quotas
A Rust CLI that auto-detects your AI provider credentials and shows usage quotas for each provider in a single place — either as an interactive TUI or as filterable JSON.
Supports five providers today:
| Provider | Auth discovery | Endpoint |
|---|---|---|
| Claude (Max/Pro subscription or API key) | $CLAUDE_CONFIG_DIR/.credentials.json → ~/.claude/.credentials.json → ANTHROPIC_API_KEY |
GET /api/oauth/usage on api.anthropic.com |
| Codex / ChatGPT subscription | ~/.codex/auth.json → OPENAI_API_KEY |
GET /backend-api/wham/usage on chatgpt.com |
| MiniMax Token Plan | MINIMAX_API_KEY → ~/.minimax |
GET /v1/api/openplatform/coding_plan/remains on api.minimax.io |
| Kimi (Coding Plan + PAYG) | MOONSHOT_API_KEY / KIMI_API_KEY |
GET /coding/v1/usages and /v1/users/me/balance |
| Z.ai / GLM Coding Plan | ZHIPU_API_KEY / ZAI_API_KEY → ~/.api-zai |
GET /api/monitor/usage/quota/limit on api.z.ai |
Install
Usage
# Interactive TUI (default)
# Headless JSON for all providers
# Pretty-printed JSON
# Filter by provider
TUI keybinds: ←↑↓→ navigate, Enter drill into a card, R refresh,
C copy selected provider's JSON to clipboard, Q quit.
Output
Each provider result reports one of:
- available — authenticated and the usage API responded. Includes a
plan name and one or more
QuotaWindowentries (5h,weekly, etc.) with used/limit/remaining and reset timestamps. - auth_required — no credentials were discoverable for that provider.
- unavailable — credentials worked but the server reported the plan is not active or the endpoint returned an error; includes a console URL.
- network_error — transport error reaching the endpoint.
License
Dual-licensed under MIT OR Apache-2.0.