DSH Hub

jiay98528-dev/dsh-plugin-weaknet-adaptor

dsh-plugin-weaknet-adaptor

UIWeb UI1 GitHub stars· updated 2026-08-17

dsh-plugin-weaknet-adaptor is a community DeepSeek Harness plugin. Read the repository README before installing.

Install

npx @deepseek-ai/dsh plugin --profile web add github:jiay98528-dev/dsh-plugin-weaknet-adaptor

Restart `dsh web` after install. Bundle APIs can change during the developer preview.

README badge

dsh-plugin-weaknet-adaptor DSH Hub badge
[![DSH Hub](https://dshhub.dev/badge/dsh-plugin-weaknet-adaptor.svg)](https://dshhub.dev/plugins/dsh-plugin-weaknet-adaptor)

Paste this into your README. The star count updates with every catalog sync.

From the README

Excerpt from jiay98528-dev/dsh-plugin-weaknet-adaptor, cleaned of badges and images.

🌐 dsh-plugin-weaknet-adaptor

Weak-network adaptor for the DeepSeek Harness — 弱网适配插件

Keep model reply chains alive on flaky networks: transparent long-backoff retries, local response cache, heartbeat auto-reconnect, and degraded-mode token economy — with a fully structured, bilingual (zh/en) settings page.

弱网下保护模型回复链条不断裂:透明长退避重试、本地响应缓存、心跳自动重连、降级省 token——全部参数结构化,设置页中英双语。


✨ Features / 特性

中文English
🔁模型流在 llm/stream 瀑布层透明拦截:降级模式下全缓冲 + 指数退避重试(默认最长等待 10 分钟),网络抖动不再打断回复链条Transparent interception in the llm/stream waterfall: degraded mode buffers and retries with exponential backoff (up to a 10-minute cap by default), so network blips no longer break the reply chain
💾完整成功响应本地缓存(请求指纹 key、LRU、TTL 可配):相同请求零 token 本地重放Local replay cache for fully successful responses (fingerprint keys, LRU, configurable TTL): identical requests replay with zero tokens
❤️断网后本地心跳任务按指数退避自动探测重连(间隔上限默认 10 分钟);任意模型流成功也立即被动恢复After an outage a local heartbeat task probes and auto-reconnects with exponential backoff (cap: 10 minutes by default); any successful model stream passively restores online state
🪙重试等待期间零模型调用(零 token);重连后可选输出上限降级与 tool-result 确定性裁剪Zero model calls (zero tokens) while waiting; optional output-cap and deterministic tool-result pruning after reconnect
🎛️24 个参数全部结构化(分组/类型/范围/默认值),设置页「弱网适配」可全部调整All 24 parameters are fully structured (group/type/range/default) and adjustable in the "Weak Network" settings page
🌍每个参数的说明与场景预期效果均为中英双语,UI 跟随系统语言Every parameter ships with bilingual (zh/en) descriptions and expected-effect notes; the UI follows the interface language

🚀 Install / 安装

npm install dsh-plugin-weaknet-adaptor

Add to your composition (cordis.yml), e.g. in the DSH profile you run:

plugins:
  - name: dsh-plugin-weaknet-adaptor

The Host half activates automatically; the browser half (settings page + status dock) is picked up through the package's dsh.client declaration and served by client-modules — no extra wiring needed. Peer dependencies (@deepseek-ai/*, cordis) are installed automatically by npm.

💡 为什么用正式包而不是动态插件? 动态插件(会话内 cordis_define 创建)只存在于进程内存,进程崩溃或重启后即丢失;本包由 cordis.yml 静态加载,重启后依然生效,真正"安装即用"。

💡 Why a real package instead of a dynamic plugin? Dynamic plugins live only in process memory and vanish on crash/restart; this package is loaded statically from cordis.yml and survives restarts.


⚙️ Configuration / 参数配置

All parameters can be changed live from Settings → Weak Network(设置 → 弱网适配). Changes persist best-effort to <workspaceRoot>/.dsh/weaknet-params.json.

Related plugins