This article is derived from the plugin's already-indexed fields (wiki / faq / compatibility_json / readme), not freshly generated by AI. Source field is noted at the end of each section.

Quick start

deepseek-visionary

— source: plugin_wiki.wiki_content

Install & verify

dsh plugin --profile web add @xlight-oss/visionary-dsh

Run the command above in your DSH Web Profile. Then enable the plugin in the plugin list.

— source: plugins.install

Key points

• visionary-zed-ext：Zed 扩展壳（仅 Zed 需要），按平台从 GitHub Releases 下载/缓存 visionary-server 并启动
• CLI：visionary-server login（可先 status --json 预检）
• MCP / DSH 原生工具：调用 deepseek_vision_login
• CLI：visionary-server vision <image> 识图（详见下文「CLI 工具」）
• MCP / DSH 原生工具：调用 deepseek_vision 传入图片路径 / base64 / data URI 即可识图

— source: plugin_wiki.readme_en (fallback readme_raw)

FAQ

What should I do if the tools don't appear in the DSH tool list after installation?

Confirm successful installation via dsh plugin --profile web add. After restarting DSH, 5 tools (deepseek_vision/ocr/status/login/logout) should appear in the tool directory. You can use dsh --profile web --dump-config to confirm the @xlight-oss/visionary-dsh layer is loaded.

Is a DeepSeek API Key required?

No. This plugin calls the DeepSeek web version vision model (chat.deepseek.com), reusing web credentials through automatic browser login. You need to run deepseek_vision_login to complete login before using it.

Why are images rejected when pasting with a pure text model? How to enable it?

Text models don't support image viewing natively, so the DSH host rejects them with MODEL_DOES_NOT_SUPPORT_IMAGES. This plugin's visionary-image-bridge is enabled by default (enabled: true), which saves the image to disk and rewrites it as guidance text, allowing the agent to call deepseek_vision for analysis. If it's turned off, you can re-enable it in Settings → Visionary.

Where are images and session data stored? Is it automatically cleaned up?

The bridge saves pasted images to ~/.deepseek-visionary/pasted (directory permission 0700, file 0600), with lazy cleanup after 7 days by default (retainHours: 168). The original image bytes in the host attachment library are permanently retained and unaffected by cleanup. To completely delete images, you need to clear the corresponding session.

How to uninstall the plugin?

Use dsh plugin --profile web remove @xlight-oss/visionary-dsh to uninstall and restart DSH. After uninstallation, pasting images with text models will revert to the original behavior of being rejected by the host.

Why does deepseek_vision report a CONTENT_EMPTY error?

This is a known upstream binary issue, fixed in visionary-server ≥0.5.x (no longer aborts when images have no OCR text). Please reinstall the latest binary according to the README instructions.

Which operating systems are supported?

The plugin itself is cross-platform Node.js. The visionary-server binary provides installation scripts for macOS, Linux, and Windows; on Windows, the plugin additionally parses the npm shim to locate the actual exe.

— source: plugin_wiki.faq_json

Compatibility

• DSH: 未声明（peerDependencies 锁定 dsh-tools/dsh-llm/dsh-attachment/dsh-settings ^0.1.0-rc.6，cordis ^4.0.1）
• Node: >=20
• Platforms: macOS, Windows, Linux

— source: plugin_wiki.compatibility_json

Pitfalls

Review the upstream repo before installing. This guide is auto-derived from indexed fields and may lag the latest release. If anything contradicts the official docs, treat the upstream source as authoritative.

— source: general rule