Plugins

Balance and live session usage for the DSH web UI: header balance button (left of the Session log), dropdown with account balance, real-time token breakdown (cache miss/hit, cache write, output), cache hit rate, and official-rate cost estimates with peak/off-peak auto-switch.

Codex-style context visibility for DeepSeek Harness: live context-window occupancy, composition breakdown, KV-cache hit rate, and a GitHub-style 53-week usage heatmap

Environment housekeeper for the agent: toolchain inventory (node/pnpm/git/gh/ffmpeg/browsers), scratch/cache scan with whitelist-guarded one-click cleanup, and the machine rules file (AGENTS.md) editor - all in the Web GUI settings.

Vision plugin for text-only DeepSeek Harness models: native image paste, layered evidence memory and cache, and intent-driven tool selection.

Vision toolkit for text-only DeepSeek: model-invokable `vision` tool, wrapper adapters for deepseek/opencode-go (v4 flash/pro), Antigravity IDE quota (default, flash/pro) / any OpenAI-compatible VLM / Gemini / local Ollama channels, evidence memory with compaction rehydration, content-hash cache, and a bilingual client panel.

A slim port of Headroom: compresses tool output, cuts up to 20% of context, without affecting the model's context cache or agent performance.

Persistent provider-reported token usage totals with per-session, per-model, and daily activity statistics.

Sidebar billing dashboard: real usage aggregated from session logs (per-model and per-day calls/tokens/cache), cost estimated in CNY from a current multi-provider pricing catalog, subscription-plan (coding/token/agent) routes exempt, and per-provider health dots.

Yellow one-line notice for a prompt cache miss after the cache was rebuilt; shows re-billed uncached input tokens, cached hit tokens, and TTFT.

Request Context Profiler for DeepSeek Harness — see what changed between model requests, and how cache reuse changed with it.

Composer-dock balance/usage readout for DeepSeek, Kimi Coding, OpenCode Go, Zhipu, MiniMax, OpenRouter and OpenAI Codex — provider-based, 2s polling, 5min cache.

Read-only performance diagnostics for DeepSeek Harness: session load (open/restore) timing, spill-hit counts, compaction count and trigger, context-injection volume (AGENTS.md/skills/tool-schema token share), and LLM cache hit rate — surfaced via the /fas

DeepSeek bills peak hours at 2x, and peak is 09-12 and 14-18 Beijing time. npx dsh-lean audit shows what your session paid and what it costs off-peak. DeepSeek 峰时按 2 倍计费，峰时正好是上班时间。

DeepSeek Harness web plugin that displays cache-hit rate with two decimal places

Guards the persisted session-projection cache: per-row JSON degradation so one misbehaving projection unit (e.g. a third-party plugin storing a Map) can never stall session titles or the whole cache again, plus a startup self-heal that backfills missing titles.

Weak-network adaptor: transparent long-backoff model-stream retries (10-min cap), local response replay cache, heartbeat auto-reconnect, and degraded-mode token economy, with a bilingual (zh/en) settings page.

Settings-page usage dashboard: per-provider/model token summary across all sessions, with today/7-day/30-day CSS-bar trends and a cache-hit rate.

Vendor-based usage dashboard: per-provider token/cache/output KPI, a 53-week heatmap, trend chart with hourly "today" view, model drilldown, cost estimation, CSV export, and health cards (TTFT, generation speed, peak context, error rate).

Agent FinOps / token analytics for DeepSeek Harness: usage collection, time-aware cost engine, cache analysis, and session drill-down