Cap your AI bill: a usage-limit dashboard in a spreadsheet

Costs last verified: 2026-10-15 (public pricing pages only; not yet run end-to-end by us). By The Agent Ledger.

Draft: pending review and a real test run before publishing.

Cap your AI bill: a usage-limit dashboard

Short answer: List every AI subscription and API key you pay for in one spreadsheet. Once a week, paste in usage from each provider's billing page, set a hard spend limit or budget alert wherever the provider offers one, and send simple, repetitive tasks to a cheaper model. The dashboard itself should cost $0–5/month (estimate to verify). What you save depends on what you cut.

The problem

AI spend leaks in two ways: subscriptions you forgot about and API usage that spikes (an agent loops, or a big model handles a job a small one could do). Fixed-price plans also have usage limits that reset on their own schedules, so "why am I rate-limited?" and "why is my bill $X?" are really the same question.

What you'll get

One sheet with every AI cost in one place: fixed plans + metered API
A weekly 10-minute update routine
Spend limits/alerts turned on wherever your providers support them
A short "routing rule" listing which tasks go to a cheaper model

Tools

A spreadsheet (Google Sheets, Excel, or a CSV + a small script)
Each provider's billing/usage page (consumer plans show usage in settings; for example, Claude says you can see where you stand in Settings > Usage, per https://claude.com/pricing)
Optional: an agent to read exported CSVs and fill in the sheet

Setup steps

Inventory. Go through your card statement and email receipts for the last 2–3 months. Add every AI-related charge as a row: vendor, plan, monthly price, billing day, account email, what you actually use it for.
Build the sheet. Two tabs:
Plans: vendor | plan | $/mo | renews on | limit type (e.g., rolling 5-hour + weekly) | last time used | keep? (Y/N/?)
Metered: week | provider | model | input tokens | output tokens | $ spent | project
Pull usage weekly. Each provider's console or settings page shows usage. Export CSV if it's offered, otherwise copy the totals by hand. (Optional: have an agent turn exported CSVs into Metered rows. Give it the files, not your account login.)
Set hard limits and alerts. In every API console that supports it, set a monthly spend limit and an alert threshold (for example, 50% and 80% of budget). For subscriptions, put a calendar reminder 2 days before each renewal. Both Anthropic and Cursor say you must cancel at least 24 hours before renewal to avoid the next charge (https://claude.com/pricing, https://cursor.com/pricing).
Add a total and a budget line. Total this month = SUM(Plans $/mo where keep≠N) + SUM(Metered $ this month). Turn the cell red when it goes over your budget.
Write a routing rule. For example: "Summaries, classification, and formatting → the cheapest model that passes a 10-sample spot check. Coding and multi-step reasoning → the main model." Test on 10 real examples before you switch.
Monthly review (15 min). Sort Plans by "last time used." Anything unused for 30+ days gets a "?" and a decision from you.

The bill (monthly, USD)

Cost of the dashboard itself | Scenario | What you pay for | Est. cost/mo | Source | |---|---|---|---| | Low | Spreadsheet on an account you already have | $0 | — | | Typical | Same, plus a few cents of API tokens if an agent parses CSVs | ~$0–1 | Token rates: https://claude.com/pricing | | High | A paid spreadsheet/automation add-on | up to ~$5 | Verify with the specific add-on |

Reference prices to put in your Plans tab (verify, they change) | Plan | Listed price | Source | |---|---|---| | Claude Pro | $20/mo monthly, $17/mo billed annually ($200 up front) | https://claude.com/pricing | | Claude Max | From $100/mo (5x or 20x Pro usage) | https://claude.com/pricing | | ChatGPT Go / Plus / Pro | $8 / $20 / $100, $200, or $500 per month | https://help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers, https://howply.com/blog/chatgpt-claude-gemini-plans-compared/ | | Cursor (plans as listed on its pricing page) | Free $0, Plus $8, Premium $30, Super $100, Ultra $200 per month; third-party models billed separately at API list price | https://cursor.com/pricing |

Model-routing example (API list prices per million tokens, input/output) | Model (Anthropic) | Input | Output | Source | |---|---|---|---| | Haiku 5.5 (prompts ≤100K tokens) | $0.50 | $2.50 | https://claude.com/pricing | | Sonnet 5.5 | $2 | $10 | https://claude.com/pricing | | Opus 5.5 | $4 | $20 | https://claude.com/pricing |

How to read this: at these list prices, the same token volume costs 4x less on Haiku 5.5 than on Sonnet 5.5, and 8x less than on Opus 5.5. That's a property of the price list, not a measured saving. Your real result depends on whether the cheaper model does the task well enough.

Where it could go over: an agent stuck in a retry loop on a metered API key. A hard spend limit is the only reliable guard, and alerts are only useful if someone reads them.

✋ What stays manual

Choosing which plans to keep or cancel. The sheet shows you the data. You make the call.

Clicking "cancel." Don't give an agent your billing login.

Setting the budget number. Only you know what the work is worth to you.

Judging quality after routing. A cheaper model that gives wrong answers isn't a saving.

Security notes

Never paste API keys into the sheet. Track keys by a nickname and their last 4 characters only.
Give the agent exported files, not account access. Billing consoles can usually also create keys and change plans.
One key per project makes it obvious which project caused a spike, and lets you revoke one key without breaking everything.
Rotate any key you've pasted into a chat, a repo, or a screenshot.
Turn on 2FA on every AI vendor account that bills your card.

Common bugs & fixes

Symptom	Likely cause	Fix
Sheet total ≠ card statement	Taxes, currency, annual plans charged in one month	Add a "billing cycle" column; split annual cost ÷ 12
You hit a limit while the sheet says you're fine	Fixed plans limit usage, not dollars (e.g., rolling windows + weekly caps)	Track "times rate-limited this week" as its own column
Metered spend jumps one week	Long contexts, a loop, or a bigger model chosen by default	Check the model column first; then check for retries

FAQ

Do I need code for this? No. A spreadsheet and 10 minutes a week is enough. Scripts are optional. Should I switch everything to the cheapest model? No. Route by task, and test on real examples first.