torta 🍰

See your AI usage as a pie chart — for Claude Code, Codex CLI, and Gemini CLI, all in one terminal view.

torta means "cake" in Italian: usage, sliced.

It reads local session data those CLIs already write to disk — no API keys, no network calls, no telemetry, nothing sent anywhere.

🍰 torta — AI usage overview

Share of total tokens
   ▄▄█████▄▄     ■ Claude Code 90.8%
 ▄██▀████████▄   ■ Codex CLI   9.2%
▄█████████████▄
███████████████
███████████████
 █████████████
  ▀█████████▀
     ▀▀▀▀▀

Claude Code
  ▄█████▄    ■ Input       0.0%
▄███▀▀▀███▄  ■ Output      0.8%
███     ███  ■ Cache read  97.1%
███▄   ▄███  ■ Cache write 2.1%
 ▀███████▀
   ▀▀▀▀▀
178.57M tokens · $87.66 · 10 sessions · last active just now

Codex CLI
  ▄█████▄    ■ Input      9.7%
▄███▀▀▀▀██▄  ■ Output     0.4%
███     ███  ■ Cache read 89.8%
███▄   ▄███  ■ Reasoning  0.1%
 ▀███████▀
   ▀▀▀▀▀
18.09M tokens · $4.93 · 3 sessions · last active 110d ago

Total: 196.66M tokens · $92.59 estimated

(a real run on a real machine — your numbers will differ)

Install

npx torta

or install it globally:

npm install --global torta
torta

Requires Node.js ≥ 22.

Usage

torta                                  # everything, all-time
torta --days 7                         # only the last 7 days
torta --provider claude                # just one provider
torta --provider claude --provider codex
torta --json > usage.json              # machine-readable, no UI

| Flag | Short | Description | | -------------------- | ----- | ---------------------------------------------------------- | | --days <n> | -d | Only include usage from the last n days (default: all time) | | --provider <name> | -p | Limit to claude, codex, and/or gemini (repeatable) | | --json | | Print machine-readable JSON instead of the interactive UI | | --version | | Show version number | | --help | | Show help |

Supported providers, and where their data comes from

There's no public "usage" API for any of these CLIs, so torta parses the session logs each one already writes locally:

| Provider | Reads from | Notes | | ---------------- | ------------------------------------ | ------- | | Claude Code | ~/.claude/projects/**/*.jsonl | Full token accounting (input/output/cache read/cache write) per message. | | Codex CLI | ~/.codex/sessions/**/*.jsonl and ~/.codex/archived_sessions/ | Full token accounting. Older Codex CLI versions (pre ~Sept 2025) didn't record token counts at all — those sessions are counted but contribute 0 tokens. | | Gemini CLI | ~/.gemini/tmp/**/ | Session/message activity is always shown. Token counts are not persisted locally by default in the Gemini CLI builds we tested against — torta reports it clearly with a warning instead of a fake $0. Enable gemini --telemetry --telemetry-target local if you want Gemini token data to exist on disk at all. |

Each provider's directory can be overridden with the same environment variable the underlying CLI itself uses: CLAUDE_CONFIG_DIR, CODEX_HOME, GEMINI_DATA_DIR.

A provider showing "(CLI not found on PATH)" just means the binary isn't currently installed — its historical usage data (if any) is still shown.

About the cost estimates

Costs are computed from a small built-in, static USD-per-million-token table (see src/pricing/index.ts) and are approximate — provider pricing changes over time and isn't fetched live. Unknown models are excluded from totals rather than priced at $0, so a total can be a partial sum. Treat the numbers shown as directional, not a bill.

Why this exists / what it isn't

There's already a much more complete tool for this — aiusage — with 20+ providers, a SQLite cache, sync backends, and a web dashboard. torta is the opposite: one screen, three providers, no persistence, no config. If you want daily/ weekly breakdowns, billing-window tracking, or historical trends, use ccusage instead.

Development

git clone <this repo>
cd torta
npm install
npm run typecheck
npm run build
node dist/cli.js

See AGENTS.md for the full technical spec — exact on-disk JSON formats for each provider, the accounting rules (Claude is per-message, Codex is cumulative-per-session, Gemini is best-effort), and how to add a new provider.

License

MIT