dsh-personal-center

A local, offline personal center for DeepSeek Harness — usage stats, cost estimation, custom instructions, global font size, and a desktop pet.

A local plugin for DeepSeek Harness (DSH) desktop / web. Adds a Personal Center section to Settings, with four tabs:

• Token Usage — real usage stats: today / cumulative tokens, session count, tool calls, a GitHub-style activity heatmap (daily / weekly / cumulative), per-model breakdown, top tools, and session review; with sub-tabs (Overview / Review / Model Cost).
• Personalization — global custom instructions (like ChatGPT / Codex "Personalization → Custom instructions"), applied to all chats on this machine.
• Appearance — a global font-size stepper (default 14, range 11–16, step 1): scales the whole UI instantly and remembers your choice.
• Desktop Pet — bitmap (black / blue whale) and vector ("round black / round blue") skins, driven by your real usage (5 emotions), with a session-status overview (double-click a session to jump into it).

Model cost estimation lives inside Token Usage: per-model pricing (peak / off-peak, per currency, official presets) plus cache-hit-rate stats.

100% local — no network, no reading chat content. See PRIVACY.md. Bilingual UI (中文 / English), follows DSH's language automatically.

📸 Screenshots

Four tabs (v1.0)

Token Usage	Personalization

Appearance (global font size)	Pet

Feature details

Token Usage · light	Review · dark

Pet panel · light	Pet panel · dark

Session status overview (double-click to jump)

Five emotion animations (happy / busy / tired / wallet-pain / dozing):

Black whale	Blue whale

New actions (v0.8.0) (thinking / waiting / celebrate / drag / wave):

Black whale	Blue whale

Model cost · dark

Personalization · per workspace (layered instructions + templates)

✨ Features

Token Usage (statistics)

• Today overview: token consumption / sessions / tool calls;
• Cumulative data: total tokens / longest chat / total sessions / cache-hit rate / estimated cost;
• Token activity: GitHub-contribution-style heatmap with daily / weekly / cumulative views;
• Per-model breakdown: tokens, requests, per-model cache-hit rate and cost split by provider + model;
• Top tools: sorted by call count (including MCP tools in mcp__<server>__<tool> form);
• Session review: recent sessions (title / date / duration / tokens / cache-hit rate), auto-excludes archived sessions.

Data source: aggregates local session logs in real time — numbers only, never reads content. See docs/common/DESIGN.md.

Model Cost (estimation)

• Per-model cost per million tokens, split by currency (¥/$);
• Peak / off-peak pricing (DeepSeek official: Beijing 9:00–12:00 & 14:00–18:00 peak, off-peak = half);
• Built-in presets (deepseek / kimi / gemini / gpt) with provider→model autocomplete, editable;
• Shows this week / this month / cumulative cost.

Personalization (layered instructions: v0.7)

…