dsh-vision-plugin

Give DeepSeek Harness text-only models a pair of eyes.

English | 中文

v1.1.0 · MIT License

---

The deepseek-v4 family is text-only: you can paste screenshots and photos into the chat, but the model can't see them. This plugin sits in between — images are first sent to a vision model (qwen3.7-plus, kimi-k3, …) which transcribes them into text, and that text is what the main model reads. Paste an image, switch to any text-only model, and it just works.

What it looks like

You: What's this?                    (pasted a Clash Verge logo)

Assistant: That's the Clash Verge logo. A black cat silhouette on the left,
           the text "Clash Verge" on the right. It's a GUI client for the
           Clash proxy core…

The image is transcribed by a vision model; the main model answers from the description, as naturally as if it could see.

Three things it does

1. Paste an image and ask — paste or drop an image into the chat and any text-only model can answer about it. No special command needed.
2. vision_analyze tool — the model can call it directly to analyze a local image file, with an optional question ("What's the value in row 3 of this table?") and an optional vision model override.
3. Settings page for the vision model — a "Vision Model" page in the DSH settings panel: pick the default vision model, see the active route. Bilingual, follows the UI language.

Quick start

Option 1: send a prompt to DSH (fastest)

Copy the block below and send it to a DSH session. The agent will install, configure, and verify everything:

Install the dsh-vision-plugin (DeepSeek Harness vision plugin v1.1.0) for me:

1. Obtain the plugin source (fetch via curl, or read from a local checkout):
   - Host half:   https://raw.githubusercontent.com/Xin-Zhang-IceMan/dsh-vision-plugin/main/plugin/vision-plugin.js
   - Client half: https://raw.githubusercontent.com/Xin-Zhang-IceMan/dsh-vision-plugin/main/plugin/vision-plugin.client.js

2. Define the dynamic Cordis plugin with cordis_define:
   - kind: "new", idPrefix: "visn"
   - code.host   = the full content of vision-plugin.js (the function body)
   - code.client = the full content of vision-plugin.client.js (the function body)
   - name: "Vision Assistant"
   - purpose: one sentence describing the plugin

3. Activate it with cordis_run (mode: "run"); approve the Client half when prompted.

4. If ~/.dsh/settings.yaml does not yet declare image capability for text-only
   models, add (hot-reloaded, no restart needed):
   llm-pi-ai:
     providers:
       opencode-go:
         apiKeyEnv: OPENCODE_GO_API_KEY
         modelOverrides:
           deepseek-v4-flash:
             input: [text, image]

5. Verify:
   - Tool.listTools shows vision_analyze
   - The settings panel has a "Vision Model" page
   - Pasting an image into the chat is transcribed automatically

Option 2: three manual steps

① Load the plugin. Pass plugin/vision-plugin.js as code.host and plugin/vision-plugin.client.js as code.client in one cordis_define call (any idPrefix, e.g. visn), then activate with cordis_run (the Client half needs one approval in the UI).

② Declare image capability. Before a message enters the session, the host checks the current model's input modalities. Declare image capability for your text-only models in ~/.dsh/settings.yaml so they accept images (the plugin transcribes them afterwards):

llm-pi-ai:
  providers:
    opencode-go:
      apiKeyEnv: OPENCODE_GO_API_KEY
      modelOverrides:
        deepseek-v4-flash:
          input: [text, image]
        deepseek-v4-pro:
          input: [text, image]
        # add every other text-only model you may switch to;
        # native vision models need no override

The file is hot-reloaded — no restart. Declare every text-only model you might switch to: otherwise pasting an image after switching to an undeclared model is rejected before the plugin can transcribe it.

③ Chat. Paste an image and ask away.

Settings page

Open DSH settings (bottom-left ⚙️) → "Vision Model" page:

• Status badge with run state and version
• Dropdown to pick the default vision model (auto-select or a specific model) — saved changes take effect immediately
• Notes showing the active model and route

The UI language follows DSH (Settings → General → Language), switching between Chinese and English automatically.

FAQ

Pasting an image fails after switching to another text-only model? That model isn't declared in settings.yaml. Add it as in step ② above — hot-reloaded, effective immediately.

What if opencode-go isn't configured? No problem. Since v1.1.0 the plugin scans all configured providers and picks the first route with a vision model. Only when no provider has one does it degrade (the conversation continues, with a note that the image is unavailable).

Does it survive a DSH restart? Dynamic plugins are process-local. After a restart, re-run cordis_run with the same pluginId/packageId. For a permanent install, migrate the code into a host-composition plugin row.

Does it cost extra? Each transcription is one vision-model call; the same image with the same question is cached, so follow-up turns reuse it. Vision model priority: settings config > call parameter > auto-select.

What happens when the vision model errors? The conversation never breaks: partial output is kept; a total failure retries once with a fallback model; if that fails too, the model is told the image is unavailable and the chat continues.

Under the hood (optional)

Flow: an image enters the session → the llm/stream listener checks the target model — native vision models (whitelist) see the image directly; text-only models get a vision-model transcription dispatched in its place. Translation requests carry a Symbol marker to prevent recursion, and results are cached by image + question.

Tunable constants live at the top of plugin/vision-plugin.js: DEFAULT_MODEL (fallback model), NATIVE_VISION_MODELS (native-vision whitelist), MODEL_PRIORITY (auto-select order).

License

MIT