pi-usage

A pi extension that displays your active model provider's subscription quota usage in the status bar. Supports z.ai (GLM Coding Plan) and OpenAI Codex subscriptions.

What it does

Shows subscription quota usage in pi's footer status bar for your active model provider:

• z.ai (GLM) — 5-hour and weekly quotas with dynamic reset times
• OpenAI Codex — Available subscription limits with provider-specific periods

Displays as a percentage used (e.g., 5h: 16% | week: 4%).

The status bar updates live every 60 seconds during long agent runs.

The status bar also shows:

• Plan level (when available)
• Time until quota reset (relative or absolute as appropriate)

Example outputs:

GLM Lite | 5h: 16% ↻ 2h 14m | week: 4% ↻ Mon
Codex Plus | 5h: 32% ↻ 1h 45m | week: 15% ↻ 23h 0m

How it works

Data sources

• z.ai (GLM): https://api.z.ai/api/monitor/usage/quota/limit with API key from ~/.pi/agent/auth.json → zai.key
• OpenAI Codex: https://chatgpt.com/backend-api/wham/usage using ~/.pi/agent/auth.json → openai-codex OAuth state

Example response:

{
  "code": 200,
  "data": {
    "limits": [
      {
        "type": "TOKENS_LIMIT",
        "unit": 3,
        "percentage": 16,
        "nextResetTime": 1777819631597
      },
      {
        "type": "TOKENS_LIMIT",
        "unit": 6,
        "percentage": 4,
        "nextResetTime": 1778262784969
      },
      {
        "type": "TIME_LIMIT",
        "unit": 5,
        "percentage": 0,
        "nextResetTime": 1780336384978
      }
    ],
    "level": "lite"
  }
}

Unit mapping (discovered from z.ai frontend source):

unit	Meaning
3	5 Hours Quota (TOKENS_LIMIT)
6	Weekly Quota (TOKENS_LIMIT)
5	Monthly Web Search/Reader/Zread (TIME_LIMIT) — parsed but not displayed in v1

Refresh behavior

• Fetches quota data immediately on pi startup (session_start event)
• Starts 60-second periodic refresh when agent begins (agent_start event)
• Stops periodic refresh and fetches quota after each agent turn (agent_end event)
• Live updates every 60 seconds during long agent runs
• Updates the status bar immediately with provider-specific formatting
• Falls back gracefully on errors (shows last known state or hides)
• All fetches are fire-and-forget (non-blocking) with a 5-second timeout

Display

Uses ctx.ui.setStatus("pi-usage", ...) to render a provider-neutral status bar entry:

GLM Lite | 5h: 16% ↻ 2h 14m | week: 4% ↻ Mon
Codex Plus | 5h: 32% ↻ 1h 45m | week: 15% ↻ 23h 0m

• ↻ shows when the quota resets (relative or absolute as appropriate)
• Plan level shown on the left (when available)
• Missing quota types are omitted from the display
• Status entry is hidden for unsupported providers

Configuration

No configuration needed. The extension reads provider credentials from ~/.pi/agent/auth.json:

{
  "zai": {
    "type": "api_key",
    "key": "your-zai-api-key-here"
  },
  "openai-codex": {
    "type": "oauth",
    "access": "access-token",
    "refresh": "refresh-token",
    "expires": 1777819631597,
    "accountId": "account-id"
  }
}

Required fields:

• zai.key for z.ai (GLM) usage
• openai-codex.access, refresh, expires, and accountId for Codex usage

The extension will hide the status bar entry if required credentials are missing for the active provider.

Environment variable (for testing):

Set PI_AUTH_DIR to override the auth file path:

export PI_AUTH_DIR=/path/to/test/auth/dir

This is useful for testing without affecting your actual pi configuration.

Installation

Quick install (recommended)

pi install npm:@unfixed3854/pi-usage

This automatically downloads and installs the extension. Restart pi to load it.

Manual install

1. Create the extensions directory if it doesn't exist:

   mkdir -p ~/.pi/agent/extensions/pi-usage
   

2. Copy the extension files to the directory:

   cp -r pi-usage/* ~/.pi/agent/extensions/pi-usage/
   

3. Restart pi to load the extension

The extension will automatically load and display your quota usage in the status bar.

Troubleshooting

Status bar not showing

Cause: Auth file is missing or required credentials are not configured for the active provider.

Fix: Ensure ~/.pi/agent/auth.json exists and contains valid credentials for your provider:

cat ~/.pi/agent/auth.json

For z.ai (GLM):

{
  "zai": {
    "type": "api_key",
    "key": "sk-..."
  }
}

For OpenAI Codex:

{
  "openai-codex": {
    "type": "oauth",
    "access": "access-token",
    "refresh": "refresh-token",
    "expires": 1777819631597,
    "accountId": "account-id"
  }
}

If the file is missing, configure your provider credentials in pi settings.

Status bar shows old data

Cause: API fetch failed or timed out, falling back to cached data.

Fix: Check the pi console for error messages. Common issues:

• Network connectivity problems
• API endpoint is temporarily unavailable
• Invalid API key or expired OAuth credentials

To force a refresh, trigger a new agent turn (ask a question in pi).

"Request timeout: fetch took longer than 5 seconds" in logs

Cause: The API request timed out after 5 seconds.

Fix: This is usually temporary. The extension will retry on the next agent turn. If the issue persists:

• Check your network connection
• Verify the active provider endpoint is accessible
• Check if z.ai or ChatGPT services are operational

Extension not loading

Cause: The extension files are not in the correct location or have syntax errors.

Fix: Verify the extension is installed correctly:

ls -la ~/.pi/agent/extensions/pi-usage/

You should see:

index.ts
api.ts
auth.ts
format.ts
types.ts
package.json
tsconfig.json

Check the pi console for any error messages during startup.

TypeScript compilation errors

Cause: The extension was not built or has type errors.

Fix: Run the typecheck to identify issues:

bun run typecheck

Fix any reported errors and rebuild the extension if necessary.

Future ideas (out of scope for v1)

• Color coding — green/yellow/red as quota fills up (needs ctx.ui.setStatus color support or ANSI codes)
• /pi-usage command — detailed breakdown with pi.registerCommand()
• Token consumption per model — requires additional API data
• Web search quota (unit 5) — z.ai returns it but it is not displayed
• Configurable refresh interval — make the 60-second interval adjustable
• Detailed error display — show minimal error message instead of hiding status bar

Architecture

The extension is structured with clear separation of concerns:

• timer.ts — Periodic refresh controller with dependency injection (fully unit testable)
• api.ts — z.ai API client with timeout and error handling
• codex.ts — OpenAI Codex usage fetcher and parser
• auth.ts — provider credential reader from ~/.pi/agent/auth.json
• format.ts — Status bar formatting utilities
• usage.ts / router.ts — provider-neutral routing and usage snapshots
• types.ts — Shared TypeScript types
• index.ts — Extension entry point, wires everything together

All modules follow TDD with co-located tests (<module>.test.ts).