🌸

Minami · Metrics

Live usage & model routing across your machines

Usage heatmap

0 turns · 0 tok
Mon
Wed
Fri
lessmoreusage by day · your local time

Live routing

Routed$0.0000
If all-Opus$0.0000
Tokens00 in · 0 out
Saved0%
Haiku00 tok
Sonnet00 tok
Opus00 tok
Opus00 tok
Fable00 tok
connecting…

Waiting for the next turn from any machine…

Model routing

Spend the least, keep the output identical. Reserve the top tier for judgement and push grunt work down to a cheaper model — the token-heavy part billed cheap, quality unchanged. Edit the table in lib/routing.ts.

grep / glob, read logs, fetch one number, check a status

Haiku 4.5· finding ≠ thinking — 5× cheaper than Opus

summarise, draft a reply, build a task, light review

Sonnet 5· near-Opus quality, ~2.5× cheaper

decide, complex code, final synthesis, anything a human reads

Opus 5· judgement work — don't downgrade

very long autonomous run, hardest reasoning

Fable 5· only when Opus visibly struggles — 2× Opus
ModelInOut
Haiku 4.5search & grunt$1$5
Sonnet 5draft & reconcile$2$10
Opus 5judgement · default$5$25
Opus 4.8legacy · pre-2026-07-29$5$25
Fable 5hardest reasoning only$10$50

$ / million tokens · Claude API · verified 2026-07-24

Usage · both machines

Turns—
Tokens—
Cost—
connecting…

Accounts · token-slayer

Loading account pool…

Projects · last touched

Loading projects…

Personal panels

Task log · Trace-back · Analytics · People around me are empty. Point MINAMI_PANELS_FILE at a JSON file (see panels.example.json) to fill them in.