deepseek-harness-usage

If you rely on DeepSeek Harness day to day, your token counts, model calls, and real spend are scattered across local session logs — there is no single local dashboard that ties them together without spinning up an external analytics service.

deepseek-harness-usage adds a read-only Usage workspace to the main sidebar with 30-day, 90-day, one-year, and all-time ranges. Summary cards surface estimated spend, total tokens, model calls, sessions, and active days. An interactive trend chart supports pointer and keyboard tooltips and lets you switch between tokens, cost, or calls, with input / output / cache token ratios broken out. A 365-day activity heatmap can be coloured by tokens, cost, or calls with quartile intensity and per-day drill-down. A built-in pricing catalog of 293 route rules covers 11+ providers — OpenAI GPT, Anthropic Claude, Google Gemini, DeepSeek, xAI Grok, Mistral, and others — and handles prompt-length tiers as well as DeepSeek's UTC peak / off-peak windows. Routes without a matching rule are labelled UNPRICED rather than estimated by guesswork.

The plugin is strictly local-first and read-only: all data flows through a same-origin, read-only endpoint, no external telemetry or pricing requests are made, and the API output never includes prompts, responses, or session IDs. A revision-aware in-memory cache ensures unchanged durable sessions are not reparsed on every refresh. It is built for Harness users who need usage auditing, cost estimation, and trend review without deploying a separate analytics stack.