@deepseek-ai/dsh-client-ui-settings-models 0.2.1-alpha.2 → 0.2.1-alpha.4

package/README.i18n.yaml

@@ -18,11 +18,11 @@
   en: db5aaf7f702e1f0c
   zh: 409a78c14fb2b6e3
 /deepseek-ai-dsh-client-ui-settings-models/use-this-package/api-keys:
-  en: c2f2a5c06ef20d5b
-  zh: 49a7eb0b66b4ed53
+  en: f4817eb2e3537ed9
+  zh: e589c193fa85473f
 /deepseek-ai-dsh-client-ui-settings-models/use-this-package/editing-a-provider:
-  en: ec0ba62a4e7185b4
-  zh: 37d1e1e7a3817d12
+  en: 3af73644655c8b7b
+  zh: d689eba772100c63
 /deepseek-ai-dsh-client-ui-settings-models/use-this-package/adding-and-deleting-providers:
   en: 9d533c65c82cd4a4
   zh: 0a5da334e127c035
@@ -30,8 +30,8 @@
   en: 589397a35f82244e
   zh: 342fbceee2c173a2
 /deepseek-ai-dsh-client-ui-settings-models/use-this-package/extension-slots:
-  en: 65e9d8822b67b6c2
-  zh: 7e83be9defda28ee
+  en: 79f8f80bdaba73cd
+  zh: f4d7496b293fbf81
 /deepseek-ai-dsh-client-ui-settings-models/understand-the-implementation:
   en: a87afbec5e339b36
   zh: 6348c81af7e650e1
@@ -42,8 +42,8 @@
   en: 6e04c00818f014cd
   zh: 2d57f8d8944e67ed
 /deepseek-ai-dsh-client-ui-settings-models/understand-the-implementation/onboarding-coordinator:
-  en: 46dea666dffe8813
-  zh: 6f56fdc365fa3642
+  en: dc515a31212e477a
+  zh: 770ba77cd9a69f73
 /deepseek-ai-dsh-client-ui-settings-models/further-exploration:
   en: 20abd8b98b6c20e2
   zh: 6dfda7955ce5f33f

package/README.md

@@ -41,18 +41,26 @@ Host configuration `credentialOnboarding` defaults to `true`. The Electron prelo
 
 ### API keys
 
-API-key inputs start empty and use `autocomplete="new-password"` to ask browsers not to autofill saved login passwords.
+API-key inputs start empty and use `autocomplete="new-password"` to ask browsers not to autofill saved login passwords. A nonempty draft has the same eye control as masked conversation text: **Show secret** reveals the current input and **Hide secret** masks it. Changing or clearing the input masks it again. Stored credentials are never read into the input.
 
-The primary field on an editor card is a single **API key** input — the page never asks for an environment-variable name. A typed key stores write-only through `credentials.set` under the profile's reference, deriving `<ROUTE>_API_KEY` when the profile has none, and the pi-ai profile records that derivation as `apiKeyEnv`, so `cordis.patch.yml` never carries a key value. Leaving a new pi-ai provider's key blank saves a reference-free profile and preserves provider-native authentication (for example the Bedrock credential chain or Vertex ADC). A row labels API-key state with a green solid dot only when a referenced credential is confirmed configured, and with a red solid dot only when a named reference is confirmed missing. A successful Apply emits a local accessible status message without echoing secret material.
+The primary field on an editor card is a single **API key** input — the page never asks for an environment-variable name. A typed key stores write-only through `credentials.set` under the profile's reference, deriving `<ROUTE>_API_KEY` when the profile has none, and the pi-ai profile records that derivation as `apiKeyEnv`, so `cordis.patch.yml` never carries a key value. Leaving a new pi-ai provider's key blank saves a reference-free profile and preserves provider-native authentication (for example the Bedrock credential chain or Vertex ADC). The field's placeholder and help tooltip use the profile's `apiKeyEnv` when it has one; otherwise they use the credential variables of the route's `nativeAuth`, excluding `baseURLEnv` address variables, in lookup order, as the configurable-provider directory reports them. These display hints omit references containing `oauth` (case-insensitively), and describe saved credentials followed by the displayed environment fallbacks. This presentation filter does not change the [credential resolution order](../../../docs/subsystems/credentials.md#resolution), the adapter's complete environment list, or an explicit reference. A filtered explicit reference still has a usable key input with the plain placeholder and no tooltip. A native route that has no key authentication omits the key field and authentication help entirely; an explicit reference still enables the field. A profile without a reference on a route without `nativeAuth`, such as a hand-declared route, shows the plain key placeholder and no tooltip. A row labels API-key state with a green solid dot only when a referenced credential is confirmed configured, and with a red solid dot only when a named reference is confirmed missing. A successful Apply emits a local accessible status message without echoing secret material.
+
+The editor uses redacted `CredentialInfo` to choose the saved-credential placeholder or environment detection, without displaying a credential-source label. Refreshing the snapshot updates an open editor without a separate probe. The hint never receives a credential value or changes field writability.
 
 ### Editing a provider
 
-The collapsed 自定义设置 fold carries the curated extras: `baseURL` for both families (the deepseek placeholder shows the public endpoint), each adapter's model catalog, and the **display name** and **API protocol** of a pi-ai route the adapter does not ship. Profile `headers` remain deployment configuration in `cordis.patch.yml` or Cordis config and have no Models-page editor. The Provider ID stays fixed: it is the settings key, the name every other namespace and every logged session references, and the stem of a credential reference the page cannot read back to move. Reasoning effort is deliberately not among the editable fields: it is a per-model capability, so a provider-scoped control could only be set to a value some models reject. Each model row edits `id`, optional display `name`, optional `contextWindow`/`maxTokens`, and input types; unrelated model fields survive edits.
+A blank key field offers a saved credential when one is detected, with no status text below the input; otherwise it offers environment authentication. Redacted `credentials.describe` checks distinguish a detected environment reference, no detected variables, and an unconfirmed read. A missing-variable hint uses the warning color and names the first variable to configure. A detected environment reference uses green. Address configuration variables do not count as credentials. Hidden authentication references still participate in detection; a generic detected hint keeps their names hidden. Detection does not validate credentials, test a connection, or inspect provider-managed files such as AWS profiles and Google application-default credentials. A nonempty input shows no authentication status text. The page reads native variable names as well as named references, in batches of at most 64, and never reads their values into the Client.
+
+An endpoint without a fixed catalog default shows the provider's `baseURLEnv` configuration names below the input. A profile endpoint remains the first placeholder choice.
+
+The collapsed 自定义设置 fold carries the curated extras: `baseURL` for both families (the placeholder shows the endpoint the profile resolves to, the public endpoint for deepseek, else the `defaultBaseURL` the configurable-provider directory reports for a pi-ai catalog route), each adapter's model catalog, and the **display name** and **API protocol** of a pi-ai route the adapter does not ship. Profile `headers` remain deployment configuration in `cordis.patch.yml` or Cordis config and have no Models-page editor. The Provider ID stays fixed: it is the settings key, the name every other namespace and every logged session references, and the stem of a credential reference the page cannot read back to move. There is no provider-scoped reasoning-effort field: effort is a per-model capability, so a provider-scoped control could only be set to a value some models reject. Each model row edits `id`, optional display `name`, optional `contextWindow`/`maxTokens`, input types, and, on a hand-declared pi-ai route, reasoning levels; unrelated model fields survive edits.
 
 The DeepSeek card edits the `llm-deepseek` endpoint, credentials, and model catalog. It uses Messages, with the default endpoint placeholder `https://api.deepseek.com/anthropic`.
 
 Expand **Customized settings → Model options** to edit each model. Both provider families share the same row layout, labels, and icons: context window and max output tokens occupy two columns, and **Input types** occupies a separate row with **Text** and **Image** checkboxes. A row without an input declaration displays the installed model’s input types, then the provider default, then Text. Known pi-ai providers load their installed catalog without endpoint I/O; opening a row does not write an override. Explicit input selections take precedence, including text-only overrides of vision models. Checkbox edits save the selected types, with at least one type required. DeepSeek writes `inputModalities`; pi-ai writes `input`. Unchecking Image for DeepSeek also removes `imagePixelBudget` and `imageMaxBytes`, because the adapter rejects those limits without image input. Clearing the input field in `cordis.patch.yml` restores adapter inheritance; **Restore default models** resets the entire catalog override. Declare only input types the upstream model can actually process.
 
+A model row on a hand-declared pi-ai route also edits its **Reasoning levels**, the `reasoningEfforts` the adapter already reads from `cordis.patch.yml`; the **?** beside the label explains the two columns. With the switch off the field is absent, so the model does not reason and the model picker offers no level; a stored `false` is kept until the switch is used. Turning it on always starts from rows for off, low, medium, and high; it does not restore levels removed earlier, and **Cancel** discards the edit instead. Each row pairs a level select — the keys `llm-pi-ai`'s configuration schema accepts in `reasoningEfforts`, read from the namespace schema like the protocol choices and today `off`, `minimal`, `low`, `medium`, `high`, `xhigh`, and `max`, each usable once — with the **API value** sent for it, which starts as the level name; a field cleared while typing stores the level name and shows it once the field is left. Off only turns reasoning off, so its value field is disabled; a value already written for it in `cordis.patch.yml` is shown and kept. Changing a row's level resets a value still spelled as the old level; a custom value carries over only between two thinking levels, so changing a row to or from Off resets it. A stored key the schema does not list keeps its row and stays selectable there. Rows can be added and deleted and are saved in row order, while the model picker lists the levels in escalation order. The section refuses what the adapter refuses: no level besides Off — including a `reasoningEfforts:` left without a value, which shows the switch on with no rows — or a value that is missing on a level besides Off or empty on any level. A refused value marks its field invalid, and leaving that field fills in the level name. The rows keep no state of their own, so a model row reused after a deletion, a reset, or a fold shows only its own model. Routes served from the installed catalog and the DeepSeek card keep the adapter's levels and show no reasoning control.
+
 ### Adding and deleting providers
 
 **Add model provider** is the one entry to the add card, offered while any editor-backed settings namespace is mounted and enabled while either mode below can proceed. The card opens on a segmented mode switch with a one-line purpose beneath it: **Third-party model provider** carries the dormant-directory provider select over the provider editor — a bare-mounted `llm-pi-ai` offers its whole installed catalog before any route exists — and **Custom model API** carries the create form for a route pi-ai does not ship: a relay, a self-hosted server, or any other OpenAI- or Anthropic-compatible endpoint. A mode is offered only while its namespace is mounted; with one mode the card shows that form alone under the mode's name, and a mode whose namespace has nothing left to adopt or no protocol to declare is disabled with the reason as its hover text. A panel mounts the first time its mode is shown and stays mounted, hidden, while the card is open and its mode stays offered, so switching modes discards neither draft; the switch locks while either panel has a write or an endpoint interrogation in flight, since a switch underneath one would orphan its answer. Each tab controls its panel through ids, and closing either panel forgets the catalog draft's target. The custom form asks for a unique **Provider ID**, an endpoint, a protocol, and at least one uniquely-identified model, because nothing can default those. The protocol picker names each protocol by its product name — OpenAI Chat Completions, OpenAI Responses, Anthropic Messages — while the stored value stays the schema identifier; a protocol the adapter adds before this page names it shows that identifier. The endpoint placeholder follows the selected protocol: OpenAI Chat Completions and Responses show `https://gateway.example/v1`, while Anthropic Messages shows `https://gateway.example` because its SDK appends `/v1/messages`. Changing the protocol preserves any address already entered. The endpoint must be a parseable HTTP or HTTPS URL; localhost, IPv4 and IPv6 literals, and custom ports remain valid. A syntax error blocks both discovery and creation at the field, while a request failure remains a separate provider error. **Fetch available models** asks the `llm/discoverModels` Remote about the endpoint the form shows, so adding a provider is one pass instead of save-then-return; the reply opens a searchable picker rather than being written, and nothing is written until **Add selected**. Each candidate displays its original model id in a monospace font to distinguish models with the same name. Rows stay on one line; hovering reveals the model name, falling back to the full id when no name is supplied. Selection and configuration retain the original model id. Each selected candidate copies its id, display name, context window, output-token cap, and disclosed input types into the editable row when disclosed, while an existing row retains its user-tuned values. Search matches model ids and optional display names without clearing hidden selections. **Select all** adds the visible results, while **Deselect all** clears the entire selection so hidden results cannot be adopted accidentally. A row is deletable only when the user layer alone carries it (removal restores the composition base), and its confirmation dialog names the provider.
@@ -65,7 +73,7 @@ After the versioned notice step completes, the DeepSeek step projects first-run
 
 The section declares two seats for plugins distributed outside this repository, typed in [`src/client/slot-contract.ts`](src/client/slot-contract.ts) and exported from `./client`. `settings.models.provider-card` (keyed) renders inside every card that shows a directory row — a saved row's card, its first-run setup posture, and the add-provider draft — dispatched with `entryKey = settingsNs` and owner props carrying the row's `ConfigurableProviderView`, its configured state, and its confirmed api-key credential state, so one registration under an adapter family's namespace receives every card of that family, hand-declared routes included; the hand-declared draft card has no directory row yet and dispatches nothing until saved. `settings.models.footer` (list) renders after the rows and the add controls. A registrant activates through `ctx.slots.inject` with a type-only import of this package's `/client` entry; without registrants both seats render nothing.
 
-The Models page includes **DeepSeek Account** (`deepseek-account`, **DeepSeek 账号** in Chinese). Its editor exposes the shared DeepSeek model catalog without API-key or base-URL inputs, and saves that catalog under the account route's own settings section (`llm-deepseek-account` by default), so an account edit never rewrites the `llm-deepseek` section the official route reads. The account row is hidden when its available model catalog is empty, including before sign-in and after sign-out; it returns when account models become available.
+The Models page includes **DeepSeek Account** (`deepseek-account`, **DeepSeek 账号** in Chinese). The row shows a **Signed in** status and offers no editor: the account listing supplies its models, and [model overrides](../../llm/llm-deepseek-account/README.md#use-this-package) live in `cordis.patch.yml`. The account row is hidden when its available model catalog is empty, including before sign-in and after sign-out; it returns when account models become available.
 
 -----
 
@@ -89,7 +97,7 @@ Concurrent `ModelsSettingsStore.load()` callers resolve after the current refres
 
 ### Onboarding coordinator
 
-The notice step owns its exact copy in `src/client/locales.ts` and its acknowledgement version in `src/onboarding-copy.ts`; on loopback it compares and writes `ui-settings-general.welcomeNoticeVersion` through the shared configuration form, and only an explicit Continue records the current version. Users who acknowledged an earlier version see the current notice again. A non-loopback browser cannot use that Host-only namespace, so acknowledgement is process-local and the notice returns after reload. The DeepSeek step targets `deepseek-official` in `llm-deepseek` and renders the existing `ProviderEditor` in credential-only mode inside the shared onboarding modal; `credentials.set` stays the only secret write, and no provider settings are changed.
+The notice step owns its exact copy in `src/client/locales.ts` and its acknowledgement version in `src/onboarding-copy.ts`. Authenticated browsers compare and write `ui-settings-general.welcomeNoticeVersion` through the shared configuration form; only an explicit Continue records the current version, and an unavailable namespace or a refused write leaves the notice unacknowledged. Users who acknowledged an earlier version see the current notice again. The DeepSeek step targets `deepseek-official` in `llm-deepseek` and renders the existing `ProviderEditor` in credential-only mode inside the shared onboarding modal; `credentials.set` stays the only secret write, and no provider settings are changed.
 
 </details>
 

package/README.zh.md

@@ -41,18 +41,26 @@ Host 配置 `credentialOnboarding` 默认为 `true`。Electron preload 标记会
 
 ### API 密钥
 
-API 密钥输入框初始为空，并通过 `autocomplete="new-password"` 请求浏览器不要自动填入已保存的登录密码。
+API 密钥输入框初始为空，并通过 `autocomplete="new-password"` 请求浏览器不要自动填入已保存的登录密码。输入非空时，显示与会话密钥遮罩一致的眼睛按钮：**显示密钥**展示当前输入，**隐藏密钥**重新遮住输入。修改或清空输入后会重新遮住密钥。已保存凭据从不读入输入框。
 
-编辑卡片上的主字段是单独一个 **API 密钥**输入框——页面从不询问环境变量名。键入的密钥经 `credentials.set` 以**只写**方式存入 profile 的引用之下，profile 没有引用时便派生 `<ROUTE>_API_KEY`，pi-ai profile 会把这次派生记录为 `apiKeyEnv`，因此 `cordis.patch.yml` 从不携带密钥值。为新的 pi-ai 提供商留空密钥会保存一个不带引用的 profile，从而保留提供商原生认证（例如 Bedrock 凭据链或 Vertex ADC）。只有确认引用的凭据已配置时，行才会以绿色实心点标示 API 密钥状态；只有确认具名引用缺失时，才会以红色实心点标示。「应用」成功后会发出本地无障碍状态消息，且绝不回显任何机密内容。
+编辑卡片上的主字段是单独一个 **API 密钥**输入框——页面从不询问环境变量名。键入的密钥经 `credentials.set` 以**只写**方式存入 profile 的引用之下，profile 没有引用时便派生 `<ROUTE>_API_KEY`，pi-ai profile 会把这次派生记录为 `apiKeyEnv`，因此 `cordis.patch.yml` 从不携带密钥值。为新的 pi-ai 提供商留空密钥会保存一个不带引用的 profile，从而保留提供商原生认证（例如 Bedrock 凭据链或 Vertex ADC）。profile 有 `apiKeyEnv` 时，字段的占位文字与帮助提示采用该引用；否则按查找顺序采用可配置提供商目录为该路由报告的 `nativeAuth` 凭据变量，并排除 `baseURLEnv` 地址变量。这些显示提示会忽略包含 `oauth` 的引用（不区分大小写），并说明优先使用已保存凭据，其后为显示的环境后备。这项显示过滤不改变[凭据解析顺序](../../../docs/subsystems/credentials.zh.md#resolution)、适配器的完整环境变量列表或显式引用。被过滤的显式引用仍有可用的密钥输入框，使用普通占位文字且不显示帮助提示。没有密钥认证的原生路由不显示密钥字段及认证帮助；显式引用仍会启用该字段。路由没有 `nativeAuth`（例如手动声明的路由）且 profile 没有引用时，显示普通的密钥占位文字，且不显示帮助提示。只有确认引用的凭据已配置时，行才会以绿色实心点标示 API 密钥状态；只有确认具名引用缺失时，才会以红色实心点标示。「应用」成功后会发出本地无障碍状态消息，且绝不回显任何机密内容。
+
+编辑器使用脱敏 `CredentialInfo` 选择已保存凭据的占位文字或环境检测，不显示凭据来源标签。快照刷新会更新打开的编辑器，无需单独探测。提示从不接收凭据值，也不改变字段的可写性。
 
 ### 编辑提供商
 
-收起的「自定义设置」折叠区承载精选的额外字段：两个家族都有 `baseURL`（deepseek 的占位符显示公共端点）、各适配器自己的模型目录，以及适配器未提供的 pi-ai 路由的**显示名称**与 **API 协议**。Profile `headers` 仍是 `cordis.patch.yml` 或 Cordis 配置中的部署配置，Models 页面不提供编辑器。Provider ID 保持固定：它是 settings 的键、其他每个 namespace 与每一条已记录会话引用的名字，也是页面读不回、因而搬不走的凭据引用词干。推理等级刻意不在可编辑字段之列：它是按模型的能力，提供商级的控件只可能被设成某些模型会拒绝的值。每个模型行可编辑 `id`、可选显示 `name`、可选 `contextWindow`/`maxTokens` 和输入类型；无关的模型字段在编辑后仍会保留。
+密钥字段留空时，已检测到已保存凭据则提示使用该凭据，输入框下方不显示状态文字；否则提示使用环境认证。脱敏的 `credentials.describe` 检查区分已检测到环境引用、未检测到变量和读取状态未确认。缺少变量时使用黄色提示，并列出应配置的首个变量名。已检测到环境引用时使用绿色。地址配置变量不计作凭据。隐藏的认证引用仍参与检测；使用不含变量名的通用检测提示。检测不验证凭据、不测试连接，也不检查 AWS profile 或 Google 应用默认凭据等提供商管理的文件。输入非空时，不显示认证状态文字。页面读取原生变量名和具名引用，每批最多 64 项，不将变量值读入 Client。
+
+目录未提供固定默认端点时，输入框下方显示提供方的 `baseURLEnv` 配置变量名。Profile 中的端点仍是占位符的首选。
+
+收起的「自定义设置」折叠区承载精选的额外字段：两个家族都有 `baseURL`（占位符显示 profile 解析出的端点，deepseek 即公共端点；没有时显示可配置提供商目录为 pi-ai 内置路由报告的 `defaultBaseURL`）、各适配器自己的模型目录，以及适配器未提供的 pi-ai 路由的**显示名称**与 **API 协议**。Profile `headers` 仍是 `cordis.patch.yml` 或 Cordis 配置中的部署配置，Models 页面不提供编辑器。Provider ID 保持固定：它是 settings 的键、其他每个 namespace 与每一条已记录会话引用的名字，也是页面读不回、因而搬不走的凭据引用词干。这里没有提供商级的推理等级字段：它是按模型的能力，提供商级的控件只可能被设成某些模型会拒绝的值。每个模型行可编辑 `id`、可选显示 `name`、可选 `contextWindow`/`maxTokens`、输入类型，以及手工声明的 pi-ai 路由上的推理等级；无关的模型字段在编辑后仍会保留。
 
 `llm-deepseek` 的 DeepSeek 卡片编辑端点、凭据和模型目录。它使用 Messages，默认端点占位符为 `https://api.deepseek.com/anthropic`。
 
 展开**自定义设置 → 模型选项**编辑模型。两个提供商家族共用相同的模型行布局、标签和图标：上下文窗口与最大输出 token 数分为两列，**输入类型**独占下一行，提供**文本**和**图片**复选框。未声明输入类型的模型行优先显示已安装模型的输入类型，其次是提供商默认值，最后回退为文本。已知 pi-ai 提供商会加载已安装目录，不向端点发送请求；打开模型行不会写入覆盖值。显式输入选择具有优先权，包括为视觉模型设置的仅文本覆盖。修改复选框会保存所选类型，且至少保留一种。DeepSeek 写入 `inputModalities`，pi-ai 写入 `input`。DeepSeek 取消勾选图片时，还会移除 `imagePixelBudget` 和 `imageMaxBytes`，因为适配器在没有图片输入时拒绝这些限制。在 `cordis.patch.yml` 中清除输入字段可恢复适配器继承；**恢复默认模型**会重置整个模型目录覆盖。仅声明上游模型实际能够处理的输入类型。
 
+手工声明的 pi-ai 路由上，模型行还会编辑**推理等级**，即适配器本来就从 `cordis.patch.yml` 读取的 `reasoningEfforts`；标签旁的 **?** 说明两列的含义。开关关闭时该字段不存在，模型不推理，模型选择器不提供任何档位；已存储的 `false` 在开关被操作前保持不变。打开开关总是从 off、low、medium 和 high 四行开始，不会恢复之前删掉的档位；要放弃修改请点**取消**。每行把一个档位选择框——可选值是 `llm-pi-ai` 配置 schema 允许作为 `reasoningEfforts` 键的档位，与协议选项一样从命名空间 schema 读取，目前为 `off`、`minimal`、`low`、`medium`、`high`、`xhigh`、`max`，每个档位只能用一次——与发送给它的 **API 参数值**配成一对，参数值初始为档位名；输入时清空的字段会存为档位名，离开字段后显示出来。Off 只表示关闭思考，因此它的参数值输入框是禁用的；`cordis.patch.yml` 中已为它写好的值会显示并保留。修改某行的档位时，仍是旧档位名的参数值会随之重置；自定义值只在两个思考档位之间保留，切到或切离 Off 都会重置。schema 未列出的已存键会保留自己的行，并可在该行中继续选择。行可以增删，并按行的顺序保存，模型选择器则按升级顺序列出档位。本节拒绝适配器会拒绝的配置：没有 Off 以外的档位（包括未写值的 `reasoningEfforts:`，此时开关显示为打开且没有任何行），或者 Off 以外的档位缺少参数值、任意档位的参数值为空字符串。被拒绝的值会把对应输入框标为无效，离开该输入框时会填入档位名。各行不保留自己的状态，因此删除、恢复默认或折叠之后被复用的模型行只显示它自己的模型。走已安装目录的路由和 DeepSeek 卡片沿用适配器的档位，不显示推理等级控件。
+
 ### 新增与删除提供商
 
 **添加模型提供商**是新增卡片的唯一入口：只要任一带编辑器的设置 namespace 已挂载就会出现，下述两种方式任一可继续时才可点击。卡片打开后顶部是分段式的方式切换，下方一行小字说明所选方式的用途：**第三方模型提供商**承载休眠目录提供商选择框与提供商编辑器——裸挂载的 `llm-pi-ai` 在任何路由存在之前就能提供其完整的已安装 catalog；**自定义模型 API**承载为 pi-ai 不提供的路由（中转站、自部署服务或其他兼容 OpenAI / Anthropic 协议的接口）准备的创建表单。某种方式只在其 namespace 已挂载时提供；只剩一种方式时卡片以该方式为标题直接显示表单；namespace 已无可采用的提供商或没有可声明的协议时，对应方式会被禁用并把原因作为悬停提示。面板在其方式首次显示时挂载，之后在卡片打开且该方式仍被提供期间保持挂载但隐藏，因此切换方式不会丢掉任何一边的草稿；任一面板有写入或端点探测进行中时滑块锁定，因为此时切换会让结果落到看不见的面板上。每个 tab 通过 id 控制其面板，关闭任一面板都会忘记目录草稿的目标。自定义表单会索要唯一的 **Provider ID**、端点、协议与至少一个可唯一识别的模型，因为没有东西能为它们兜底。协议选择框以产品名称显示各协议——OpenAI Chat Completions、OpenAI Responses、Anthropic Messages——存储的值仍是 schema 标识符；适配器新增而本页尚未命名的协议直接显示其标识符。端点占位示例随所选协议变化：OpenAI Chat Completions 与 Responses 显示 `https://gateway.example/v1`，Anthropic Messages 显示 `https://gateway.example`，因为其 SDK 会追加 `/v1/messages`。切换协议会保留已输入的地址。端点必须是可解析的 HTTP 或 HTTPS URL；localhost、IPv4 与 IPv6 字面地址以及自定义端口仍然有效。语法错误会在字段处阻止询问与创建，请求失败则继续作为独立的提供商错误显示。**获取可用模型**通过 `llm/discoverModels` Remote 查询表单显示的端点，因此新增提供商一次即可完成，而非先保存再返回；回复打开的是可搜索选择器而非直接写入，只有点击**添加所选**才会写入。候选项使用等宽字体显示原始模型 id，以区分同名模型。每行保持单行，悬停可查看模型名称，未提供名称时回退到完整 id。勾选与配置仍使用原始模型 id。每个选中候选会在提供商公布相应信息时，把 id、显示名、上下文窗口、最大输出 token 数和已公布的输入类型复制进可编辑行；已经存在的行保留用户调整过的值。搜索会匹配模型 id 与可选显示名称，且不会清除隐藏项的勾选状态。**全选**会加入可见结果，而**取消全选**会清空全部勾选，以免意外采用隐藏结果。只有用户层单独携带某行时，该行才可删除（删除会恢复组合基线），其确认对话框会指名该提供商。
@@ -65,7 +73,7 @@ API 密钥输入框初始为空，并通过 `autocomplete="new-password"` 请求
 
 本分区为仓库外分发的插件声明两个席位，类型定义在 [`src/client/slot-contract.ts`](src/client/slot-contract.ts) 并从 `./client` 导出。`settings.models.provider-card`（keyed）渲染在每张展示目录行的卡片内部——已保存行的卡片、其首次运行 setup 形态、以及「添加提供商」草稿卡——以 `entryKey = settingsNs` 分发，owner props 携带该行的 `ConfigurableProviderView`、其 configured 状态与已确认的 api-key 凭据状态，因此以某适配器家族的 namespace 注册一次即可收到该家族的全部卡片，含手工声明的路由；手工声明的草稿卡尚无目录行，保存之前不分发。`settings.models.footer`（list）渲染在行列表与新增控件之后。注册方通过 `ctx.slots.inject` 激活，并以 type-only import 引入本包 `/client` 入口；没有注册方时两个席位均不渲染任何内容。
 
-Models 页面包含 **DeepSeek 账号**（`deepseek-account`，英文为 **DeepSeek Account**）。其编辑器展示共享的 DeepSeek 模型目录，不提供 API Key 或 Base URL 输入框；目录保存到账号路由自己的设置段（默认为 `llm-deepseek-account`），因此账号侧的编辑不会改写 official 路由读取的 `llm-deepseek` 段。账号可用模型目录为空时隐藏账号行，包括登录前和退登后；账号模型恢复可用时重新显示。
+Models 页面包含 **DeepSeek 账号**（`deepseek-account`，英文为 **DeepSeek Account**）。该行显示 **已登录** 状态，不提供编辑器：模型来自账号的模型列表，[模型覆盖](../../llm/llm-deepseek-account/README.zh.md#use-this-package)写在 `cordis.patch.yml` 中。账号可用模型目录为空时隐藏账号行，包括登录前和退登后；账号模型恢复可用时重新显示。
 
 -----
 
@@ -89,7 +97,7 @@ Models 页面包含 **DeepSeek 账号**（`deepseek-account`，英文为 **DeepS
 
 ### 引导协调器
 
-预览版说明在 `src/client/locales.ts` 中持有精确文案，并在 `src/onboarding-copy.ts` 中持有确认版本；回环时它通过既有 settings API 比较并写入 `ui-settings-general.welcomeNoticeVersion`，且只有显式点击「继续」才会记录当前版本。已确认旧版的用户会再次看到当前说明。非回环浏览器无法使用这个仅限宿主的 namespace，因此确认只保留在进程内，刷新后说明会再次出现。DeepSeek 步骤面向 `llm-deepseek` 中的 `deepseek-official`，在共享引导模态框内以仅凭据模式渲染既有 `ProviderEditor`；`credentials.set` 仍是唯一的机密写入，且不改变任何提供商设置。
+预览版说明在 `src/client/locales.ts` 中持有精确文案，并在 `src/onboarding-copy.ts` 中持有确认版本。认证浏览器通过共享配置表单比较并写入 `ui-settings-general.welcomeNoticeVersion`；只有显式点击「继续」才会记录当前版本；命名空间不可用或写入被拒时，声明保持未确认状态。已确认旧版的用户会再次看到当前说明。DeepSeek 步骤面向 `llm-deepseek` 中的 `deepseek-official`，在共享引导模态框内以仅凭据模式渲染既有 `ProviderEditor`；`credentials.set` 仍是唯一的机密写入，且不改变任何提供商设置。
 
 </details>