Introduction¶

The positioning of DSH is “everything is a plugin”; whether an agent can view images or search for public information depends on the tools it has. However, the default DSH chat does not itself forward the tasks of image understanding and web search (README original text: two capabilities that DSH chat does not forward on its own). Therefore, when the default model is a pure text model, or after switching models, the agent loses these two capabilities.

The kinfxhk/dsh-model-senses introduced below is the plugin that fills this gap. It is maintained by Kinfxhk, classified under “Network Tools” in the plugin category. It provides two actions—seeing images and searching—via DSH’s currently deployed default API; after switching the default model, the tool will follow along on the next call. Its directory page description is: See images and search the web from any DSH agent through the live deployment-default API. MIT.

What is it¶

One-sentence positioning: Attach the model_senses tool to any DSH agent so it can view images and search the web, and these capabilities follow the currently active deployment default API, requiring no extra plugins or configuration.

It does not add new API keys or modify model settings, but directly reuses provider credentials already configured within DSH.

Core Features¶

model_senses provides three actions.

Image Viewing Routing¶

The routing order for see is fixed, making it easy to determine which backend is answering:

1. First try a visual model within the same API;
2. If not, use the deployment default;
3. Then use other connected APIs that have visual capabilities;
4. The response will explain who answered (selected／same-api／default／cross-api).

So even if the default model is switched to a pure text model, the ability to view images will not be lost.

Search¶

• Default API is xAI: uses xAI native web_search, includeX: true adds x_search;
• Other APIs: uses DSH’s ctx.web, returns up to 8 source URLs;
• Native search failure automatically falls back to harness web, no silent failure.

Credentials¶

• Non-xAI: handed to ctx.llm adapter, using your originally configured provider credentials;
• xAI: can use XAI_API_KEY, or the xAI OAuth grant saved in DSH; the grant auto-refreshes, and the token won’t appear in tool results.

Installation and Activation¶

The plugin requires DSH 0.2.1-rc.5 or higher (from engines.dsh in package.json). Installation is divided into three steps, with commands from the README.

1. First, get the source code; choose one of two methods: git clone, or Code → Download ZIP on GitHub.

2. Install via plugin_manager in DSH, fill target with the absolute path of the plugin folder:

plugin_manager(action: "install_bundle", target: "C:\\path\\to\\dsh-model-senses")

3. Restart DSH. Afterward, the agent will have the model_senses tool, requiring no extra configuration.

To confirm the version, you can check package.json:

"name": "@local/model-senses"
"version": "1.1.1"
"license": "MIT"
"engines": { "dsh": ">=0.2.1-rc.5" }

Typical Usage¶

After installation and restart, no configuration is needed; trigger it directly in the conversation.

1. Viewing images: Let the agent use see to pass a local image (png／jpeg／webp／gif) or https image URL, then give a question. If the current model cannot see it clearly, the plugin will select a backend that can based on the routing rules above, and mark selected／same-api／default／cross-api in the result.

2. Searching: Let the agent use search to check public information. When the default API is xAI, it uses native web_search, adding x_search if includeX: true; for other APIs, it uses ctx.web, returning a maximum of 8 source URLs. Native search failure will automatically switch to harness web.

3. Troubleshooting: Use status to view the current route to provider／model／backend; add probe: true to perform a real-time check to confirm if the backend is actually available.

Suitable Scenarios and Notes¶

Suitable for: Developers whose default model is a pure text model but still want the agent to view images; those who want search capabilities to follow the current deployment default API rather than maintaining a separate provider set; those who want to double-check routing and credential status outside the sidebar.

Notes to consider:

• The plugin runs with the current DSH process permissions; it does not add API keys or modify model settings itself, but sends images to the routed API. The plugin itself does not store images.
• Image limit is 8 MiB, answer max 12,000 words; see timeout is 120s, search timeout is 180s.
• Search results are a maximum of 8 source URLs.
• It is recommended to read the source code once before installing to ensure patches like cordis.patch.yml meet your expectations; the plugin is under the MIT license.

It is part of Kinfxhk’s plugin series¶

README lists the same batch of plugins. The positioning of Model Senses is “view images + web search, following the live deployment default API,” requiring no extra plugins. The series also includes Attention Center (sidebar notification inbox), Computer Use (sidebar “Computer” page, requires Windows + Cua driver), Model Shelf (sidebar model rack), and Perplexity Guard (fixes Perplexity Opus／Sonnet 5.5、GPT-6.1 Sol requests and provider settings, requires Perplexity API key).

Conclusion¶

The value of dsh-model-senses is straightforward: it fills the gap of image viewing and web search that DSH agents cannot get by default, and makes them follow the currently active default API, reducing the issue of capability loss after switching models. It introduces no new credential complexity, at the cost of one extra code review and one restart. After confirming your DSH version is 0.2.1-rc.5 or higher, follow the steps above to install.

Plugin Directory Page: https://www.skillhub.cn/plugins/Kinfxhk/dsh-model-senses

Source Code Repository: https://github.com/Kinfxhk/dsh-model-senses