@tian.zuo/pi-usage

Release notes: changelog · GitHub releases

Show OpenAI Codex, GitHub Copilot, Z.ai (GLM Coding Plan), Z.ai Coding Plan (China), and DeepSeek account usage from inside the pi coding agent, plus a /tokens dashboard of the token and cost history pi records locally.

/usage opens a menu with the current usage for every configured provider, and a compact meter is shown in the footer whenever the active model belongs to a supported provider.

OpenAI Codex · Plus
  5h limit:         [████████████████░░░░] 78% left · resets 14:20 on 27 Jul

  Weekly limit:     [████████████░░░░░░░░] 60% left · resets 09:00 on 2 Aug

GitHub Copilot · Business
  Premium credits:  [██████░░░░░░░░░░░░░░] 31% left · 7,787 / 25,000 credits
  Quota resets: 2026-08-01

GLM Coding Plan · Lite
  5h tokens:        [████████████░░░░░░░░] 59% left · resets 16:22

GLM Coding Plan (China) · Pro
  5h tokens:        [████████████████░░░░] 82% left · resets 18:05

DeepSeek
  Balance:          ¥27.00
  Topped up: ¥27.00

/tokens opens an interactive token and cost dashboard with an ASCII/Unicode bar chart, peak breakdown, and scrollable model rankings (up to top 10 models):

────────────────────────────────────────────────────────────────────
 tokens · local pi usage
 14 session files · dedup by message id

 1d [7d] 30d MTD
 ←/→ or h/l window · ↑/↓ or j/k scroll · 1-4 jump · Tab · Esc close
────────────────────────────────────────────────────────────────────
 Last 7 days (Aug 23 – Aug 29) · per day, tokens
 1.3M tokens · $41.27
 7 requests · in 892K · out 163K · cache 255K
 cost at list prices (subscription plans may cover it)

 peak 415K on Aug 25
       ██
       ██ ██
    ██ ██ ██    ██
    ██ ██ ██ ██ ██
 ██ ██ ██ ██ ██ ██ ██
 ██ ██ ██ ██ ██ ██ ██
 23    25    27    29

 top models by tokens (1–5 of 7 · ↑/↓ scroll)
   1. anthropic/claude-3-7-sonnet  747K  $25.62
   2. openai-codex/gpt-5.6  453K  $14.55
   3. zai/glm-4-plus  110K  $1.10
   4. google/gemini-2.5-pro  85K  $0.75
   5. deepseek/deepseek-chat  40K  $0.20
────────────────────────────────────────────────────────────────────

Commands

• /usage — open the usage menu. A cancellable loading spinner is shown while the provider endpoints are queried (press Esc to cancel). Pick Refresh to re-query, Close to dismiss. In non-interactive modes it prints a one-line summary instead.
• /tokens — local token/cost history recorded by pi itself, no provider calls. Shows a bar chart with a 1.2M tokens · $39.82 headline for Today (per hour), Last 7 days, Last 30 days, and Month to date (per day). Navigate with ←/→ (or 1-4), scroll top models with ↑/↓ (or j/k, up to top 10), toggle the chart metric with Tab (tokens ⇄ cost), rescan with r, close with Esc/q/Enter. In non-interactive modes it prints one summary line per window.

Statusline

When the active model provider is Codex, Copilot, Z.ai, Z.ai Coding Plan (China), or DeepSeek, the footer shows a compact Azure Blue meter such as codex 60% wk, copilot 31% credits, copilot 49% premium, zai 59% 5h, zai-cn 82% 5h, or deepseek ¥27.00, refreshed at most every five minutes (results are cached to avoid hammering the endpoints).

How /tokens works

pi records every assistant message's usage (input/output/cache tokens, total, and list-price cost) in session files under ~/.pi/agent/sessions/ (or $PI_CODING_AGENT_DIR/sessions). /tokens streams those JSONL files, filters assistant messages with usage, and deduplicates by message id so replayed or resumed copies (repro.jsonl, forks) are counted once. Files whose name-encoded start date is more than 7 days older than the window are skipped; non-standard names are always scanned. Cost is the model's list price recorded at request time — subscription plans (Codex, Copilot, GLM Coding Plan) may cover it, which the panel notes as cost at list prices.

Install

pi install npm:@tian.zuo/pi-usage

Try it without installing permanently:

pi -e npm:@tian.zuo/pi-usage

License

MIT © Tian Zuo