tonik

Score and tune how you use AI — right in your terminal.

tonik reads your local Claude Code session logs and shows a beautiful dashboard of how efficiently you're using AI: token spend, cost, cache efficiency, and a single score that tells you whether things are getting better or worse. Then it points at the concrete waste and tells you how to fix it.

No account. No API key. No data leaves your machine — it just reads the logs Claude Code already keeps locally.

$ npx tonik

╭─ Usage · tinaba96 · 14d ──────────────╮ ╭─ Score ────╮
│ ▁▂▃▅▇▆▄▃▂▄▆█             ↓ 18%        │ │            │
│ in 4.2M   out 1.1M   cost $38.20      │ │  ⬢ 72/100  │
╰───────────────────────────────────────╯ │   ▲ +6     │
╭─ Health ──────────────────────────────╮ ╰────────────╯
│ sessions 63   turns/session 7.4 ⚠     │
│ cache hit 41% ⚠                        │
╰────────────────────────────────────────╯
╭─ Findings ────────────────────────────────────────────╮
│ ✂ low cache reuse — $9.10 spent re-creating context   │
│ ✂ 7.4 turns/session — try splitting prompts           │
╰────────────────────────────────────────────────────────╯

Install

npx tonik          # run without installing
npm i -g tonik     # or install globally

Usage

tonik                 # dashboard for the current project (last 14 days)
tonik --global        # all projects combined
tonik --since 7d      # change the time window
tonik --theme card    # switch visual theme (dashboard | minimal | card)
tonik --json          # machine-readable output (for CI / scripts)

How the score works

The score is intentionally transparent — it's the sum of three explainable sub-scores, each grounded in numbers you can verify:

| Sub-score | Max | Based on | | --- | --- | --- | | Efficiency | 40 | Cache hit rate (reused context ÷ created context) | | Conversation | 30 | Avg human turns per session (fewer, well-scoped is better) | | Cost discipline | 30 | Token spend trend over the window (improving earns more) |

No magic, no black box. If you don't like the weighting, it's open source.

Roadmap

• tonik fix — apply the suggested improvements, not just list them
• More themes and custom color palettes
• A tonik-core library so other tools can reuse the analysis

License

MIT © Takahiro Inaba