dsh-usage-guard

When you work with the DeepSeek Harness Web GUI, token consumption and API costs tend to scatter across individual sessions, making them hard to quantify and easy to overlook until the bill arrives. dsh-usage-guard plugs exactly that gap: it silently hooks every assistant message in the background, folds usage into daily buckets keyed by provider and model, survives restarts, and even backfills from historical session logs on first run—no manual bookkeeping required.

Open Settings → Usage to see a full dashboard: six stat cards summarizing tokens, sessions, messages, active days, current streak, and top model; a GitHub-style yearly activity heatmap with weekday and month labels plus a hover detail card for each day; a multi-series trend chart that plots input, output, cache-hit, cache-write, and cost on dual axes with a 7-day / 30-day toggle; and a donut chart breaking down usage by model. The cost-estimation module ships with DeepSeek list prices and lets you set custom per-million-token rates for input, output, cache-read, and cache-write. Models without a configured price are simply excluded from cost tallies rather than guessed at. If you want a hard guardrail, the daily-limit guard offers two modes: warn shows a global banner, while block aborts over-limit LLM requests before they ever hit the network.

It is well suited to individual developers who want usage visibility without changing their workflow, and to teams that call multiple models or providers and need per-model cost accounting. The plugin is bilingual (zh/en), follows DSH theme tokens, and has zero external runtime dependencies—install and go.