MIT-licensed dsh plugin that adds pixel-level image reading, OCR and local editing to text-only models via a 'visual-twin' adapter — DSH-native thumbnails, 4 OCR engines, and an optional external VLM bridge.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-09-03
License
MIT
01What can it help you accomplish?
Give text-only models (DeepSeek / text-only) the ability to 'see' images and read documents
Structured pixel-level evidence (hue / structure / material analysis + real OCR + optional VLM description) the model reasons over as plain text
DeepSeek Harness users running text-only models who need image / document understanding without a vision encoder
Give plain-text models a DSH-native thumbnail experience
A '(视觉)' model variant with native thumbnails; pasted images enter the session as image blocks auto-analyzed into a text path + local evidence
DSH operators who want image blocks rendered for text-only models without triggering UNSUPPORTED_CONTENT
Edit and batch-process images locally
CPU-only edits (resize / rotate / filter / composite / watermark / background removal / perspective) — images never leave the machine
Users who need to fix or batch-process images on-device without any cloud upload
02How to install into DeepSeek Harness
Prerequisites
- DeepSeek Harness (dsh) 0.1.1-rc.2 or a compatible 0.1.x build — the README states this release specifically targets dsh 0.1.1-rc.2 and dsheac 5.1.0
- Optional local OCR engines: PaddleOCR or the macOS-native Vision engine (paddle / rapid are cross-platform)
Installation steps
- 01
Install as a dsh plugin, then enable the '(视觉)' variant for the model you want — picturereader generates the variant and shows native thumbnails when you paste an image
- 02
Optionally run `scripts/setup-macos.mjs` to compile the macOS-native Vision OCR engine
Verify the integration
Not specified by the author
03DSH integration and capability boundaries
Native dsh plugin: wraps any text-only model in a 'visual-twin' adapter so it gains DSH-native image thumbnails and pixel/OCR image-reading tools, with an optional external VLM bridge.
Pixel-level image analysis (image_scan / image_ocr)
pasted image or image block→structured pixel evidence: hue / structure / material analysis + real OCR text
Visual-twin adapter (DSH-native thumbnails)
text-only model selection→'(视觉)' model variant with native thumbnail rendering and auto image-block analysis
Local image editing (image_edit)
local image file→edited image via CPU-only Pillow / OpenCV actions (resize / rotate / filter / composite / watermark / background removal / upscale)
images are processed on-device and never leave the machineLocal pixel toolchain (scan / OCR×4 / crop / palette / compare / batch)
images→crop, palette extraction, compare, and batch OCR across 4 engines
Document-to-image & optional external VLM bridge
pdf / word / excel / ppt documents or an external vision endpoint→documents rendered to images; optional semantic understanding via an external OpenAI-compatible VLM
optional external VLM calls require a configured endpoint and network access
04Who is it for? When not to use it?
Good for
- DeepSeek Harness users running text-only models who need image / document understanding without a vision encoder
- DSH operators who want image blocks rendered for text-only models without triggering UNSUPPORTED_CONTENT
- Users who need to fix or batch-process images on-device without any cloud upload
05Compatibility, maintenance and safety notes
- The external vision API is a plug-and-play enhancement, not a required dependency — simple content is handled locally by the pixel-level toolchain and OCR.
- OCR engine options are shown per platform: the macOS-native Vision engine only on macOS, the Windows engine only on Windows, while paddle / rapid are cross-platform.
- Local image analysis and editing run fully on-device; images never leave the machine.
MIT · active maintenance (no release tag yet; peerDependencies target dsh 0.1.1-rc.2)
06Frequently asked questions
What does picturereader do for a text-only model?
It wraps any text-only model (e.g. DeepSeek) in a 'visual-twin' adapter so the model gets DSH-native thumbnails and pasted images become image blocks auto-analyzed into a text path plus local pixel / OCR evidence — the model only ever receives plain text, so it never hits UNSUPPORTED_CONTENT.
Do I need an external vision API?
No. The external OpenAI-compatible VLM bridge is optional and plug-and-play. Simple content is handled locally by the pixel-level toolchain and OCR; the model only calls the external API when it decides the image is worth the extra semantic understanding.
Which OCR engines are available?
Four engines ship in the local toolchain. Engine choice is shown per platform: the macOS-native Vision engine only on macOS, the Windows engine only on Windows, while PaddleOCR and RapidOCR are cross-platform. Run scripts/setup-macos.mjs to compile the macOS Vision engine.
Is my data sent anywhere?
Local image analysis and editing run entirely on-device — images never leave the machine. Only the optional external VLM bridge, if you configure an endpoint, makes outbound calls; everything else stays local.
07Related DSH workflows
modlens
by liustack
Vision bridge for text-only models: paste an image, get structured JSON evidence (OCR, layout, semantics).
agent-vision-toolkit
by anionex
Vision toolkit and skills that give text-only LLMs eyes — multi-image understanding, image Q&A, OCR, frontend UI restoration and GUI automation, with optional agent integration.
dsh-vision-router
by ysr666
Eyes for text-only DeepSeek Harness agents: built-in free vision chain (no key) + pixel-level vision tools (Q&A, grounding, crop, pixel diff, colors, OCR, SVG trace, cutout, screenshots). One-command install, no Python, image turns work like ordinary tool-calling turns.
dsh-vision-toolkit
by anionex
为纯文本 DSH Agent 提供 10 个结构化视觉工具：意图感知图片问答、长截图 OCR、原始像素 grounding、UI 还原、像素 diff 等
08Data and sources
通过「视觉孪生 adapter」让纯文本模型在 DSH 里获得原生缩略图体验
DSH 原生缩略图也需要模型被声明为「支持图片」才会渲染。
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-09-03. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.