# quotamate

> quotamate tells you which AI account will run out of allowance and when. It tracks the
> quota on each connected subscription, measures how fast it is being consumed, and
> projects the exhaustion time, so you can see which accounts are usable right now and
> which will hit their ceiling before they reset.
>
> It answers that at three scales: one person's own accounts, a shared team space, or an
> organization containing many spaces.

## What it tracks

24 providers are read today:

- **Claude**: Pro and Max. Session, weekly and per-model windows, each with its reset time.
- **Codex**: ChatGPT Plus, Pro and Business plan limits, plus the per-feature windows.
- **OpenRouter**: Credit balance and per-key spend. An inference key is enough.
- **Tavily**: Plan and per-key credits remaining, from your ordinary search key.
- **Ollama Cloud**: The hosted subscription: 5-hour and weekly limits, the same shape as Claude.
- **ElevenLabs**: Characters used against your plan limit, with the reset date.
- **Firecrawl**: Credits and tokens, metered separately because either can run out first.
- **DeepSeek**: Prepaid balance, and whether the account can still make calls at all.
- **PostHog**: Events, recordings and feature-flag requests against your organisation limits.
- **Cloudinary**: Storage, bandwidth and transformations against your plan credits.
- **SendGrid**: Emails left in the current billing period.
- **Twilio**: Account balance, covering SMS and Verify together.
- **OpenAI Platform**: API spend this month. Pay-per-token billing, separate from ChatGPT.
- **SerpAPI**: Searches left on your plan, plus any extra credits, with the renewal date.
- **Netlify build minutes**: Build minutes used against the minutes included in your plan.
- **Apify**: Every limit tracked separately, because the one that blocks you is rarely the biggest.
- **Neon**: Compute and storage against your project quota. Exhausting it suspends the database.
- **Upstash**: Spend and commands this month, added up across every database and vector index on the account.
- **Deepgram**: Prepaid balance, in dollars or hours depending on how the project was funded.
- **Fal.ai**: Credit balance, read with an admin key.
- **Cloudflare Workers**: Workers requests per day, against the free ceiling where one applies.
- **GitHub Actions**: Actions spend for the period, with minutes and storage kept apart.
- **Anthropic Console**: API spend this month. Pay-per-token billing, separate from a Claude subscription.
- **Routera**: Tokens billed this period on the router, plus requests. Tokens, not money.

## Being added next

Endpoints confirmed, not yet live: Gemini API (Google Cloud).

## The three scales

- **Individual**: your own connected accounts, private by default, free permanently.
- **Space (team)**: members share accounts into a space and see each other's plan,
  meters and remaining headroom, each labelled with its owner. Sharing exposes usage,
  never the credential, and only the owner can share or revoke.
- **Organization**: rolls up every space, which team is closest to a limit, how many
  accounts are at risk, and which providers the org runs on. An account shared into
  several spaces is counted once.

## How the projection works

Usage is sampled every couple of minutes. A least-squares fit over recent samples gives
a burn rate in percent per hour, which is projected forward to the ceiling. A window that
is flat or falling gets no projected date rather than a far-future one, and a projection
built on too few samples is marked low-confidence instead of being published as fact.

Each account resolves to one of four states: **usable**, **at risk** (projected to run
out before it resets), **exhausted**, or **not reporting**.

## Spend over a chosen period

For providers reporting money or credits, usage can be totalled over last 24 hours, last
7 days, month to date, 30 days or 90 days. "What have I spent this month" is a different
question from "what is my balance now", and both are answerable.

## Key facts

- Reading usage does not consume usage. These are metadata endpoints; they run no model.
- Data is collected roughly every two minutes and retained as a time series.
- Credentials are encrypted with AES-256-GCM under a key derived per user, and every
  decryption is recorded in an audit log.
- Sharing an account into a team space shares its usage and owner, never the credential.
- Individual use is free permanently. Teams $2/user/month, organizations $5/user/month.
  Everything is free during the beta.

## Known limitations

- Several provider usage endpoints, Claude's and Codex's among them, are undocumented.
  They are read-only, but the providers can change them without notice.
- Some providers report consumption with no ceiling at all. Those show a figure rather
  than a percentage, unless you set your own cap, and a cap you set is labelled as
  yours rather than presented as the provider's.

## Not supported, and why

- Google Gemini (consumer): The consumer subscriptions, Google One AI Premium and the individual Gemini Code Assist tier, expose no machine-readable per-user quota. The only surface carrying it is cloudcode-pa.googleapis.com, which is internal, undocumented and not offered to third parties. The Gemini API on Google Cloud is a different matter and is on the list to build.
- Together AI: No balance or usage API exists. Sixteen candidate paths were probed and every one returned 404 while known-good control paths on the same host returned 200, so this is absence rather than a missed URL.
- Ollama (local): Runs on your own machine with no account and no quota, so there is nothing that can run out. Not a gap in coverage, a category difference. Ollama Cloud, the paid hosted service, is supported and is a separate entry.

## Links

- Home: https://quotamate.ai
- App: https://app.quotamate.ai
- Pricing: https://quotamate.ai/pricing.md