@timiliang/pi-quotas

Public feature fork of latentminds-ai/pi-quotas, preserving the original MIT license and author attribution.

Includes first-class support for the China-region GLM Coding Plan (zai-coding-cn).

Quota monitoring for Pi. Shows remaining usage and rate limits for Anthropic, OpenAI Codex, GitHub Copilot, OpenRouter, Synthetic, Grok, Z.ai, OpenCode Go, and Kimi Code directly in your Pi session.

Interface

/usage writes one compact quota block into the transcript. Every provider uses its own bright colour across the heading, progress bars, values, and reset times; each quota window stays on one line and the entry never enters LLM context. The footer separately shows only the provider used by the active model.

Install

From npm (recommended):

pi install npm:@timiliang/pi-quotas

From source:

git clone https://github.com/LJC-god/pi-quotas.git
pi install ./pi-quotas

Try without installing:

pi -e npm:@timiliang/pi-quotas

Commands

Command	Description
/usage	Compact quota entry for connected providers
/usage --refresh	Bypass the quota cache and append a new entry
/quotas	Compatibility alias for /usage
/anthropic:quotas	Anthropic quotas only
/codex:quotas	OpenAI Codex quotas only
/github:quotas	GitHub Copilot quotas only
/openrouter:quotas	OpenRouter quotas only
/synthetic:quotas	Synthetic quotas only
/grok:quotas	Grok quotas only
/zai:quotas	Z.ai quotas only
/zai-cn:quotas	GLM China quotas only
/opencode-go:quotas	OpenCode Go quotas only
/opencode-go:setup	Configure OpenCode Go dashboard quota access
/opencode-go:clear	Clear saved OpenCode Go dashboard quota access
/kimi:quotas	Kimi Code quotas only
/tokens	Cross-session token/cost usage
/quotas:settings	Toggle individual features on or off

Features

Static usage transcript

Run /usage to append a compact static quota entry to the transcript. Provider headings are colour-coded and every quota window uses one aligned row containing its progress bar, usage, counts or currency, and reset time. There is no modal, border, close key, or unused vertical padding. Use /usage --refresh to bypass the normal provider cache.

The combined entry hides providers with no configured subscription, credentials that cannot report subscription usage, and successful responses with no quota windows. Provider-specific commands use the same static renderer and remain available for detailed authentication or API diagnostics. /quotas remains as a compatibility alias.

Footer status widget

When your active model is from a supported provider, the Pi footer shows only that provider's quota, balance, or budget. It never lists every configured provider and does not pin OpenCode Go when another model is selected. The compact provider label changes immediately after model selection, and data refreshes every 60 seconds and on each turn. Colours shift from green → amber → red as usage climbs. A dim ~ marks last-known data retained through a temporary network failure.

The footer uses ASCII structural markers so Windows terminals do not need to render decorative separator or reset glyphs:

Codex | 7d: 87% left (reset in 5d23h24m) | cap: OK

Quota warnings

Automatic notifications when projected usage is on track to exceed limits before the window resets. Warnings escalate from warning → high → critical based on your consumption pace.

Per-feature toggles

Use /quotas:settings to enable or disable:

• Combined /usage command and /quotas compatibility alias
• Per-provider commands (/anthropic:quotas, /codex:quotas, /github:quotas, /openrouter:quotas, /synthetic:quotas, /grok:quotas, /zai:quotas, /zai-cn:quotas, /opencode-go:quotas, /kimi:quotas)
• Current-provider footer quota status
• Quota warning notifications
• Defer to Synthetic — when both pi-quotas and pi-synthetic are loaded, pi-quotas hides its own Synthetic footer to avoid showing duplicate quota information. Enabled by default; disable if you prefer to see both footers.

Settings can be saved globally (~/.pi/agent/extensions/quotas.json) or per-project (.pi/quotas.json). Run /reload after changing command visibility.

Supported providers

Provider	Windows	Details
Anthropic	5h, 7d, per-model 7d, extra usage	Utilization percentages; optional overage budget in local currency
OpenAI Codex	5h, 7d, credits, spend cap	Rate-limit percentages; credit balance; spend-cap reached/OK
GitHub Copilot	Premium/chat/completions per month	Remaining/entitlement counts with overage indicators
OpenRouter	Monthly budget, daily/weekly/monthly usage	USD spending tracking with cents precision; optional per-key budget limits; UTC-based period resets
Synthetic	Subscription, search/hour, free tools, weekly tokens, 5h limit	Request counts and token budgets; rolling five-hour rate limit; weekly token regen
Grok	Weekly credits, per-product usage, on-demand spend	SuperGrok credit usage from the xAI CLI billing endpoint
Z.ai	5h, 7d, monthly web searches	Token utilisation percentages (rolling 5h/7d windows); monthly web-search count limit
GLM China	Session/month coding credits, token windows, monthly tools	China Coding Plan usage from the BigModel monitor endpoint; absolute credit counts when available
OpenCode Go	Rolling 5h, weekly, monthly USD	USD spend tracking against tier limits; cross-session token/cost aggregation via the /tokens command
Kimi Code	Rolling 5h, weekly	Coding Plan request allowances with reset times

Credentials

pi-quotas reads existing Pi auth entries from ~/.pi/agent/auth.json:

• zai-coding-cn - Z.ai/GLM China Coding Plan API key (ZAI_CODING_CN_API_KEY)

• anthropic — Anthropic OAuth token

• openai-codex — Codex access token (also reads ~/.codex/auth.json for the account ID)

• github-copilot — GitHub Copilot OAuth token (falls back to gh auth token if needed)

• openrouter — OpenRouter API key (Bearer token)

• synthetic — Synthetic API key (set the SYNTHETIC_API_KEY environment variable)

• xai — Grok/xAI OAuth access token

• zai — Z.ai (Zhipu AI / GLM Coding Plan) API key

• opencode-go — the model API key remains managed by Pi through /login opencode-go; dashboard quota access is configured separately with /opencode-go:setup

• kimi-coding — Kimi Code OAuth access token

For OpenCode Go, run /opencode-go:setup. It uses exactly two guided input screens. Step 1 shows the clickable https://opencode.ai/auth entry point and the expected https://opencode.ai/workspace/<id>/go page format directly above the workspace URL/ID input. Step 2 shows F12 > Application > Storage > Cookies > opencode.ai > auth > Value directly above the masked cookie input and asks for only the auth value. The command does not open a browser. It validates both values before saving them to ~/.config/opencode/opencode-quota/opencode-go.json. /opencode-go:clear removes only this dashboard credential; it does not remove Pi's model API key. OPENCODE_GO_WORKSPACE_ID and OPENCODE_GO_AUTH_COOKIE remain supported and take precedence over the managed file.

No additional setup is required for most other providers if Pi can already use them. Global Z.ai and GLM China are independent providers and both appear when both are configured. The global key is sent only to api.z.ai with Bearer authentication; the China key is sent only to open.bigmodel.cn using that endpoint's raw Authorization format. For Synthetic, export SYNTHETIC_API_KEY in your shell or Pi environment.

Requirements

• Pi >= 0.61.0

Changelog

See CHANGELOG.md for release notes and recent changes.

License

MIT © Latent Minds Pty Ltd

Acknowledgements

This project was inspired by @aliou/pi-synthetic.