dsh-awesome-model-setting

LIMITATIONS

Known limitations

- **DSH currently supports only `text` and `image` as input modalities.** The harness content model, both LLM adapters, the underlying `@earendil-works/pi-ai` library and the attachment pipeline all agree on this. Audio, video and PDF input do not exist yet. The checkboxes are read from the live settings schema, so a future modality would appear automatically without a plugin update. — **DSH 目前的输入模态只有 `text` 和 `image`**：harness 内容模型、两个 LLM 适配器、底层 `@earendil-works/pi-ai` 库与附件管线一致如此。音频、视频、PDF 输入目前并不存在。勾选框的值域是从实时设置 schema 读取的，所以将来新增模态会自动出现，无需更新插件。 - A route whose `models` list you have narrowed only serves the models you listed; catalog models outside that list are not reachable and are therefore not shown. — 一旦你收窄了某条路由的 `models` 列表，适配器只服务你列出的模型；列表之外的目录模型无法访问，因此也不会显示。 - The page edits model entries only. Provider lifecycle (adding/removing a provider, storing an API key) stays in the shipped **Models** page. — 本页只编辑模型条目。供应商的增删与 API 密钥录入仍由官方 **模型** 页负责。 - Chinese UI strings are currently hard-coded. — 界面文案目前是中文硬编码。