Subscription Pace

Subscription Pace

Track several AI subscription allowances in one panel: how much of each you have used, whether it will last until the reset, and which needs attention.

What it shows

A single chart with every subscription as its own line, plus a filter to look at one, two or three at a time. Each row names the percentage used, the day the current rate runs the allowance out, and when it resets.

Vercel Token Pass and OpenCode Go occupy slots 1 and 2, read live. Slots 3 to 5 take any kind from the catalogue: Claude, ChatGPT / Codex, Copilot, Cursor, Gemini, the Anthropic and OpenAI APIs, or a figure you type in.

Why percent of allowance

Providers do not agree on units. Vercel publishes dollars. OpenCode publishes percentages and no history at all. So every slot is measured the one way they share:

usedPercent = spend / that subscription's own allowance × 100

That is the chart's axis, fixed at 0 to 110% so the allowance rule stays visible whether a line sits at 6% or at 100%. Dollars appear on the row in one of two forms: measured, when the provider publishes money, or marked ≈ with the percentage it came from, when you entered the limit. A plan on the plugin's own suggested limit shows none — it says "no dollar limit set", because a figure the plugin chose is not one it may multiply out and call a spend.

There is no combined percentage. It was removed rather than fixed: the arithmetic was sound but its inputs were not. It was weighted by each slot's allowance, and OpenCode's is a per-model monthly limit ($15–$100) the endpoint never names — the same 27% is $27 on a $100 model and $4.05 on a $15 one, so any weight is a guess.

The header carries a count of subscriptions in trouble instead — over the limit, or on pace to be before the next reset. That second clause is what an average never could say: 60% used is fine until you know it took three days. Nor is there a "closest to its limit", since a plan at 95% on day 30 is not worse than one at 60% on day 3. Each row carries its own verdict.

Where the numbers come from

• Vercel Token Pass — the AI Gateway spend report, read with the key bb already holds.
• OpenCode Go — its /usage endpoint, sampled hourly to build the trend it does not publish.
• ChatGPT / Codex — BB's own Codex provider, which reads the plan's quota.
• Everything else — a figure you enter, dated to the day you last changed it.

Only two settings are credentials: your Vercel AI Gateway key and your OpenCode key. Both are stored in BB's own secret files, never returned to the interface, and never sent anywhere except the provider they belong to.

What it will not do

• It does not guess. A provider that cannot be read shows as unreadable, never as zero — a subscription with an unknown balance is not a subscription with an empty one.
• It does not invent history. OpenCode publishes no series, so the plugin records one sample a day and says so. Codex publishes only its current quota, so that row is a marker, not a curve.
• It does not project a figure that has already run out. When the allowance is spent before the reset, the row names the last day it was spent rather than forecasting a future one.

Setup

1. Add your Vercel AI Gateway key and OpenCode key in the plugin settings.
2. Check the allowance and renewal day on each slot — they start from a guess based on the subscription you picked. OpenCode Go has no default: its percentage is of a per-model limit the endpoint does not name. Enter your model's limit and the row estimates the spend, marked ≈; leave it blank to see only the share.
3. For the Codex slot, enable BB's Codex provider and sign in — bb plugin enable provider-codex. Without it that slot reads as unreadable rather than 0%.

Slots 1 and 2 are on unless you switch them off. Slots 3 to 5 are off until you name a subscription, so an empty slot never starts measuring something you did not choose.