Tokan — DSH Token Analytics

English | 中文

A DeepSeek Harness (DSH) plugin that adds a Token Analytics dashboard to the Web UI Settings panel: per-tool token attribution, cost breakdown, cache analysis, optimization signals, historical session replay, and JSON/CSV export.

DeepSeek Harness (DSH) 插件，在 Web 设置面板中提供 Token 分析仪表板：每个工具的 Token 归因、成本分解、缓存分析、优化信号、历史会话重放，以及 JSON/CSV 导出。

---

Features

Token Usage Dashboard

Real-time metrics for the selected session:

• Total Tokens: input + output
• Input (New): tokens not served from cache
• Input (Cache): tokens served from cache hits
• Output: generated tokens
• Cache Hit Rate: percentage of input from cache
• Cost: total cost (when pricing is available for the provider/model)

Optimization Signals

Detects tool operations that deviate from the session's per-tool baseline:

Severity	Threshold	Meaning
💡 OPPORTUNITY	≥1.5x AND ≥200 tokens	Possible optimization
🟡 WARNING	≥3.0x AND ≥800 tokens	Unusual operation
🔴 CRITICAL	≥6.0x AND ≥3000 tokens	Extreme operation

Thresholds are fixed for consistent results (see lib/waste.js).

Visual Analytics

• Token distribution by tool (horizontal bar chart)
• Tool usage summary (donut chart)
• Attribution confidence indicators (HIGH/MEDIUM/LOW)

Historical Session Replay

On startup, the host half automatically replays persisted session logs and rebuilds every session's report. Each session is fully isolated — no cross-session contamination. Sessions with corrupt logs are skipped with a warning instead of aborting the restore.

Cross-Session Summary

An aggregate view across all restored sessions: total tokens, cost, waste signals, and per-session comparison.

Export Functionality

Format	Scope
JSON	Current session / All sessions / Summary
CSV	Current session / All sessions / Summary

Easter Eggs

Syncs with the my-skin plugin's global saturation slider.

---

功能特性

Token 使用仪表板

所选会话的实时指标：

• 总 Token 数：输入 + 输出
• 输入（新）：非缓存命中的 Token
• 输入（缓存）：缓存命中的 Token
• 输出：生成的 Token
• 缓存命中率：输入中来自缓存的百分比
• 成本：总成本（当提供商/模型定价可用时）

优化信号

检测偏离会话内工具基线的操作：

严重性	阈值	含义
💡 OPPORTUNITY	≥1.5x 且 ≥200 Token	可能的优化机会
🟡 WARNING	≥3.0x 且 ≥800 Token	异常操作
🔴 CRITICAL	≥6.0x 且 ≥3000 Token	极端操作

阈值固定以确保结果一致（见 lib/waste.js）。

可视化分析

• 按工具的 Token 分布（水平柱状图）
• 工具使用摘要（环形图）
• 归因置信度指示器（高/中/低）

历史会话重放

启动时宿主半边自动重放持久化的会话日志，重建每个会话的报告。会话完全隔离，无跨会话污染。日志损坏的会话会带警告跳过，不会中断整体恢复。

跨会话摘要

所有已恢复会话的聚合视图：总 Token、成本、浪费信号与逐会话对比。

导出功能

格式	范围
JSON	当前会话 / 所有会话 / 摘要
CSV	当前会话 / 所有会话 / 摘要

彩蛋

与 my-skin 插件的全局饱和度滑条同步

Screenshots / 截图

Cross Session

Per Session

Optimization Signals / 优化信号

Attribution Details / 归因详情

Token / Cost Timeline / Token 成本时间线

Signal Details / 信号详情

---

Installation / 安装

Prerequisites / 前提条件

• DeepSeek Harness (DSH) installed, with its dsh CLI on PATH
• Node.js >= 22.19 or >= 24
• pnpm (the profile plugin manager forwards to it)

Install into a DSH profile / 安装到 DSH profile

DSH loads plugins as profile bundles:

dsh plugin --profile web add github:fthuu/Tokan-dsh-token-analytics

dsh plugin initializes the profile on first use, runs pnpm add in the profile directory, and reconciles the dsh.profile.bundles list automatically. Then boot the Web GUI:

dsh web

Open the printed URL and find Settings → General → Token Analytics.

---

Version History / 版本历史

See CHANGELOG.md for detailed version history. Latest: v0.37.1.

详细版本历史请查看 CHANGELOG.md。最新版本：v0.37.1。

---

License / 许可证

This project is licensed under the GPL-3.0-only License - see the LICENSE file for details.

本项目采用 GPL-3.0-only 许可证——详情请查看 LICENSE 文件。

SPDX-License-Identifier: GPL-3.0-only
Copyright (c) 2026 fthuu

Author

• Xiaohongshu/Rednote: @Epho
• GitHub: https://github.com/fthuu