dsh-vision-assist
DeepSeek Harness 视觉助手插件:给没有视觉能力的模型配一个可随时切换的多模态识别模型。输入框图片自动落盘并改写为文本提示,主模型调用 vision_recognize 工具即可完成看图;识别模型在插件设置卡片里热更新切换。
569 results
DeepSeek Harness 视觉助手插件:给没有视觉能力的模型配一个可随时切换的多模态识别模型。输入框图片自动落盘并改写为文本提示,主模型调用 vision_recognize 工具即可完成看图;识别模型在插件设置卡片里热更新切换。
DSH web plugin: open and edit .docx files in the better-sidebar editor through SuperDoc (superdoc.dev) — a browser-native DOCX editor. Assets are self-hosted from this package's node_modules; edited documents export back to disk atomically.
Bridge Google AI subscription (Antigravity/Gemini OAuth, Code Assist API) into DeepSeek Harness as an OpenAI-compatible model endpoint
DSH plugin: give the composer area one uniform translucent frosted-glass material.
Audit DeepSeek Harness disk usage (sessions, attachments, rebuildable caches), classify it by risk, and move anything you choose to delete into the plugin's own recoverable trash. The model can query, never delete.
GitHub operations for DeepSeek Harness: releases, tags, a guarded generic API pass-through, workflow runs, secrets and sanitized mirror publication.
Use the ChatGPT models from the local Codex desktop app inside DeepSeek Harness: a first-class LLM provider plus conversation-access tools, backed by the Codex app-server and the desktop app's own ChatGPT sign-in.
Report-and-ledger plugin for DSH: front-matter 汇报 as the first-class agent-to-agent interaction primitive, with upward/downward delivery, co-authorship, CC, and a complete transfer-path audit trail; plus a conversation timeline tab.
Personal assistant bundle: structured personal data in SQLite with capture, query, and review tools
DSH小助手:原生 Swift 悬浮窗负责展示与交互,宿主插件负责把它接到 DSH 会话状态(含今日工作总结)
dsh plugin: shows the model name the upstream actually reported for each assistant step, backfilling pi-ai replay metadata from a non-invasive fetch observer when the adapter omits it
Managed secret store for DSH Desktop. Operator-owned keys, injected into the assistant's shell as $DSH_* variables, with a settings-page panel and three operator-controlled access modes.
Ask TypeSafe (Jev) for a structured decision from inside DeepSeek Harness: route, classify, score or verify with typed questions, and get probabilities your code can branch on.
Frosted-glass (acrylic) window surfaces for the DeepSeek Harness Desktop application.
DSH plugin (dsh 0.1.5 and 0.1.7+): a ctx.web search provider that searches with the session's currently selected model and its gateway-native search tool (vercel:perplexity_search and friends), replacing the model-hardcoded built-in DeepSeek search provider.
Assorted DSH Desktop UI tweaks: edit-and-resend with branch switching, sidebar branch-family grouping, unified branch titles, and auto summary titles with a pixel shimmer effect.
GitHub accelerator for DSH: hosts takeover + per-domain loopback SNI passthrough (no TLS MITM, no certificate install), a local CONNECT proxy, and a PAC fallback. The toggle lives in the session top bar.
Jev quality bench for DeepSeek Harness: shadow-accounted dangerous-command judging, injection screening for fetched content, canary A/B drills, and an optional gate mode — with the ledger, the report and the settings card needed to judge whether any of it is worth keeping.
Single-user remote authentication for DeepSeek Harness: password, TOTP and revocable sessions.
A thin Feishu/Lark WebSocket adapter for the durable DSH assistant-delivery core.
给同事的运维请求通道:同事在飞书私聊机器人提一件事,你在审批卡上点「同意」后,由你指定的那条 DSH 会话带着你配好的工具与人设去办,结果以卡片回到提交人的飞书(Markdown 正常渲染),不需要公网入口。四个特点:配对码把发送者绑成管理员,机器人只服务指定账号;非管理员请求走审批卡片,同意/拒绝都校验点击人身份;请求串行排队,前一条整轮 turn 跑完才注入下一条;目标会话启动时主动 resume,不用先有人在界面上点开。
DeepSeek Harness plugin: remember the conversation you were last in and reopen it after a restart — implemented as a first-class plugin (host half persists the pointer, client half records and reopens) instead of patching engine files.
DSH 局域网密码门禁:手机/平板输密钥直连 3080,内置 randomUUID polyfill 与移动端登录页。
DeepSeek Harness plugin that captures the raw HTTP request and response bodies of model provider calls, shows the request body as a collapsible size-annotated JSON tree in a Conversation View tab and on a loopback viewer page, and assembles SSE responses back into reasoning, content and tool calls.