Free vision OCR with adaptive tile recognition for long documents and Markdown/Word/PNG/Excel export.
Install
# from npm (prebuilt)
dsh plugin --profile web add aura-vision
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:Ck-epsilon/aura-vision
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
This plugin publishes its README in Chinese only.
免费视觉识别插件(DeepSeek Harness web profile 永久插件)。设计语言 Aura:宛如天生如此,若有似无。
特性
- 免费通道:智谱 GLM-4V-Flash(免费档)优先;可配置任意 OpenAI 兼容多模态接口;Pollinations 匿名兜底。
- 长文档识别:glm-4v-flash 输出上限 1024 tokens(API 实测)——自适应网格切块(单块目标 1100px、最多 3×3、8% 重叠)逐块识别突破上限,切块模式用逐块纯转录提示词;>3200px 温和归一。
- Aura UI:高透毛玻璃(55% 底色 + 28px 模糊)、三星机身式小圆角、深浅主题自洽配对(CSS
light-dark())、若有似无的次级信息。 - 历史:缩略图 + 原图分离存储;收藏、过滤、删除、清空;导出 Markdown(原图内嵌 base64,单文件自包含)。
- 结果导出:MD / Word(.doc) / PNG 长图 / Excel(含表格时);双击预览图与详情图全屏放大;清除图片联动清除结果。
安装(永久)
# 推荐:npm 安装
dsh plugin --profile web add aura-vision
# 或从 GitHub:
dsh plugin --profile web add "github:Ck-epsilon/aura-vision"
# 或本地路径:
dsh plugin --profile web add "file:<绝对路径>"
其他机器使用前提:本机已 npm login(仅发布者需要);使用者只需装好 dsh 并执行第一条命令。重启 dsh 后生效:任意会话输入框右侧出现「识图」按钮。数据存 <workspaceRoot>/.aura-vision/(随工作区可迁移),key 存本机凭证库。
开发经验
完整开发经验教训见配套文档 LESSONS.md(运行时契约、深浅主题、模型实证法、大图策略、迭代纪律、陷阱速查表)。
兼容性
已在 DeepSeek Harness 0.1.0-rc.7 与 0.1.0-rc.8(web profile)验证。
常见问题
- 安装/升级后按钮没反应、key 保存不上? dsh 在启动时装载插件,安装或升级后需重启 dsh 一次(1.0.0 的 RPC 端点发现 bug 已在 1.0.1/1.0.2 修复)。重启后输入框右侧出现「识图」按钮即为生效。
- 识别长文档很慢? 大图会切块逐块顺序调用(免费档防限流),块数越多耗时越长,属预期。
- 结果被截断? glm-4v-flash 输出上限 1024 tokens(免费模型硬限制);插件已用自适应切块 + 转录优先提示词缓解,更完整结果请配置更长输出的模型并设为当前。
- 换机器后历史还在吗? 历史随工作区
<workspaceRoot>/.aura-vision/迁移;key 存各机凭证库,需在新机器重新填写一次。 - 图片会发给谁? 仅发给当前所选后端(智谱或你配置的接口),本机不向任何第三方上传。
- 如何卸载?
dsh plugin --profile web remove aura-vision。
架构
lib/host.js:宿主半边——AuraVisionService extends TypertRemoteService(命名空间aura-vision),方法经@Remote标记由网关 SRC 模式直连(无需构建期描述符)。lib/client.bundle.js:浏览器半边——window.__ModuleLoader__.load工厂产物,ctx.connection.rpc.call('/api', 'aura-vision/<method>', { args })调用宿主。cordis.patch.yml:bundle 补丁层,一行注册双半边。
License
MIT
Links
More in this category
liustack/modlens★ 4172
Vision bridge for text-only models: paste an image, get structured JSON evidence (OCR, layout, semantics).
ysr666/dsh-vision-router★ 1137
Free vision for text-only agents: built-in keyless vision chain plus pixel tools (Q&A, grounding, crop, pixel diff, colors, OCR, SVG trace, cutout, screenshots); paste an image to use it.
Anionex/dsh-vision-toolkit★ 888
Vision for text-only models: paste an image and the model switches to a Vision Toolkit variant for image Q&A, multi-image comparison, long-screenshot OCR, screenshot-to-UI reproduction, element grounding, and pixel diff. No API key by default — images are processed by the author-hosted free service, 100 per machine per day; configurable to your own provider.
fandc520/dsh-comfyui★ 102
Drive a local or remote ComfyUI server from DeepSeek Harness: comfyui_run / comfyui_object_info / comfyui_workflow tools generate and edit images and videos, with a workflow library (graph extraction: per component / main flow / all), a load area with resolution auto-match, a live queue, SDXL and Wan 2.1 templates, a companion skill, and a same-origin media proxy.
dickpy/dsh-imagegen★ 100
AI image generation for the DSH Web GUI: text-to-image and image-to-image through a configurable OpenAI-compatible endpoint (gpt-image-2 / gpt-image-1 / dall-e-3), with an api_url/api_key settings card and a sidebar split-pane generation studio.
sunxin-ai/dsh-design-qa★ 44
Design-fidelity QA for text-only models: a `deepseek_vision` tool borrows an eye from any OpenAI-compatible vision route, so the model can judge whether an implementation matches its mock — shipped with the benchmark behind that judgement (four fixtures, 23 injected defects, raw transcripts) and the questioning discipline it depends on.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.