MIT 授權的 DeepSeek Harness 視覺外掛，透過 `dsh plugin add` 原生安裝到 Web 或 Headless Profile，提供 10 個視覺工具、視覺任務手冊與內建免費 Gemini 3.7 Flash 服務，支援圖片問答、OCR 和 UI 還原。
DSH 整合
原生執行環境
作者聲明
安全稽核
未稽核
最後核實
2026-08-21
授權條款
MIT
01它能幫你完成什麼？
Complete high-value visual tasks in DeepSeek Harness: read long screenshots, rebuild a UI from a screenshot or sketch, restore graphics, or operate a GUI from screenshots
Markdown OCR output with chunks and audit info, rebuilt HTML/CSS pages, SVG / transparent-PNG graphics, and verified GUI states — produced by 10 composable tools guided by the bundled vision-skills playbooks
Developers running DeepSeek Harness who want text-only agents to do more than caption images — OCR, UI restoration, grounding and screenshot-driven GUI automation
Give a text-only DeepSeek Harness model a multimodal-style experience: paste an image and ask questions about it directly
Task-aware image answers with intent-focused evidence (e.g. where the error is, where the button is) instead of generic captions, with native thumbnails, session history and artifact previews kept intact
DeepSeek Harness Web users on text-only models who want to paste an image and ask directly, without changing models or copying paths
02如何將外掛接入 DeepSeek Harness？
先決條件
- A running DeepSeek Harness environment with a Web, Headless, or Desktop profile
- Python 3.11+ is preferred; when none is found the plugin downloads a hash-verified standalone Python (about 35 MB) automatically on first use
安裝步驟
- 01
Run `dsh plugin --profile web add @anionex/dsh-vision-toolkit` to install into a Web Profile
$ dsh plugin --profile web add @anionex/dsh-vision-toolkit
- 02
Optionally install into a Headless Profile with `dsh plugin --profile headless add @anionex/dsh-vision-toolkit`
$ dsh plugin --profile headless add @anionex/dsh-vision-toolkit
- 03
For DSH Desktop, open DSH Terminal from the tray and run `dsh plugin --profile desktop add @anionex/dsh-vision-toolkit`, then restart DSH Desktop (the Desktop 2.0.1 built-in marketplace has known installation issues; the terminal command is the reliable path)
$ dsh plugin --profile desktop add @anionex/dsh-vision-toolkit
- 04
Restart the running Web Profile
驗證整合成功
- Open Settings → Vision Toolkit; the free provider is already configured — run Test vision model to confirm it is reachable
03DSH 整合程度與能力邊界
Native DeepSeek Harness plugin installed with the dsh CLI into Web / Headless / Desktop Profiles, exposing 10 vision tools, a bundled vision-skills Skill, Settings panel, Credentials, Artifacts and Web UI presentation
10 composable vision tools
images (pasted, in the session workspace, or long screenshots) plus a task-focused question→image Q&A, pixel coordinates and boxed previews, PNG/JPEG crops, SVG traces, pixel-diff heatmaps with JSON, long-screenshot Markdown OCR, transparent PNGs, palettes, and HTML screenshots
vision_html_screenshot renders local pages via Chrome/Chromium/Edgetool outputs write result files (crops, SVGs, heatmaps, JSON) into the workspacevision-skills Skill playbooks
a visual task prompt, e.g. `(Use vision-skills) Rebuild this image into HTML.`→agent workflows for long-screenshot OCR, UI restoration from screenshots/designs, graphic/logo restoration, sketch-to-structured-code (Mermaid/Graphviz), and screenshot-driven GUI operation
Built-in free Gemini 3.7 Flash vision service
none (no API key required)→a working vision provider configured out of the box after installation
images are sent to the shared service endpoint https://vision.anionex.me/v1first start downloads a hash-verified standalone Python (~35 MB) from a domestic mirror (GitHub fallback) and installs locked runtime dependencies (Pillow, NumPy, vtracer) from a PyPI mirrorAutomatic image input routing in DSH Web
an image pasted into a DSH Web conversation→the text-only model transparently switches to its `(Vision Toolkit)` variant with a reusable workspace path and task-focused description; thumbnails, session history and artifact previews stay intact
04適合誰？何時不該用？
適合
- Developers running DeepSeek Harness who want text-only agents to do more than caption images — OCR, UI restoration, grounding and screenshot-driven GUI automation
- DeepSeek Harness Web users on text-only models who want to paste an image and ask directly, without changing models or copying paths
不適合
- The built-in free service is a shared endpoint with safeguards: 100 images per machine per day, up to 5 images per request, 4 MiB per image, 20,000,000 decoded pixels, and 4,096 output tokens; when capacity is reached it returns 429 with Retry-After. Higher volume requires bringing your own vision provider.
- First-time setup needs network and disk access to prepare the isolated Python runtime: it downloads a hash-verified standalone Python from a domestic mirror (falling back to the GitHub release) and installs locked dependencies (Pillow, NumPy, vtracer) from a Tencent Cloud PyPI mirror before falling back to the official PyPI index.
05相容性、維護與安全提醒
- The built-in free service is a shared endpoint with safeguards: 100 images per machine per day, up to 5 images per request, 4 MiB per image, 20,000,000 decoded pixels, and 4,096 output tokens; when capacity is reached it returns 429 with Retry-After. Higher volume requires bringing your own vision provider.
- First-time setup needs network and disk access to prepare the isolated Python runtime: it downloads a hash-verified standalone Python from a domestic mirror (falling back to the GitHub release) and installs locked dependencies (Pillow, NumPy, vtracer) from a Tencent Cloud PyPI mirror before falling back to the official PyPI index.
- vision_html_screenshot requires Chrome, Chromium, or Edge — without one only HTML screenshot rendering is unavailable while the other tools still work; on DSH Desktop 2.0.1 the built-in marketplace has known installation issues, so installing via DSH Terminal is the reliable path.
MIT · actively maintained (latest release v0.1.36, 2026-08-20)
06常見問題
如何把 DSH Vision Toolkit 安裝到 DeepSeek Harness？
執行 `dsh plugin --profile web add @anionex/dsh-vision-toolkit`（或 `--profile headless` / `--profile desktop`），然後重新啟動對應 Profile。它是原生 dsh 外掛而非 MCP，工具直接接入 Profiles、工作階段、設定、Artifacts 與 Web UI。
使用視覺能力需要 API key 嗎？
不需要。安裝後內建的免費 Gemini 3.7 Flash 視覺服務已自動設定完成。如需更高配額或私有端點，可在「設定 → Vision Toolkit」中更換 provider，並將 API key 存為 DSH Credential。
有什麼相容與前置條件？
需要一個執行中的 DeepSeek Harness Profile。優先使用 Python 3.11+；未偵測到時外掛會在首次使用時自動下載經雜湊校驗的獨立 Python（約 35 MB）。vision_html_screenshot 另需 Chrome、Chromium 或 Edge。
免費服務有哪些限制？
共享免費端點配額為每台機器每天 100 張圖、單次請求最多 5 張圖、單張 4 MiB、解碼像素上限 20,000,000。達到容量上限時回傳附帶 Retry-After 的 429，而非無聲失敗。
我的圖片會被傳送到哪裡？
預設設定下圖片會傳送到共享免費服務 https://vision.anionex.me/v1。也可以在設定中切換為自己的 OpenAI 相容或 Anthropic Messages 端點，Credential 金鑰僅儲存參考，不會被讀回瀏覽器。
07相關的 DSH 工作流程
modlens
作者 liustack
The first vision plugin for DeepSeek Harness, and the vision bridge for every text-only coding agent. Paste an image, get structured JSON evidence (OCR, layout, semantics). | 全網最強 DeepSeek Harness 外掛視覺外掛，為 DeepSeek、GLM 等純文本模型外掛視覺能力，貼上圖片即得結構化 JSON 證據（OCR、版面、語義）。
agent-vision-toolkit
作者 anionex
為純文本模型"看圖“設計更好的視覺工具箱和技能，支援多圖理解，圖片問答，前端UI還原、GUI 自動化等，並可選無縫接入多個主流agent，直接識別貼上圖片｜ A vision toolkit and skill designed for text-only llms — image Q&A, long-screenshot OCR, frontend UI restoration, and GUI automation, with optional seamless integration for Codex, Claude Code, Pi, Oh My Pi, and OpenCode
dsh-vision-router
作者 ysr666
為純文本 DeepSeek Harness 智慧體提供「視覺」能力，內建免金鑰視覺鏈路與畫素級視覺工具，一條命令安裝，無需 Python。
dsh-image-gen
作者 shanliuling
直接在 DeepSeek Harness 對話中生成圖片，無需切換外部工具。
08資料與來源
A more powerful vision toolkit—give text-only models in DeepSeek Harness eyes: image Q&A, long-screenshot OCR, UI restor…
dsh plugin --profile web add @anionex/dsh-vision-toolkit
Native DSH integration: those capabilities live inside Profiles, sessions, Settings, Artifacts, and the Web UI, with a f…
此頁面根據專案公開文件、儲存庫中繼資料與 DSH Plugins 的結構化解析所產生；最後核實於 2026-08-21。發現錯誤？提交更正。
最佳 DeepSeek Harness 外掛
從全目錄挑出的 12 個值得優先安裝的外掛，涵蓋各個分類。