dsh-model-provider-label

English | 中文

dsh-model-provider-label — show the provider in the "current model" button

2.0 (2026-10-21): rewritten as a slot plugin. DSH 0.2.1 exposes the conversation.input.model.label render slot, so this plugin now registers a renderer instead of patching the compiled dsh-client-ui-model-selection bundle. The old patch.mjs is deprecated and kept only for historical reference — do not run it against 0.2.1, where its anchors no longer match.

1. The problem

DSH's model selector already groups models by provider: once you expand the list, every group has the provider name above it (the MenuGroup title, taken from the provider's display name). But after you select a model, the button above the input box shows only the model name:

deepseek-v4.1-flash           ← is this 「阿里云(公司)」 or 「阿里云(个人)」?

When several providers offer a model with the same name (for example, both 阿里云(公司) and 阿里云(个人) expose deepseek-v4.1-flash, and both llm-deepseek and 和仁平台 expose deepseek-flash), the group title cannot help once the selection is made — it is gone.

With this plugin:

和仁平台 · deepseek-flash · Default
└─ dimmed ─┘   └─ normal ─┘   └─ effort ┘

The value of the "Model" row in the menu root panel carries the provider as well.

2. How it works

The plugin registers a renderer for the official render slot conversation.input.model.label:

• client.js — registers the renderer through ctx.slots.inject(...), so it mounts only while the plugin that declares the slot (@deepseek-ai/dsh-client-ui-model-selection) is active, and rolls back when this plugin unloads.
• index.js — a no-op host entry. A profile bundle row is a host plugin row, so the package must be insertable as a Cordis plugin; all the work happens in the browser.
• cordis.patch.yml — inserts that host row.
• package.json — declares dsh.bundle and injects the model-selection client package through dsh.client.inject, so the slot provider is loaded first.

The renderer puts the provider name in a dimmer caption color before the model name. Because the provider comes first, ellipsis truncation drops the tail of the model name first and never swallows the provider. Rows in the expanded model list (placement: "option") keep the bare model name: that list is already grouped by provider, so repeating the name there would only add noise.

3. Install

Requires DSH 0.2.1 or newer (the slot was introduced in 0.2.1).

dsh plugin --profile web add github:liufeicc/dsh-model-provider-label

Restart the host (dsh web) after adding the package. For local development, link the directory instead:

dsh plugin --profile web add link:/path/to/dsh-model-provider-label

4. Upgrading DSH

Nothing under node_modules is patched, so upgrading dsh does not remove the plugin. Only a future change to the slot API would require updating client.js.

5. History: the 1.0 patch

1.0 edited <dsh-install>/node_modules/@deepseek-ai/dsh-client-ui-model-selection/lib/client.js in place, inserting five anchored changes marked LOCAL-PATCH:provider-label. Every dsh upgrade overwrote them, and the anchors drifted. patch.mjs is still in the repository for reference; it is not part of the published files and must not be run against 0.2.1.

6. Known limitations

• Only the current selection (the button and the menu's "Model" value) is changed. The expanded list and the /model command popup keep their bare model names; they already show the provider in their group titles.
• The accessibility text (aria) keeps the default model name.

License

This project is released under the MIT License. See LICENSE for details.