pi-ollama-usage

A Pi extension that shows your Ollama Cloud quota usage in the footer status line:

Mo:$0.72 1.2%

$0.72 is the exact spend of the current calendar month; 1.2% is that spend as a share of the plan's included credit (the share is omitted while the plan — and therefore its allowance — is unknown). Labels are dimmed and the percentage is color-coded by usage stress — green below 70%, yellow from 70%, red from 90% of the month's credit. Set NO_COLOR to disable colors.

Uses the authenticated GET https://ollama.com/api/usage surface with your Ollama Cloud API key — the same data as the ollama.com/settings dashboard. What that API actually exposes (verified 2026-11-01):

• Exact consumption — totals.usage_usd in dollars, plus request_count and input/cached/output tokens; buckets[] breaks the same numbers down per day. Only three windows exist: range=24h, 7d, 30d (anything else is rejected with range must be 24h, 7d, or 30d).
• Month-to-date is computed, not reported — the quota resets per calendar month, but the API only offers rolling windows. The extension sums the daily buckets whose from is on/after the 1st of the current month, which is exact (the 30d window always reaches past it).
• The plan — POST https://ollama.com/api/me returns Plan (pro, max, team, …) and nothing else about billing; the older limits block is gone from /api/usage. So the allowance is the one number the API does not expose: it comes from PLAN_CREDITS in extensions/index.ts. To calibrate it against the dashboard, read allowance = dollars ÷ dashboard percentage (e.g. $70.71 at 70% → ≈ $101) and put that number in the table.
• Per-model request counts and 30-day activity cost (via /ollama-usage).

The indicator only appears while an Ollama Cloud model is active (provider ollama-cloud, or the built-in ollama provider with a :cloud model).

Install

pi install git:github.com/MoemenMostafa/pi-ollama-usage

(An npm package may follow; until then install from git.)

Setup

Works out of the box if you are signed in to the Ollama app/CLI (ollama signin) — the extension reuses the local daemon's signed identity (~/.ollama/id_ed25519), the same one used for cloud model requests. No API key needed.

Alternatively, create an API key at ollama.com/settings/keys and make it available in one of these ways (checked in order):

1. Pi provider auth — /login ollama or an ollama-cloud entry from pi-ollama-cloud
2. OLLAMA_API_KEY environment variable (e.g. in ~/.zshrc: export OLLAMA_API_KEY=...)
3. ~/.pi/agent/auth.json → { "ollama-cloud": { "type": "api-key", "key": "..." } } or the same under "ollama"

Commands

• /ollama-usage — one-shot full usage report (monthly window, per-model request counts, 4-week activity cost)
• /ollama-usage status — show whether the footer indicator is on/off
• /ollama-usage status on|off — enable/disable the footer indicator (on by default)
• /ollama-auth — interactive auth TUI:
  • Test current credential — live check against ollama.com/api/usage, shows current usage if it works
  • Set / update API key — paste a key, it is validated live, then saved to ~/.pi/agent/auth.json (file mode tightened to 0600)
  • Remove stored key — deletes the ollama / ollama-cloud entries from auth.json (env vars and Pi provider auth untouched)

In non-interactive modes (pi -p, JSON/RPC), /ollama-auth prints the active credential as text instead of opening dialogs.

Footer placement

The indicator is a plain extension status (ctx.ui.setStatus), so it always renders on the extension-status line below the footer — pi's built-in footer and pi-open-tui's footer alike. Its colors are kept by pi-open-tui ≥ 0.4.0, which preserves extension ANSI styles.

It is deliberately not inlined into the host TUI's own stats row. That needed a local patch of pi-open-tui's footer.ts, and the patch broke the TUI whenever pi-open-tui touched that file (0.4.0 renamed the status sanitizer, and the stale patch made every render throw ReferenceError: sanitizeStatus is not defined). No part of pi is patched any more — see CHANGELOG 1.2.0.

Behavior

• Refreshes on session start, model switch, and after turns — rate-limited to at most once per minute (Ollama answers 429 too many requests if you poll harder than that).
• Output is color-coded by usage stress (footer indicator and /ollama-usage report): green below 70%, yellow from 70%, red from 90%. NO_COLOR disables colors.
• Mo:$x.xx is the exact calendar-month-to-date spend; the percentage is that spend against the plan credit and is hidden when the plan is unknown (the API does not report the allowance).
• Hides automatically when you switch away from a cloud model, when no API key is available, or on API errors (footer is cosmetic; errors surface only in /ollama-usage).
• Fetches are bounded: 15s timeout, no redirects, malformed/oversized responses are rejected.

License

MIT