/pro/enterprise; define soft% and hard actions.

Discussion: Token quota vs dollar quota—which for a multi-model gateway?

Recap: Quotas are the monthly walls around token spend. Continue with Spend Alerts & Budgets.