dsh-token-stats

Making hundreds or thousands of DeepSeek API calls a day through DSH leaves you with almost no visibility into daily token spend, per-model cost, or account balance. dsh-token-stats replays the real provider usage recorded in DSH session logs (session.jsonl.zstd), reading inputTokens, outputTokens, cacheReadTokens, cacheWriteTokens, and reasoningTokens directly from the API response — no estimation, no packet sniffing.

The web dashboard opens with six stat cards (cumulative tokens, today usage with call count, per-session peak, longest chat, current streak, all-time streak). Below sits a GitHub-style 12-month activity heatmap with daily, weekly, and cumulative views; hover for per-day details, click a cell to drill into per-session breakdowns with titles, and export any range as a BOM-prefixed CSV. Per-model and per-workspace spend rankings make budget leaks obvious, while the five-part cost breakdown (input, output, cache-read, cache-write, reasoning) displays full thousands-separated precision. A persistent session badge tracks live token consumption in the chat header, and a balance capsule polls the DeepSeek balance endpoint near-real-time; when the balance drops below a configurable threshold the indicator turns red to prevent silent call failures.

Built for developers and small teams who rely on DSH as their primary front-end and need ongoing visibility into token spend, model-level cost, and account balance. All data stays local under $DSH_HOME/token-stats/; the API key is resolved in the host process only and never shipped to the browser.