dsh-token

When running local inference with DSH, a token bill usually gives you only a total, yet what really moves the budget is cache hit rate, per-session spend, and which requests are quietly burning tokens. dsh-token is a local analytics plugin built to answer exactly those questions. All data is read from your session logs, with zero upload and no online dependency.

The dashboard is an Apple-style four-tab layout (Today, Activity, Sessions, Settings) that splits every token into four segments: miss, cache hit, cache write, and output, then estimates how much caching saves you. A global filter bar spanning time, model, session, and working directory keeps every panel in sync, and the resulting URL is shareable. Drill into a single session for per-request detail, a context-growth curve, and anomaly spikes. A monthly budget ring, multi-model stacked trend chart, 24-bucket hourly histogram, and a 12-month activity heatmap round out the picture.

If your DSH workflow involves multiple models, many sessions, and multi-week inference campaigns-especially when cache cost sensitivity or team-level budget reporting matters-dsh-token turns a fuzzy usage number into a local, drillable, shareable, and attributable dashboard without introducing any external service.