dsh-monitor

When you run inference workloads through DeepSeek Harness, three questions tend to go unanswered: how much did this session actually cost, how much quota does the current provider still have, and exactly where did the tokens land. dsh-monitor stitches those fragments into a single, readable bill. It wraps the llm/stream pipeline, captures usage on every model call, and bills precisely against official peak and off-peak tiered prices. A small chip in the session header shows the running cost and token breakdown in real time, and because the projection is event-sourced, the numbers stay accurate across refreshes and reloads. Beyond a single session, the Token Ledger rolls up today, weekly, monthly, and cumulative counts with provider distribution, project distribution, an activity heatmap, and a per-model breakdown.

The provider usage panel rounds out the full picture. DeepSeek official balance is built in and requires zero configuration; the OpenCode Go plan exposes 5-hour, weekly, and monthly usage percentages along with reset times; and a custom preset lets you point at any HTTP endpoint, inject an API key via the {apiKey} placeholder in headers, and define dotted JSON paths so each item renders as a percentage, a number, a money value, or plain text. The price table can be synced in one click from the official DeepSeek pricing page, each model may declare its own peak and off-peak windows in UTC with optional day-of-week restrictions, and dual-currency tables (USD / CNY) follow the UI language automatically.

It is built for developers and teams who need fine-grained visibility into API spend, track quotas across multiple providers, and want to understand which models and time windows are the most cost-effective. If you only fire off the occasional inference call, it sits quietly in the corner; the moment you start caring about exactly where every token and every cent went, dsh-monitor becomes the little ledger sitting right next to your session.