🚀 feat: launch v1.0 — next-generation frontend built from the ground up (#4265)

* feat: add parameter coverage for the operations: copy, trim_prefix, trim_suffix, ensure_prefix, ensure_suffix, trim_space, to_lower, to_upper, replace, and regex_replace

* fix: CrossGroupRetry default false

移除gorm:"default:false"，避免每次 AutoMigrate时都执行ALTER TABLE `tokens` MODIFY COLUMN `cross_group_retry` boolean DEFAULT false
且bool默认false不影响原有功能

* feat: check-in feature integrates Turnstile security check

* feat: add support for Doubao /v1/responses (#2567)

* feat: add support for Doubao /v1/responses

* fix: fix model deployment style issues, lint problems, and i18n gaps. (#2556)

* fix: fix model deployment style issues, lint problems, and i18n gaps.

* fix: adjust the key not to be displayed on the frontend, tested via the backend.

* fix: adjust the sidebar configuration logic to use the default configuration items if they are not defined.

* feat: add plans directory to .gitignore

* fix: 修复 gemini 文件类型不支持 image/jpg

* fix: fix the proxyURL is empty, not using the default HTTP client configuration && the AWS calling side did not apply the relay timeout.

* fix: batch add key backend deduplication

* Merge pull request #2582 from seefs001/fix/tips

fix: add tips for model management and channel testing

* fix(gin): update request body size check to allow zero limit

* feat: add regex pattern to mask API keys in sensitive information

* fix(task): 修复使用 auto 分组时 Task Relay 不记录日志和不扣费的问题

问题描述：
- 使用 auto 分组的令牌调用 /v1/videos 等 Task 接口时，虽然任务能成功创建，
  但使用日志不显示记录，且不会扣费

根本原因：
- Distribute 中间件在选择渠道后，会将实际选中的分组存储在 ContextKeyAutoGroup 中
- 但 RelayTaskSubmit 函数没有从 context 中读取这个值来更新 info.UsingGroup
- 导致 info.UsingGroup 始终是 "auto" 而不是实际选中的分组（如 "sora2逆"）
- 当 auto 分组的倍率配置为 0 时，quota 计算结果为 0
- 日志记录条件 "if quota != 0" 不满足，导致日志不记录、不扣费

修复方案：
- 在 RelayTaskSubmit 函数中计算分组倍率之前，添加从 ContextKeyAutoGroup
  获取实际分组的逻辑
- 使用安全的类型断言，避免潜在的 panic 风险

影响范围：
- 仅影响 Task Relay 流程（/v1/videos, /suno, /kling 等接口）
- 不影响使用具体分组令牌的调用
- 不影响其他 Relay 类型（chat/completions 等已有类似处理逻辑）

* 🚀 feat(web): port legacy v2 frontend changes into new UI (deployments, check-in, ollama) + align APIs

Bring over the key frontend functionality introduced in merge `efa3301` and integrate it cleanly into the new `web/src` architecture and design system.

- **Model deployments (io.net)**
  - Align frontend endpoints and payloads with backend deployment routes (`/api/deployments/*`)
  - Add missing deployment operations: details, logs (container-aware), update config, rename, extend duration
  - Improve create-deployment flow (proper request shape, name availability check, price estimation parity)

- **System settings**
  - Enhance io.net deployment settings: allow testing connection with an unsaved API key and add “how to get API key” guidance

- **Channels / Ollama**
  - Improve Ollama model management: live fetch via base_url with fallback to channel fetch, selection + apply flows, delete confirmation
  - Refactor for feature-layer consistency: extract Ollama parsing/normalization utilities into `features/channels/lib`

- **Quality**
  - Ensure TypeScript typecheck passes after refactor and new dialogs/components integration

* Merge pull request #2590 from xyfacai/fix/max-body-limit

fix: 设置默认max req body 为128MB

* docs: update readme

* i18n: add missing translations

* fix(gemini): fetch model list via native v1beta/models endpoint

Use the native Gemini Models API (/v1beta/models) instead of the OpenAI-compatible
path when listing models for Gemini channels, improving compatibility with
third-party Gemini-format providers that don't implement OpenAI routes.

- Add paginated model listing with timeout and optional proxy support
- Select an enabled key for multi-key Gemini channels

* refactor(gemini): 更新 GeminiModelsResponse 以使用 dto.GeminiModel 类型

* fix: remove Minimax from FETCHABLE channels

* fix(minimax): 添加 MiniMax-M2 系列模型到 ModelList

* feat: add doubao video 1.5

* 🤢 chore: remove useless file

* feat: /v1/chat/completion -> /v1/response (#2629)

* feat: /v1/chat/completion -> /v1/response

* fix: clean propertyNames for gemini function

* fix: support snake_case fields in GeminiChatGenerationConfig

* chore: update dependencies and lockfile for improved compatibility

- Updated @clerk/clerk-react to version 5.59.3
- Updated @hookform/resolvers to version 5.2.2
- Updated @lobehub/icons to version 2.48.0
- Updated various Radix UI components to their latest versions
- Updated @tanstack/react-query and related packages for better performance
- Updated axios, i18next, and other libraries for security and feature enhancements
- Updated lockfile to include configVersion and ensure consistency across environments

* Merge pull request #2647 from seefs001/feature/status-code-auto-disable

feat: status code auto-disable configuration

* fix: chat2response setting ui (#2643)

* fix: setting ui

* fix: rm global.chat_completions_to_responses_policy

* fix: rm global.chat_completions_to_responses_policy

* Merge pull request #2627 from seefs001/feature/channel-test-param-override

feat: channel testing supports parameter overriding

* chore: update dependencies and lockfile for improved compatibility

- Updated @lobehub/icons to version 4.0.3
- Updated ai to version 6.0.27
- Updated various libraries including axios, react-day-picker, and streamdown for security and feature enhancements
- Updated devDependencies for eslint, prettier, and typescript for better performance and compatibility
- Updated lockfile to ensure consistency across environments

* chore: update lockfile and Vite configuration for improved build process

- Updated lockfile to version 1 for better compatibility and consistency
- Enhanced Vite configuration to support production optimizations, including code minification and chunking for dependencies
- Added environment-specific console and debugger removal for production builds

* chore: migrate from Vite to Rsbuild for build process

- Added Rsbuild configuration for development and production builds
- Updated package.json scripts to use Rsbuild instead of Vite
- Replaced @tailwindcss/vite with @tailwindcss/postcss in dependencies
- Introduced postcss.config.mjs for Tailwind CSS integration
- Updated TypeScript configuration to include Rsbuild config
- Removed Vite configuration file to streamline the build process

* refactor: optimize user data handling and API calls

- Replaced direct API calls to get user data with cached user information from auth-store in ModelsFilter and SummaryCards components.
- Improved session management in RootComponent and Authenticated route to utilize localStorage for user authentication status, reducing unnecessary API requests.
- Added caching for setup status checks to enhance performance during navigation.

* feat: enhance session validation in authenticated route

- Implemented session verification to check user authentication status via API call only once per session.
- Updated beforeLoad logic to redirect users to the login page if session validation fails or if no user information is available in localStorage.
- Improved user data handling by updating the auth store with fresh user information upon successful session verification.

* refactor: improve useMediaQuery hook for better SSR handling

- Enhanced the useMediaQuery hook to check for window availability before accessing matchMedia, preventing errors during server-side rendering.
- Simplified state initialization and change handling by using a dedicated function to determine initial matches.
- Updated event listener management for improved performance and clarity.

* feat(hooks): export useMediaQuery from hooks index

* refactor: update useMediaQuery imports to use unified hooks index

* fix(rsbuild): fix loadEnv API usage and removeConsole type

* feat: customizable automatic retry status codes

* refactor(hooks): use useSyncExternalStore for better SSR handling in useMediaQuery

* refactor: simplify embedded file structure in main.go

- Updated the embedded file directive to include the entire web/dist directory instead of individual assets, streamlining the build process.

* refactor: replace DropdownMenu with Sheet component in ProfileDropdown

- Updated the ProfileDropdown component to use a Sheet for user interactions instead of a DropdownMenu.
- Enhanced user info display with improved layout and styling.
- Added navigation links and sign-out functionality within the Sheet.

* refactor: streamline ProfileDropdown layout and improve user info display

- Removed unused Badge component and secondary text from user display.
- Enhanced styling for user info section and navigation links.
- Updated sign-out functionality to use a button for better accessibility.

* feat: add System Settings link for super admin in ProfileDropdown

- Introduced a new link to System Settings in the ProfileDropdown, visible only to users with the SUPER_ADMIN role.
- Updated imports to include the Settings icon and adjusted the component logic accordingly.
- Removed the Settings entry from the sidebar data to streamline navigation.

* feat: codex channel (#2652)

* feat: codex channel

* feat: codex channel

* feat: codex oauth flow

* feat: codex refresh cred

* feat: codex usage

* fix: codex err message detail

* fix: codex setting ui

* feat: codex refresh cred task

* fix: import err

* fix: codex store must be false

* fix: chat -> responses tool call

* fix: chat -> responses tool call

* feat(i18n): add missing translations

* fix(i18n): restore missing translations for "360" and add "User Menu" in multiple locales

- Reintroduced the translation for "360" in English, French, Japanese, Russian, Vietnamese, and Chinese locales.
- Added the "User Menu" translation in the same languages to enhance user interface consistency.

* fix: openAI function to gemini function field adjusted to whitelist mode

* feat: TLS_INSECURE_SKIP_VERIFY env

* fix: for chat-based calls to the Claude model, tagging is required. Using Claude's rendering logs, the two approaches handle input rendering differently.

* refactor(system-settings): restructure settings sections and navigation

- Replaced SettingsAccordion with a unified SettingsSection component across various settings sections for consistency.
- Introduced a section registry to manage general settings sections dynamically.
- Updated navigation items in the system settings sidebar to utilize the new section registry.
- Enhanced the GeneralSettings component to support section-based content rendering based on user selection.

* fix(system-settings): remove type assertion for quotaDisplayType in GeneralSettings

- Eliminated the type assertion for quotaDisplayType in the GeneralSettings component to improve type inference and maintain cleaner code.

* refactor(system-settings): update zod import syntax in general settings

- Changed the import statement for zod from a default import to a namespace import for better clarity and consistency in the codebase.

* fix: the login method cannot be displayed under the aff link.

* feat(system-settings): implement generic settings page and enhance navigation

- Added a new generic SettingsPage component to handle loading states, data fetching, and section rendering.
- Integrated section registry for general and authentication settings to streamline navigation and content management.
- Updated URL utility functions to improve query parameter handling for active navigation states.
- Enhanced the system settings sidebar to include authentication section items dynamically.

* refactor(system-settings): replace SettingsAccordion with SettingsSection across authentication settings

- Updated BasicAuthSection, BotProtectionSection, OAuthSection, and PasskeySection to use the new SettingsSection component for consistency.
- Introduced a section registry to manage authentication settings dynamically, enhancing navigation and content rendering.

* feat(system-settings): enhance request limits settings with new section and unified component

- Added a new Request Limits section to the system settings sidebar, integrating it with the section registry for improved navigation.
- Replaced SettingsAccordion with SettingsSection in RateLimitSection, SensitiveWordsSection, and SSRFSection for consistency.
- Updated RequestLimitsSettings to utilize the new SettingsPage component for better data handling and rendering.
- Implemented a search schema for request limits to streamline navigation and section management.

* feat(system-settings): integrate content settings sections with unified component and registry

- Added a new Content section to the system settings sidebar, incorporating it into the section registry for improved navigation.
- Replaced SettingsAccordion with SettingsSection in multiple content-related components for consistency.
- Created a section registry to manage content settings dynamically, enhancing the rendering and navigation experience.
- Updated the ContentSettings component to utilize the new section registry and streamline content display.

* feat(system-settings): enhance integrations settings with unified section registry and components

- Introduced a new section registry for integrations settings, consolidating various settings components for better organization and navigation.
- Replaced SettingsAccordion with SettingsSection in multiple integration-related components for consistency.
- Updated IntegrationSettings to utilize the new SettingsPage component, improving data handling and rendering.
- Added a new integrations section to the system settings sidebar, enhancing user experience and accessibility.

* feat(system-settings): unify model settings with new section registry and components

- Introduced a section registry for model settings, consolidating various model-related components for improved organization and navigation.
- Replaced SettingsAccordion with SettingsSection in multiple model settings components for consistency.
- Updated ModelSettings to utilize the new SettingsPage component, enhancing data handling and rendering.
- Added a new Models section to the system settings sidebar, improving user experience and accessibility.

* feat(system-settings): enhance maintenance settings with unified section registry and components

- Introduced a new section registry for maintenance settings, consolidating various maintenance-related components for improved organization and navigation.
- Replaced SettingsAccordion with SettingsSection in multiple maintenance components for consistency.
- Updated MaintenanceSettings to utilize the new section registry, enhancing data handling and rendering.
- Added a new Maintenance section to the system settings sidebar, improving user experience and accessibility.

* feat(system-settings): update section titles for improved clarity and consistency

- Renamed various section titles across content, integrations, maintenance, models, and request limits to enhance clarity and better reflect their functionalities.
- Adjusted titles such as 'Dashboard' to 'Data Dashboard', 'API Info' to 'API Addresses', and 'Update Checker' to 'System maintenance' for improved user understanding.
- Ensured consistency in naming conventions across all settings sections to streamline user experience and navigation.

* feat(nav-group): enhance collapsible menu behavior and URL matching logic

- Added controlled state management for collapsible menu items to automatically expand based on active sub-item paths.
- Updated the URL matching logic in checkIsActive to improve handling of query parameters and ensure accurate navigation state detection.
- Refactored the collapsible component to utilize the new state management, enhancing user experience in the sidebar navigation.

* feat(system-settings): update system settings navigation and redirect logic

- Changed the link in the profile dropdown to point directly to the general section of system settings with a search parameter for section identification.
- Implemented a redirect in the general settings route to ensure users are directed to the default section if no section parameter is provided, enhancing navigation consistency.

* feat(system-settings): unify route configuration for settings sections

- Refactored route configuration for various system settings sections (auth, content, general, integrations, maintenance, models, request limits) to utilize a new `createSettingsRouteConfig` function.
- This change consolidates the repetitive logic of creating search schemas and handling redirects, improving code maintainability and readability.
- Enhanced navigation by ensuring default sections are loaded when no section parameter is provided.

* feat(url-utils): enhance URL handling and matching logic

- Introduced a new utility function `urlToString` to convert various URL formats (string and object) into a standardized string format.
- Updated the `checkIsActive` function to utilize `urlToString`, improving the accuracy of URL matching and handling of query parameters.
- Refactored URL comparison logic to ensure consistent behavior across different URL types, enhancing navigation state detection.

* feat(system-settings): validate DataExportDefaultTime for improved data handling

- Introduced a new function `validateDataExportDefaultTime` to ensure the `DataExportDefaultTime` value is either 'week', 'hour', or 'day', defaulting to 'hour' for unexpected values.
- Updated the `DataExportDefaultTime` assignment in the settings section to utilize this validation function, enhancing data integrity and user experience.

* perf(system-settings): Improve the i18n of system settings content

- Changed button labels in various sections to use consistent capitalization and translation functions, enhancing user experience.
- Updated validation messages in schemas to utilize translation functions for improved internationalization support.
- Ensured all user-facing strings are properly translated, improving accessibility for non-English users.

* fix(system-settings): update ApiInfoFormValues type inference for improved schema validation

- Changed the type inference for ApiInfoFormValues to utilize ReturnType of createApiInfoSchema, ensuring accurate type representation and enhancing type safety in the API info section.

* fix(chat-settings): improve validation logic for chat settings schema

- Updated the validation logic to ensure that null values are correctly handled and that only objects are accepted as valid items in the chat settings schema.
- Simplified error handling by removing the error message from the catch block, providing a consistent user-facing message for invalid JSON strings.

* fix(system-settings): enhance validation error handling in uptime-kuma schema

- Updated the validation logic for category name, URL, and slug fields to use an object format for error messages, improving clarity and consistency in user feedback.
- Ensured that all validation messages are properly structured to enhance internationalization support.

* fix(i18n): add translations for Uptime Kuma group management

- Added English, French, Japanese, Russian, Vietnamese, and Chinese translations for "Add Uptime Kuma Group" and "Edit Uptime Kuma Group" to enhance internationalization support.
- Included validation messages for category name and slug fields across multiple languages to improve user feedback and accessibility.

* fix(system-settings): improve validation error message structure for SystemName

- Updated the validation logic for the SystemName field to use an object format for error messages, enhancing clarity and consistency in user feedback.
- This change aligns with recent improvements in internationalization support across the system settings schemas.

* perf(i18n): add new validation error message translations

- Added translations for the new validation error message "Invalid JSON format or values out of allowed range" in English, French, Japanese, Russian, Vietnamese, and Chinese.
- This update enhances internationalization support by ensuring users receive clear feedback across multiple languages.

* fix(i18n): update Japanese translation for payment method configuration message

- Corrected the Japanese translation for the message regarding payment methods configuration to use the term "メソッド" instead of "方法" for improved accuracy and consistency in user feedback.
- This change enhances the clarity of the message for Japanese-speaking users.

* fix(i18n): remove unnecessary loading messages from French translations

- Removed the French translations for "Loading settings...", "Loading maintenance settings...", and "Loading content settings..." to streamline the localization file.
- This change improves the clarity and relevance of the translations provided to users.

* fix(i18n): add translations for Uptime Kuma group management in multiple languages

- Added French, Japanese, Russian, Vietnamese, and Chinese translations for "Add Uptime Kuma Group" and "Edit Uptime Kuma Group" to enhance internationalization support.
- This update improves user experience by providing clear and consistent messaging across different languages.

* fix(validation): enhance pricing schema error messages and add translations

- Updated the pricing schema to include localized error messages for validation, ensuring users receive clear feedback when input values are invalid.
- Added new translations for "Exchange rate is required" and "Exchange rate must be greater than 0" in English, French, Japanese, and Chinese to improve internationalization support.
- This change enhances user experience by providing accurate and contextually relevant messages across multiple languages.

* fix: codex Unsupported parameter: max_output_tokens

* fix(model-mapping-editor): simplify JSON parsing logic in useEffect

* fix: jimeng i2v support multi image by metadata

* refactor(models): restructure models section handling and improve UI components

- Replaced tab-based navigation with section-based navigation for better clarity and organization.
- Introduced a new section registry to manage model sections, including 'metadata' and 'deployments'.
- Updated the ModelsContent component to reflect the new section structure and added a Create Deployment button.
- Removed the ModelsTabs component as it was no longer needed.
- Enhanced internationalization support by adding new translations for section descriptions and management tasks.
- Adjusted sidebar configuration to accommodate the new section structure.

* fix: update warning threshold label from '5$' to '2$'

* fix: video content api Priority use url field

* fix: update abortWithOpenAiMessage function to use types.ErrorCode

* feat(deployment): introduce CreateDeploymentDrawer component and update dialog references

- Replaced the CreateDeploymentDialog with a new CreateDeploymentDrawer component for improved user experience.
- Added comprehensive form handling for deployment creation, including validation and price estimation features.
- Updated internationalization files to include new translations for UI elements and descriptions related to deployment configuration.
- Enhanced the ModelsContent component to integrate the new drawer for creating deployments.

* perf(i18n): enhance internationalization for models table and columns

- Updated labels and titles in the ModelsTable and useModelsColumns components to utilize translation functions for improved localization.
- Changed static text for vendor and sync status to dynamic translations, enhancing user experience for non-English speakers.
- Updated empty state messages in the ModelsTable to support internationalization, ensuring clarity for all users.

* fix: fix email send

* fix: issue where consecutive calls to multiple tools in gemini all returned an index of 0

* fix: replace Alibaba's Claude-compatible interface with the new interface

* fix: Only models with the "qwen" designation can use the Claude-compatible interface; others require conversion.

* feat: log shows request conversion

* feat: optimized display

* feat: optimized display

* feat: optimized display

* fix: codex rm Temperature

* Revert "fix: video content api Priority use url field"

* feat: requestId time string use UTC

* feat(qwen): support qwen image sync image model config

* feat: sync old ui

* feat: more ui sync

* feat: replace theme

* fix build

* refactor(web): revert theme colors and variables in CSS

Updated color variables for light and dark themes to improve consistency and visual appeal.

* feat(deployment): enhance deployment access guard and model deployment settings

- Introduced loading phase management in the DeploymentAccessGuard component to provide better user feedback during connection checks.
- Updated the ModelsContent component to prefetch the deployments list while checking connection status, improving data readiness.
- Implemented a caching mechanism for connection status in useModelDeploymentSettings to optimize performance and reduce unnecessary API calls.
- Enhanced loading states and error handling for improved user experience during deployment settings retrieval and connection testing.

* feat(i18n): add new translations for connection and loading states across multiple languages

- Introduced translations for "Checking connection" and "Loading configuration" in English, French, Japanese, Russian, Vietnamese, and Chinese.
- This update enhances the internationalization support, providing clearer user feedback during connection checks and loading phases.

* refactor(pagination): adjust layout and styling for pagination component

- Updated the pagination component to improve layout by removing unnecessary width constraints and enhancing responsiveness.
- Increased minimum width for pagination text to ensure better visibility and alignment across different screen sizes.

* feat(i18n): implement translations for various UI elements across multiple components

- Updated several components to utilize the translation function for titles and placeholders, enhancing internationalization support.
- Added new translation entries for "Filter by name or key..." and "Log Type" in English, French, Japanese, Russian, Vietnamese, and Chinese.
- This update improves user experience by providing localized text in the ChannelsTable, SummaryCards, ApiKeysTable, RedemptionsTable, UsageLogsTable, and UsersTable components.

* feat(i18n): integrate translation support in SummaryCards component

- Added the useTranslation hook to the SummaryCards component to enhance internationalization.
- This update allows for localized text rendering, improving user experience for diverse language speakers.

* feat(dashboard): refactor dashboard structure and introduce section-based navigation

- Removed the tab navigation in favor of a section-based approach, enhancing user experience by providing clearer context for the dashboard content.
- Introduced a new section registry to manage dashboard sections, allowing for easier expansion and maintenance.
- Updated sidebar configuration to reflect the new section structure, ensuring proper navigation links are displayed.
- Added translations for new section titles and descriptions to support internationalization.

* feat(i18n): update time range labels and enhance translation support

- Changed time range labels from shorthand (e.g., '1D') to full text (e.g., '1 Day') for better clarity.
- Updated various components to utilize the translation function for time range labels, improving internationalization.
- Added new translation entries for time ranges in English, French, Japanese, Russian, Vietnamese, and Chinese, enhancing user experience across languages.

* feat(dashboard): enhance type safety and improve component structure

- Updated the Dashboard component to use specific types for model data and filters, enhancing type safety.
- Introduced new types for announcements and FAQs, improving clarity and maintainability.
- Refactored LogStatCards and UptimePanel components to utilize AbortController for better data fetching management.
- Optimized the rendering of announcements and FAQs by using unique keys based on item IDs.
- Improved theme management in ModelCharts by caching the ThemeManager import to reduce dynamic imports.

* feat(agents): add comprehensive guidelines for React and Next.js development

- Introduced a new set of best practices and optimization techniques for React and Next.js applications, aimed at enhancing performance and maintainability.
- Included detailed rules covering various aspects such as event handling, API routes, rendering strategies, and state management.
- Added extensive documentation in AGENTS.md and SKILL.md to support developers in adhering to these practices.
- This update serves as a foundational resource for improving code quality and efficiency in React-based projects.

* chore(web): update package.json dependencies

- Removed outdated dependencies including @base-ui/react, @clerk/clerk-react, and others to streamline the project.
- Updated remaining dependencies to their latest versions for improved performance and security.
- This cleanup enhances the overall maintainability of the project.

* feat(usage-logs): implement section-based navigation and enhance log management

- Introduced a section registry for usage logs, allowing for better organization and navigation between different log categories (common, drawing, task).
- Updated the UsageLogsContent component to dynamically render titles and descriptions based on the selected section.
- Refactored UsageLogsTable and UsageLogsPrimaryButtons components to accept the active log category as a prop, improving modularity.
- Enhanced sidebar configuration to support new section navigation, ensuring users can easily access different log types.
- Updated routing to redirect to the default section if none is specified, improving user experience.

* feat(i18n): enhance internationalization across usage logs components

- Integrated the useTranslation hook in various components related to usage logs, including CommonLogsStats, UsageLogsTable, and column helpers.
- Updated labels, titles, and messages to utilize translation functions, improving localization support.
- Added new translation entries for log-related terms in English, French, Japanese, Russian, Vietnamese, and Chinese, enhancing user experience for diverse language speakers.

* feat(datetime-picker): integrate dayjs for date formatting

- Added dayjs as a dependency to the project for improved date handling.
- Updated the DateTimePicker component to use dayjs for formatting dates, enhancing consistency and readability of date displays.

* feat(date-handling): replace date-fns with dayjs for improved date management

- Updated the project to use dayjs instead of date-fns for date formatting and manipulation, enhancing consistency across components.
- Refactored DatePicker, DateTimePicker, and other components to utilize dayjs for date-related functionalities.
- Added a new dayjs configuration file to extend its capabilities with relative time support.
- Updated AGENTS.md to reflect the new technology stack, emphasizing the use of dayjs for date handling.

* refactor(agents): streamline front-end development guidelines and update technology stack

- Revised AGENTS.md to condense front-end development standards and best practices, making it more accessible for developers and AI assistants.
- Updated the technology stack section to reflect current dependencies, emphasizing the use of Bun, React 19, TypeScript, and other key libraries.
- Enhanced the document structure with a new table format for better readability and navigation, including a comprehensive table of contents for quick access to sections.

* feat(i18n): enhance date picker and datetime picker localization support

- Integrated internationalization support in DatePicker and DateTimePicker components by adding locale handling for multiple languages (English, French, Japanese, Russian, Vietnamese, Chinese).
- Updated the calendar component to accept a locale prop, ensuring proper localization of month and weekday labels.
- Improved user experience by allowing date selection to adapt based on the user's language preference.

* feat(layout): add SectionPageLayout component for structured page layouts

- Introduced a new SectionPageLayout component to facilitate structured layouts for pages with sections, enhancing the organization of content.
- Added subcomponents for Title, Description, Actions, and Content to improve clarity and maintainability of page structures.
- Updated AGENTS.md to include guidelines on avoiding unnecessary destructuring of props for better code readability.

* feat(layout): refactor components to use SectionPageLayout for improved structure

- Replaced AppHeader and Main components with SectionPageLayout across multiple features including Channels, Dashboard, ApiKeys, Models, Redemption Codes, Usage Logs, Users, and Wallet.
- Enhanced page organization by utilizing SectionPageLayout's Title, Description, Actions, and Content subcomponents, improving clarity and maintainability.
- This update standardizes the layout structure across the application, facilitating a more cohesive user experience.

* feat(usage-logs): enhance URL state management and redirection logic

- Added useEffect to synchronize column filters with URL search changes, preventing infinite loops caused by inline references.
- Improved redirection logic in usage logs to clear 'type' from the URL when the section is not 'common', enhancing user experience and URL cleanliness.

* fix(usage-logs): disable global filter and update DataTableToolbar props

- Disabled the global filter in the UsageLogsTable component to streamline the user interface.
- Updated the DataTableToolbar component to accept a null customSearch prop, enhancing flexibility in toolbar configuration.

* feat(routes): implement section-based routing for system settings and dashboard

- Introduced section-based routing for system settings and dashboard features, enhancing navigation and organization.
- Updated route definitions to include dynamic sections, allowing for more granular access to settings and dashboard components.
- Refactored existing routes to redirect to default sections when no specific section is provided, improving user experience.
- Added new section routes for models, usage logs, and system settings, ensuring consistency across the application.
- Removed deprecated routes to streamline the routing structure and improve maintainability.

* refactor(usage-logs): update column helper functions to require config parameter

- Modified createFailReasonColumn and createProgressColumn functions to require a config parameter instead of allowing it to be optional.
- Simplified destructuring of config to enhance clarity and ensure necessary properties are always provided, improving code reliability.

* refactor(usage-logs): improve section ID validation and routing logic

- Introduced a type guard function, isUsageLogsSectionId, to validate section IDs, enhancing type safety and reducing the need for casting.
- Updated UsageLogsContent to utilize the new validation function for determining the active category, improving clarity and reliability.
- Refactored routing logic to use isUsageLogsSectionId for section validation, ensuring proper redirection to the default section when necessary.

* refactor(calendar): update locale documentation for i18n support

- Revised the locale prop documentation in the Calendar component to specify the use of react-day-picker for internationalization, clarifying the expected locale setup for users.

* chore(i18n): remove redundant user information description from locale files

- Removed the user information description from English, French, Japanese, Russian, Vietnamese, and Chinese locale files to streamline translations and improve clarity.

* chore(i18n): streamline locale files by removing redundant entries

- Removed unnecessary entries from English, French, Japanese, Russian, Vietnamese, and Chinese locale files to enhance clarity and reduce clutter.
- Adjusted translations for consistency and improved user experience across multiple languages.

* chore(sidebar): remove deprecated usage logs route from sidebar config

- Eliminated the '/usage-logs' entry from the sidebar configuration to streamline navigation and improve clarity in the sidebar structure.

* refactor(redemption-codes): enhance internationalization support and improve UI consistency

- Updated various components to utilize translation functions for user-facing strings, ensuring a consistent experience across different languages.
- Added meta labels for table columns to improve accessibility and clarity.
- Revised confirmation and action texts in dialogs and tooltips to leverage translation, enhancing user experience.
- Updated locale files to include new translations for improved clarity and consistency.

* feat(masked-value-display): add MaskedValueDisplay component for sensitive data handling

- Introduced a new MaskedValueDisplay component to display masked values with a popover for full value visibility and a copy button for easy access.
- Updated api-keys-columns and redemptions-columns to utilize the new component, enhancing code reusability and UI consistency.
- Revised translation keys in locale files to remove colons for improved clarity.

* refactor(url-utils): simplify query parameter matching logic in checkIsActive function

- Updated the checkIsActive function to streamline the logic for matching URLs with and without query parameters.
- Removed unnecessary checks for query parameters when matching base paths, improving clarity and maintainability of the code.

* fix(channels-table): update group filter label to use translation function

- Replaced hardcoded 'All Groups' label with a translation function call to enhance internationalization support in the ChannelsTable component.

* chore(api-keys): remove deprecated API key action messages and related exports

- Deleted the api-key-actions.ts file, which contained action messages for enabling, disabling, and deleting API keys.
- Updated index.ts to remove the export of getApiKeyActionMessage, streamlining the codebase by eliminating unused functionality.

* refactor(i18n): enhance internationalization support across various components

- Updated multiple components to utilize translation functions for user-facing strings, ensuring a consistent experience across different languages.
- Revised constants and labels in the channels and redemption codes features to use i18n keys, improving maintainability and clarity.
- Ensured that success and error messages leverage translation functions, enhancing user experience and accessibility.
- Streamlined the handling of i18n keys in the constants files for better organization and clarity.

* refactor(i18n): enhance translation support across various components

- Updated multiple components to utilize translation functions for user-facing strings, ensuring a consistent experience across different languages.
- Revised pagination and status labels to use i18n keys, improving maintainability and clarity.
- Enhanced response time formatting to support internationalization, allowing for localized display of time values.
- Updated locale files to include new translations for improved clarity and consistency.

* docs(AGENTS): add type checking requirement for TypeScript changes

- Included a new guideline stating that type checks must be executed after modifying TypeScript or TSX code, ensuring no type errors are left unresolved.
- Updated the document to reflect this addition in the relevant section for better clarity on coding standards.

* feat(combobox-input): add ComboboxInput component for enhanced token selection

- Introduced a new ComboboxInput component to facilitate token name selection with search and filtering capabilities.
- Integrated the ComboboxInput into the UsageLogsFilterDialog for improved user experience when filtering by token name.
- Updated locale files to include new translations for user-facing strings related to token filtering.

* feat(combobox): integrate translation support for custom value prompt

- Added translation functionality to the Combobox component, replacing hardcoded text with a translatable string for the custom value prompt.
- Utilized the useTranslation hook from react-i18next to enhance internationalization support, ensuring a consistent user experience across different languages.

* refactor(i18n): improve Chinese translations for consistency and clarity

- Adjusted spacing in various Chinese translations to enhance readability and maintain consistency across the locale file.
- Updated multiple user-facing strings to ensure proper formatting and alignment with localization standards.

* feat(calendar): add CalendarDropdown component for enhanced dropdown functionality

- Introduced a new CalendarDropdown component to improve user interaction with dropdown selections in the calendar.
- Implemented state management for dropdown visibility and selection handling, enhancing the overall user experience.
- Updated styling for dropdown elements to ensure consistency and better alignment with the UI design.

* fix(balance-query-dialog): handle null currentRow and improve usage query logic

- Updated the BalanceQueryDialog component to safely access currentRow properties using optional chaining.
- Added a check to ensure currentRow is not null before proceeding with usage queries, preventing potential runtime errors.
- Refactored the handleQueryCodexUsage function to use a local variable for currentRow, enhancing code clarity.

* feat(i18n): add new translations for batch creation and channel updates

- Added new translation strings for batch creation instructions across multiple languages, enhancing user guidance.
- Included translations for the "Update Channel" prompt to improve clarity in channel configuration settings.
- Ensured consistency in terminology across locale files for better user experience.

* feat(channel-mutate-drawer): improve API key input handling and update translations

- Refactored the API key input logic in the ChannelMutateDrawer component to enhance readability and maintainability.
- Added new placeholder translations for batch creation and existing key prompts in multiple languages, improving user guidance.
- Ensured consistency in translation strings across locale files for better user experience.

* feat(fetch-models-dialog): implement sorting for model categories

- Added a new function to sort model categories alphabetically, placing 'Other' at the end for easier navigation.
- Updated the rendering logic in the FetchModelsDialog component to utilize the new sorting function for both new and existing models, enhancing user experience.

* refactor(wallet-stats-card): standardize props usage and improve layout consistency

Standardizes props usage and improves layout consistency in wallet stats card

Refactors the wallet stats card component to:
- Use props directly instead of destructuring for consistency
- Add min-w-0 to prevent content overflow
- Adjust text sizing with break-all for proper wrapping
- Implement responsive font sizes (3xl on mobile, 4xl on larger screens)
- Improve leading and tracking for better readability

Refactor wallet stats card for consistency and layout

Standardizes props usage and improves layout consistency in wallet stats card

- Uses props directly instead of destructuring for consistency
- Adds min-w-0 to prevent content overflow
- Adjusts text sizing with break-all for proper wrapping
- Implements responsive font sizes (3xl on mobile, 4xl on larger screens)
- Improves leading and tracking for better readability

* feat(web): add subscription management and admin settings UI

* feat(web): add subscription management and admin settings UI

- Add subscription management module (plans, pricing, toggle status, and related dialogs/tables with Stripe/Creem integration notes)
- Add channel affinity (rules and cache stats), Waffo integration, performance, and Grok model sections to system settings, with extended types and section registry
- Add status code mapping validation/risk warnings, upstream update hooks, and utilities for channels; add available models and sidebar module cards to user profile
- Add chat2link route and useMinimumLoadingTime, useTableCompactMode shared hooks

Made-with: Cursor

* fix: remove duplicate GenerateOAuthCode and add missing TaskBulkUpdate

- remove duplicate GenerateOAuthCode from github.go since oauth.go already has the generic version.
- add model.TaskBulkUpdate for bulk update by upstream task_id strings, fixing task_video.go build failure.

* feat(router): add chat2link and subscriptions routes

- register /chat2link page route under authenticated layout.
- register /subscriptions/ page route under authenticated layout.
- update auto-generated routeTree type definitions and route mappings.

* feat(docker): add development environment setup with Docker Compose

- Introduced docker-compose.dev.yml for local development, including services for new-api, Redis, and PostgreSQL.
- Created Dockerfile.dev for backend-only builds, optimizing the development workflow.
- Updated makefile to include new commands for starting backend services and frontend development.

* feat(web): complete i18n coverage for setup wizard and add language switcher

- wrap all hardcoded English strings in setup-wizard, database-step, usage-mode-step, and complete-step with t() calls, covering step titles, descriptions, form validation messages, and fallback strings.
- add LanguageSwitcher component to the top-right corner of the setup page so users can switch language during initial setup.
- register 25 dynamic i18n keys in static-keys.ts and provide full translations for zh/en/ja/fr/ru/vi.

* feat(i18n): internationalize default version text in workspace-switcher

- remove hardcoded 'Unknown version' default, use t('Unknown version') for i18n fallback
- add "Unknown version" translation entries across all 6 locale files (zh/en/fr/ru/ja/vi)

* feat(i18n): add full i18n coverage for channel-affinity settings page

- replace Chinese t() keys with English keys across three channel-affinity components to align with new frontend i18n conventions.
- add 51 translation entries to all 6 locale files (en/zh/ja/fr/ru/vi) covering main page, rule editor, and cache stats dialog.
- register section-registry dynamic keys in static-keys.ts.

* feat(i18n): add full i18n coverage for Waffo payment settings page

- replace Chinese i18n keys with English keys in waffo-settings-section.tsx for consistency.
- wrap previously hardcoded labels (Pay Method Type / Pay Method Name) with t().
- add 26 Waffo-related translation entries across all 6 locale files (en/zh/fr/ru/ja/vi).

* feat(i18n): add missing translations for global model settings page

- add all 6 locale translations for 3 missing t() keys in global-settings-card.
- register dynamically used 'Grok' key in static-keys.ts for i18n scanner coverage.

* feat(i18n): add full i18n coverage for Grok model settings page

- add translations in all 6 locales (en/zh/fr/ja/ru/vi) for grok-settings-card t() calls.
- cover violation fee toggle, amount input, and official docs link labels.
- include section-registry descriptionKey translation entries.

* feat(i18n): add full i18n coverage for performance settings page

- migrate all t() keys from Chinese to English to align with project conventions.
- add translations for all 6 locales (en/zh/ja/fr/ru/vi) covering disk cache,
  system monitoring, log management, and stats dashboard sections.
- remove 71 obsolete Chinese-keyed entries from every locale file.

* fix(i18n): add 116 missing English translation keys across all locales

- scan all t() calls to identify English keys used in code but absent from locale files.
- add translations for zh/en/fr/ja/ru/vi, keeping key sets and sort order consistent.
- covers system-settings, channels, models, auth, wallet and other modules.

* fix(i18n): add missing translations for log cleanup quick-select and confirm dialog

- wrap quick-select button labels (24 hours ago / 7 days ago / 30 days ago) with t().
- replace hardcoded English strings in purge confirm dialog with t() calls and date interpolation.
- add 5 new translation keys across all 6 locale files (zh/en/fr/ja/ru/vi).

* refactor(web): unify all time display with dayjs formatting

- replace all toLocaleString/toLocaleDateString/toLocaleTimeString and manual padStart concatenation with dayjs.format().
- standardize output: datetime as YYYY-MM-DD HH:mm:ss, date as YYYY-MM-DD, time as HH:mm:ss.
- add formatDateTimeStr, formatDateStr, formatTimeStr dayjs-based helpers in lib/format.ts.
- update 12 files across core utils and feature components.

* refactor(web): replace native datetime-local input with DateTimePicker in announcements

- swap browser-native datetime-local for the project's DateTimePicker component to match the UI used in log cleanup and other pages.
- convert between Date objects and ISO strings to bridge the form's string-based schema.

* refactor(web): replace native HTML elements with design system components

- replace ~35 native <button> with <Button> across pricing, profile, channels modules
- replace native <input>/<textarea>/<label> with <Input>/<Textarea>/<Label> for consistent form styling
- replace native <table> with <Table> components, <details>/<summary> with <Collapsible>
- replace decorative <hr> with <Separator> to ensure global UI consistency

* refactor(web): enhance profile components with design system consistency

- update ProfileSecurityCard to use buttons for security actions, improving accessibility and styling.
- modify AccountBindingsTab layout to a grid for better responsiveness and visual alignment.
- refactor NotificationTab to utilize icons for notification methods, enhancing user experience and clarity.

* fix(i18n): complete i18n coverage for profile page components

- wrap passkey card status badges (enabled/disabled, backup state) and last-used text with t()
- fix hardcoded button labels in security dialogs (change password, access token, delete account)
- internationalize all 2FA dialog strings (setup, disable, backup codes)
- fix email bind dialog description and button state text missing i18n
- wrap remaining hardcoded strings in notification tab and checkin calendar
- add all missing translation entries to zh.json and en.json

* fix(i18n): enhance error messages with translations for deployment access and settings

- wrap connection error messages in DeploymentAccessGuard and IoNetDeploymentSettingsSection with t() for internationalization.
- add missing translation key for "io.net model deployment is not enabled or api key missing" in all locale files (en, fr, ja, ru, vi, zh).

* 🧹 chore(web): resolve all ESLint errors and warnings

Align the Vite/React frontend with the current ESLint flat config and
React Compiler–related rules by fixing violations instead of broad
suppression where practical.

- Replace `any` with concrete types (`unknown`, `Record<string, unknown>`,
  domain types) where upstream/API shapes allow
- Fix duplicate imports, unused bindings, `no-console`, and empty blocks
- Address react-hooks issues: reorder declarations, memoize unstable
  callbacks (`useCallback`), extend dependency arrays, and use targeted
  disables only where sync-from-props in `useEffect` is intentional
- Refactor `motion.create` usage in ai-elements shimmer to avoid creating
  components during render (static-components)
- Stabilize TanStack Query/Mutation hook usage (query keys, `mutate` in
  deps) and add narrowly scoped rule disables where the linter conflicts
  with library patterns
- Disable `react-hooks/incompatible-library` in ESLint config for
  TanStack Table / RHF false positives
- Add file-level `react-refresh/only-export-components` disables for
  registry/provider/column modules that intentionally mix exports

`bun lint` completes with 0 errors and 0 warnings.

* ✨ feat(web): add subscription management to sidebar and align drawer with project conventions

- Register "Subscription Management" nav item in the admin sidebar group
  with CreditCard icon pointing to /subscriptions
- Add subscription module to sidebar config defaults and URL mapping so it
  integrates with the admin sidebar modules toggle in system settings
- Add subscription entry to sidebar-modules-section moduleMeta for the
  maintenance settings UI
- Refactor SubscriptionsMutateDrawer to follow the same patterns used by
  users, redemption-codes, and other mutate drawers:
  - Use shadcn Form/FormField/FormItem/FormControl/FormLabel/FormMessage
    instead of raw register() + Label + manual error display
  - Move SheetFooter outside the form with form attribute association
  - Use SheetClose for the cancel button
  - Reset form state on drawer close
  - Align SheetContent width (sm:max-w-[600px]) and spacing conventions

* ✨ feat(web): overhaul UI/UX with Vercel Geist design alignment

Refactor the entire frontend UI/UX to align with Vercel/OpenAI design
principles, covering layout, animations, skeleton loading, and overall
visual polish.

Motion & Page Transitions:
- Add centralized motion system (lib/motion.ts) with Vercel-style
  transition presets, stagger variants for tables, cards, and sidebars
- Implement AnimatedOutlet for route-level page enter animations
  using TanStack Router pathname keying
- Add PageTransition, StaggerContainer, StaggerItem, CardStagger,
  and TableStagger wrapper components for progressive reveal effects

Skeleton Loading — Vercel Geist Style:
- Replace shadcn default `animate-pulse` with Geist-style shimmer
  sweep animation (linear-gradient + background-position keyframes)
- Add `--skeleton-base` / `--skeleton-highlight` CSS variables tuned
  for both light and dark themes with neutral oklch tones
- Override auto-skeleton-react inline styles via CSS to unify all
  skeleton elements under the same shimmer effect
- Update TableSkeleton with varied column widths for a natural feel
- Add ContentSkeleton and QuerySkeleton wrappers for auto-skeleton
  integration with React Query error/loading states
- Respect prefers-reduced-motion: disable shimmer for accessibility

Layout & Sidebar:
- Upgrade sidebar expand/collapse transitions to cubic-bezier easing
- Add hover micro-interactions (background-color, color, transform)
  to sidebar menu buttons with smooth 150ms transitions
- Fix oklch color compatibility in sidebar outline variant
- Integrate AnimatedOutlet into AuthenticatedLayout for unified
  route-level animations

Theme & CSS:
- Streamline theme.css with cleaner oklch color definitions
- Add CSS table row stagger-in animations with nth-child delays
- Fix hover-scrollbar color bug (hsl → color-mix for oklch compat)
- Add content-auto utility for long list rendering optimization

Cleanup:
- Remove deprecated skeleton-wrapper.tsx
- Remove unused imports and dead code across components
- Add empty-state, error-state, and loading-state utility components

* 🐛 fix(docker): track bun.lock to fix Docker build failure

Remove `web/bun.lock` from `.gitignore` so the lock file is committed
to version control. The Dockerfile `COPY web/bun.lock .` instruction
requires this file to be present in the build context, and ignoring it
caused the build to fail with a "not found" error.

* ⬆️ chore(web): upgrade dependencies and fix all type/lint errors

Upgrade all frontend dependencies to latest stable versions:
- lucide-react 0.562 → 1.7 (major: brand icons removed)
- shiki 3.x → 4.x, eslint 9.x → 10.x, knip 5.x → 6.x
- @rsbuild/core 1.3 → 1.7, @types/node 24 → 25
- tailwindcss/postcss 4.1 → 4.2, motion 12.25 → 12.38
- @tanstack/react-query 5.90 → 5.95, zod 4.3.5 → 4.3.6
- react 19.2.3 → 19.2.4, axios 1.13.2 → 1.13.6
- prettier 3.7 → 3.8, typescript-eslint 8.52 → 8.57
- Add missing optional deps: @xyflow/react, embla-carousel-react

Resolve all TypeScript compilation errors introduced by upgrades:
- Replace lucide-react brand icons (Github) with react-icons/si
- Fix react-hook-form Control/Resolver generics for zod v4
- Fix Record<string, unknown> type constraints across API utils
- Fix axios interceptor return types in lib/api.ts
- Add type assertions for useSettings/useStatus hook returns
- Resolve Badge variant, spread type, and route path mismatches

Resolve all ESLint 10 errors:
- preserve-caught-error: attach cause to re-thrown errors
- no-useless-assignment: refactor redundant variable assignments
- prefer-as-const: use `as const` over literal type assertions
- no-unused-vars: prefix type-only schemas with underscore

Update tsconfig lib from ES2020 to ES2022 for Error.cause support.

* 🐛 fix(web): stop pricing model row from centering its content

Wrapping the row in shadcn <Button variant='ghost'> inherits
`justify-center`, and the inner flex container had no width, so
`justify-between` collapsed and the row appeared centered.

* feat: add Waffo payment integration and related UI components

- Introduced Waffo payment method with support for custom icons and settings.
- Updated payment settings section to include Waffo settings.
- Added Waffo payment request handling in the wallet API.
- Enhanced wallet recharge form to support Waffo payment methods.
- Implemented hooks for Waffo payment processing.
- Updated localization files for new Waffo-related strings.
- Added new payment type and icon for Waffo in constants and UI components.
- Refactored topup info handling to include Waffo payment methods and configurations.

* feat(profile): add admin-only upstream model update notification setting

* fix(web): make sidebar module user settings actually take effect

Previously, saving sidebar module preferences in profile had no effect
because the client ignored user-level sidebar_modules entirely. This
fix wires user config into useSidebarConfig so the sidebar updates
immediately without a page refresh.

Changes:
- Add UserPermissions type with sidebar_settings/sidebar_modules fields
- Refactor useSidebarConfig to merge admin × user config with AND logic
- Sync sidebar_modules to auth store on save for immediate UI updates
- Conditionally render SidebarModulesCard based on user permissions
- Treat null/empty user config as "do not narrow" for legacy users

* feat(web): add custom OAuth provider CRUD and login button support

Migrate custom OAuth from v1 to v2:
- Admin CRUD UI with provider table, form dialog, preset templates, and OIDC discovery
- Login page renders dynamic buttons for custom OAuth providers
- Fix account bindings display showing "Not bound" text when already bound

* feat(web): add ServerAddress, SMTPForceAuthLogin, CreateCacheRatio and group special usable settings

Migrate missing v1 system settings to v2:
- ServerAddress input in General > System Information
- SMTPForceAuthLogin toggle in Integrations > Email
- CreateCacheRatio JSON editor in Models > Ratio
- Group special usable group rules editor in Models > Ratio

* feat(web): wire user subscriptions dialog to users table row actions

The UserSubscriptionsDialog component already existed but had no entry point
in the users table dropdown menu. Add "Manage Subscriptions" menu item.

* chore(web): update i18n translations for new settings and custom OAuth

* 💎 refactor(web): redesign pricing page with flat, typography-driven layout

* 🌐 chore(i18n): complete missing translations and normalize project config

- Add 425+ missing translations across fr, ja, ru, zh, vi locales
  for subscription management, sidebar navigation, Grok settings,
  upstream model updates, pricing page, and other UI components
- Add 37 missing i18n keys used in t() calls but absent from locale
  files (pricing filters, display options, audio/cache labels, etc.)
- Fix stale tech stack info in CLAUDE.md, AGENTS.md, and project.mdc:
  React 18 → 19, Vite → Rsbuild, Semi Design → Radix UI + Tailwind
- Fix i18n key format description: "Chinese source strings" → English
- Deduplicate .cursor/rules/project.mdc to avoid triple-loading the
  same rules already present in root CLAUDE.md and AGENTS.md
- Add i18n-translate Cursor skill for repeatable translation workflow

* 🎨 refactor(web): redesign dashboard with flat, typography-driven layout

Replace Card-based dashboard components with a flat, border-driven design
system consistent with the pricing page, following the ui-style.mdc conventions.

Overview section:
- StatCard: replace Card wrapper with flat flex layout, monospace tabular
  values, uppercase tracking-wider labels, layered opacity hierarchy
- PanelWrapper: replace Card/CardHeader/CardContent with rounded-lg border
  container and border-b header
- SummaryCards: merge three stat cards into a single bordered container
  with divide-x separators; decouple border from stagger animation to
  prevent border deformation during entrance transitions
- ApiInfoPanel/Item: full-width list rows with border-b separators,
  monospace route names, layered opacity for URLs and descriptions
- AnnouncementsPanel: native button rows with hover:bg-muted/40, i18n for
  "Click for details" hint
- FAQPanel: lighter border-border/60 accordion dividers, muted answer text
- UptimePanel: uppercase tracking-wider group headers with bg-muted/30
  background, monospace uptime percentages, fine-grained border opacity

Models section:
- LogStatCards: replace Card with rounded-lg border + divide-x grid,
  fix react-hooks/exhaustive-deps by destructuring props before useEffect
- ModelCharts: replace Card+Tabs with bordered container + custom
  segmented control matching ui-style spec
- Suspense fallbacks: match new flat skeleton layout with accurate
  column structure

Animation:
- Wrap models section in FadeIn with staggered delay
- Keep CardStagger for overview panel grid (each panel has own border)

Other:
- Add ui-style.mdc cursor rule documenting the design language
- Disable react-refresh/only-export-components for src/routes/** in
  eslint config (TanStack Router route files always export Route objects)
- Fix zh.json: "Token-based" translation "基于令牌的" → "按量计费"

* ✨ refactor(web): adopt flat dot-and-text design for all status badges

Replace the bordered/colored-background StatusBadge and Badge components
across the entire frontend with a minimal flat design: a small colored
dot followed by colored text, eliminating visual noise from heavy
borders, backgrounds, and rounded pill shapes.

Key changes:

- Redesign StatusBadge to use dot + text instead of bordered pill style,
  removing cva-based background/border variants in favor of exported
  dotColorMap and textColorMap lookup tables
- Add children prop support to StatusBadge for flexible content rendering
  alongside the existing label prop
- Migrate all Badge usages (except pricing page) to StatusBadge with
  appropriate variant mappings (default→info, secondary→neutral,
  outline→neutral, destructive→danger)
- Consolidate adjacent multi-badge groups into single-dot layouts with
  dot separators (·) to reduce visual clutter in:
  - Channel balance columns (used + remaining)
  - Channel type column (type + IO.NET indicator)
  - User invite info column (invited + revenue + inviter)
  - Usage log stats bar (usage + RPM + TPM)
  - Usage log time/FRT column (time + FRT + stream status)
  - Subscription plan counts (active + expired)
  - Channel affinity scope/regex/key-source columns
  - Prefill group card headers (type + ID)
- Export dotColorMap and textColorMap for direct use in custom inline
  layouts that need consistent status colors without the full component

* ✨ refactor(web): redesign public layout and landing page with modern UI

Overhaul the public-facing layout, header, and homepage to deliver a
polished, animation-rich landing experience inspired by contemporary
SaaS design patterns.

Header:
- Replace sticky header with fixed floating navbar that compacts into
  a pill-shaped glass-morphism bar on scroll (backdrop-blur + ring)
- Add smooth 700ms cubic-bezier transitions for scroll-based shrinking
- Build full-screen mobile menu overlay with staggered entry animations
- Remove background color from logo container, show logo image directly

Homepage sections:
- Hero: gradient text title, radial gradient + grid pattern background,
  interactive terminal demo showcasing API request/response
- Terminal demo: auto-cycles through gpt-4o, claude-sonnet-4-20250514,
  gemini-2.5-pro, deepseek-chat with smooth cross-fade transitions,
  clickable model badges, dual theme support (light/dark), fixed height
- Stats: animated counters driven by IntersectionObserver with
  cubic-bezier easing, supports integer and decimal modes
- Features: Bento grid layout with gap-px border technique, each card
  includes contextual visuals (model list, security badge, workflow)
- How It Works: new three-step process section (Configure → Connect →
  Monitor) with connecting gradient line and numbered badges
- CTA: gradient mesh background with scale-in scroll animation
- Footer: streamlined brand column + link columns layout

New components:
- AnimateInView: IntersectionObserver-based scroll animation component
  supporting fade-up, fade-in, scale-in, fade-left, fade-right
- HeroTerminalDemo: themed terminal with model carousel and live
  request/response preview

CSS:
- Add landing page scroll-triggered keyframe animations
- Add terminal demo animations (blink cursor, spinner, pulse indicator)
- Respect prefers-reduced-motion throughout

i18n:
- Add 17 new translation keys across all 6 locales (en/zh/fr/ja/ru/vi)

* ✨ feat(web): align usage logs and channels with legacy UI

Usage logs
- Show Refund (type 6) in detail dialog and hide conversion chain for refunds
- Sync filter dialog state from URL for model, token, group, username, and requestId

Channels
- Support optional stream flag in channel test API, actions, and test dialog
- Show upstream model update badges (+added / -removed) on fetchable channel types
- Add form fields and drawer toggles for upstream model update check and auto-sync
- Persist upstream model update flags in channel settings JSON for fetchable types

i18n
- Add locale strings for upstream model update UI (en, zh, fr, ja, ru, vi)

* 🐛 fix(web): prevent transient vertical scrollbar on tables during animations

Add overflow-y-clip to the shared Table container (data-slot=table-container)
alongside overflow-x-auto. Setting overflow-x to auto implicitly pairs with
overflow-y: auto in browsers, which made the table shell briefly show a
vertical scrollbar during route enter motion (y/blur) and table row stagger.

Remove the redundant descendant selector workaround from the model pricing
GroupPricingSection; behavior is now covered globally by the Table component.

* 🏗️ refactor(web): redesign console layout with fixed header, scrollable content, and pinned footer

Overhaul the authenticated console layout to match the OpenAI dashboard
pattern: header and page title bar stay fixed at the top, only the
content area scrolls, and table pagination is pinned to the bottom.

Layout architecture:
- Lock SidebarInset to full viewport height (h-svh) so all inner
  regions are controlled by flexbox instead of document scroll
- Convert Main from a generic div to a semantic <main> flex container
  with overflow-hidden, removing the legacy `fixed` prop and
  `data-layout` attribute
- Strip scroll-shadow logic and `fixed` prop from Header/AppHeader;
  the header is now naturally fixed as a shrink-0 flex child
- Restructure SectionPageLayout into three flex regions: a shrink-0
  title bar, a flex-1 overflow-auto content area, and a shrink-0
  footer portal target with empty:hidden
- Add min-h-0 to AnimatedOutlet wrappers to prevent flex overflow

Footer portal system:
- Introduce PageFooterProvider / PageFooterPortal (React Context +
  createPortal) so deeply nested table components can render their
  DataTablePagination into the fixed footer without prop drilling
- Migrate all 8 data tables (api-keys, channels, users, models,
  deployments, usage-logs, subscriptions, redemption-codes) to use
  PageFooterPortal for pagination

Page-level fixes:
- Profile: wrap content in a scrollable flex child with proper padding
- SystemSettings: remove overflow-auto from wrapper to avoid nested
  scrollbars (sub-pages manage their own scroll)
- Playground / Error pages: remove obsolete `fixed` props

API keys UX improvement:
- Replace inline key show/hide toggle with a Popover-based reveal,
  removing toggleKeyVisibility and keyVisibility state from the
  provider context

Cleanup:
- Remove dead CSS rule for body:has([data-layout='fixed'])
- Remove unused `fixed` prop from Header, AppHeader, and Main types
- Export PageFooterPortal from layout barrel file

* 💅 refactor(web): polish table UI consistency and add pagination transitions

- Standardize primary action buttons (Create, Add, Search) to size="sm"
  across all pages for visual consistency with channels and models
- Redesign NumericSpinnerInput with minimal inline style: plain text by
  default, hover-revealed +/- buttons, click-to-edit — replacing the
  clunky bordered input with stacked chevron arrows
- Fix vertical scrollbar in channels group column by replacing
  overflow-x-auto with overflow-hidden (redundant with +N collapse)
- Simplify API keys group column: replace colorful StatusBadge pairs
  with clean typography using opacity hierarchy and dot separators
- Move API key copy loading indicator from key text to the copy button
  itself, eliminating layout shift during key resolution
- Reduce page title from text-2xl to text-lg and subtitle to text-sm
  in SectionPageLayout for a more compact header
- Add smooth opacity transition (duration-150) on all 7 server-paginated
  tables during background data fetches (isFetching && !isLoading),
  with pointer-events-none to prevent interaction during loading
- Constrain usage logs Details column width (size: 200, maxSize: 220)

* 🐛 fix(web): restore missing padding on system settings content

The console layout refactor in d2150469 moved padding ownership from
Main onto each route, but SystemSettings was missed — its Outlet
wrapper had no padding, so the content area sat flush against the
sidebar and top nav. Add `px-4 pt-6 pb-4` to match the vertical
rhythm used by SectionPageLayout and the Profile page.

* 📱 refactor(web): standardize mobile responsive layout across all table pages

Unify mobile experience for all data table pages (channels, keys, models,
deployments, usage-logs, users, redemption-codes, subscriptions) with a
consistent layout pattern and cleaner header area.

DataTableToolbar:
- Redesign mobile layout: full-width search input + collapsible filter
  toggle button with active filter count badge
- Filters, additional search, and reset button collapse into an
  expandable section on mobile, keeping the default view compact
- Desktop layout remains unchanged

SectionPageLayout:
- Tighten mobile spacing (padding, gaps) for higher content density
- Scale down title (text-base) and description (text-xs) on mobile
- Shrink action button gaps on small screens

ChannelsPrimaryButtons:
- Move Tag Mode and Sort by ID toggles into the "More" dropdown on
  mobile (via DropdownMenuCheckboxItem), freeing header space
- Desktop toggle switches remain visible outside the dropdown

MobileCardList (shared component):
- Compact list-item layout with title + badge header row and
  side-by-side key fields, replacing individual card components
- Structured (CompactRow) and fallback (FallbackRow) rendering modes
  driven by column meta (mobileTitle, mobileBadge, mobileHidden)

New MobileCardList integration:
- Users table: username as title, status as badge; hide id,
  display_name, invite_info on mobile
- Redemptions table: name as title, status as badge; hide id,
  created_time, expired_time, used_user_id on mobile
- Subscriptions table: plan title as title, enabled as badge; hide id,
  sort_order, reset, payment, total_amount, upgrade_group on mobile

Column meta updates:
- Add mobileTitle/mobileBadge/mobileHidden meta across all 8 table
  column definitions for consistent mobile field prioritization

Minor fixes:
- Hide Subscriptions Stripe/Creem alert on mobile
- Disable card hover animations on mobile via CSS media query

* 🐛 fix(web): sync favicon with custom system logo

Favicon stayed at the hardcoded /logo.png while document.title already
followed system_name, leaving tab icon and site branding out of sync.
Apply the logo as favicon from localStorage cache on startup, refresh
from getStatus(), and re-apply when useSystemConfig finishes preloading.
Extract applyFaviconToDom helper into lib/dom-utils with idempotent guard
to avoid redundant DOM writes.

* ✨ feat(web): add channel affinity rule templates and CreateCacheRatio visual editing

Port missing features from legacy frontend (b8650b9 merge) to the new
React frontend:

- Add Codex CLI and Claude CLI channel affinity rule templates with
  header passthrough presets (pass_headers operations for Originator,
  Session_id, X-Codex-*, X-Stainless-*, Anthropic-*, etc.)
- Introduce "Add Rule" dropdown menu with blank, Codex CLI, and Claude
  CLI template options in the channel affinity settings page
- Add "Fill Templates" button to batch-append both CLI templates with
  duplicate name resolution and confirmation dialog
- Support templateKey prop in RuleEditorDialog to pre-fill form fields
  from selected template, auto-expanding advanced settings when a
  param_override_template is present
- Add CreateCacheRatio support to the model ratio visual editor, edit
  dialog, and form — previously only editable in JSON mode, now fully
  integrated into the visual table column, add/edit dialog fields, and
  save/delete handlers

* 🐛 fix(web): fix content-type detection bugs in About and Home pages

- Fix About page URL detection: replace naive `startsWith('https://')`
  with proper `new URL()` validation to support both http and https, and
  handle untrimmed input that previously caused silent misdetection
- Fix About page HTML detection: remove overly broad `startsWith('<')`
  and `endsWith('>')` checks that could misclassify Markdown or XML
  content; align with LegalDocument's regex-only `isLikelyHtml` approach
- Fix Home page URL detection: same `startsWith('https://')` bug,
  replaced with `new URL()` protocol validation
- Refactor About page to use early-return pattern instead of deeply
  nested ternary expressions for better readability
- Replace About loading spinner with Skeleton placeholder consistent
  with LegalDocument
- Add `prose prose-neutral dark:prose-invert` typography classes to
  About HTML/Markdown rendering for proper dark mode support
- Remove unused `Code` icon import from About page

* ✨ feat(web): port missing features from legacy frontend and complete i18n

Backport and enhance several features from the old frontend (web/old)
that were missing or incomplete in the new React frontend:

- Playground & channel test: parse structured JSON error responses from
  SSE streams and non-streaming API calls, extract error codes, and
  display actionable UI for `model_price_error` (admin settings link)
- User management: replace local quota manipulation with atomic
  server-side quota adjustments (add/subtract/override) via dedicated
  API endpoint, making the quota field read-only in the edit drawer
- Subscriptions: display next quota reset time for active subscriptions
- Dashboard: limit model ranking chart to top 20 models with an "Other"
  bucket, add dimension tooltips with sorted values and totals to model
  call trend and user consumption trend charts
- i18n: add 24 new translation keys across all 6 locales (en, zh, fr,
  ja, ru, vi) for the newly introduced UI elements and messages

* 🎨 feat: add backend-configurable frontend theme switching (default/classic)

Introduce runtime frontend theme switching so administrators can switch
between the new frontend (Radix UI + Tailwind) and the classic frontend
(Semi Design) from the settings page without restarting the server.

Directory restructuring:
- Move new frontend from web/ to web/default/
- Move classic frontend from web/old/ to web/classic/
- One-frontend-per-folder layout for extensibility

Backend (injection pattern):
- Add setting/system_setting/theme.go with GlobalConfig.Register("theme")
  so the DB key "theme.frontend" is handled automatically by
  handleConfigUpdate — no switch-case in updateOptionMap needed
- Use atomic.Value in common.GetTheme()/SetTheme() for lock-free
  concurrent reads on the hot path (static file middleware)
- Add themeAwareFileSystem that delegates to the correct embedded FS
  based on the current theme at request time
- Embed both frontends into the binary via go:embed
- Add controller validation for theme.frontend values
- Expose theme in GET /api/status response

Frontend settings UI:
- New frontend: add "Frontend Theme" select in System Information section
  using Radix UI Select + react-hook-form + Zod validation
- Classic frontend: add "前端主题" select in Personalization section
  using Semi Design Form.Select

Build system:
- Update Dockerfile with multi-stage builds for both frontends
- Update Makefile with separate build targets for each frontend
- Update GitHub Actions release workflow for dual frontend builds

i18n:
- New frontend: add translations for all 6 locales (en/zh/fr/ja/ru/vi)
- Classic frontend: add translations for all 7 locales (en/zh-TW/ja/fr/ru/vi)
- Fix zh "AI Proxy Library" → "AI 代理库"

Documentation:
- Update CLAUDE.md, AGENTS.md, .cursor/rules/project.mdc to reflect
  the new web/default/ and web/classic/ directory structure

* ✨ feat(web): add allow_speed passthrough for Claude channels, fix multi-key index and inference_geo scope

- Add `allow_speed` toggle for Anthropic (type 14) channels to control
  Claude inference speed mode passthrough, with full form schema,
  settings persistence, and UI switch
- Fix `allow_inference_geo` to also apply to Anthropic (type 14) channels,
  not just OpenAI (type 1), matching the backend behavior for Claude data
  residency region control
- Fix multi-key management dialog to display 1-based key indices instead
  of 0-based (#{key.index + 1})
- Fix TypeScript type error in section-registry by adding type assertion
  for theme.frontend enum
- Add i18n translations for all new keys across 6 locales (en, zh, fr,
  ja, ru, vi)

* 🧹 chore: clean up editor configs, consolidate agent skills, and set classic as default theme

- Add .cursor/ to .gitignore and remove tracked editor config files
  (.cursor/rules/, .cursor/skills/) from version control
- Consolidate .agents/skills/vercel-react-best-practices by keeping only
  the compiled AGENTS.md and removing redundant SKILL.md and 57 individual
  rule files under rules/
- Change default frontend theme from "default" to "classic" in both
  common/constants.go init and setting/system_setting/theme.go

* feat: Frontend Tiered Pricing, Waffo Payments, and Rsbuild 2 Upgrade (#24)

* feat(ui): add codex extra limits, key last used, and admin audit surfaces

- codex usage dialog: render `additional_rate_limits` with `RateLimitGroupSection` and typed base/secondary window data.
- api keys table: add "Last Used" column from `accessed_time`.
- usage log details: show top-up audit and manage operator for admins; extend `LogOtherData` audit fields; broaden IP display; warn when legacy records lack audit data.
- billing history: show user id badge for admins; add zh i18n for new strings.

* feat(web): add dynamic pricing breakdown and Waffo Pancake payments

- add billing-expr parsing and DynamicPricingBreakdown; surface tiered_expr in model list/details.
- extend PricingModel with billing_mode, billing_expr, and pricing_version for backend parity.
- add Waffo Pancake integration settings, amount/pay APIs, hook, and recharge flow wiring.
- update payment confirm/recharge UI and Chinese locale strings.

* feat(pricing): add tiered billing editor and tool price settings

- introduce tier-expr and extend billing-expr (time/param conditions, combine/split helpers, editor utilities) for visual tiers and request rules.
- support tiered_expr in model ratio dialog, form, and visual editor with billing_setting fields and default JSON placeholders.
- add TieredPricingEditor and tool price settings UI plus i18n updates.

* chore(web): bump rsbuild to v2 and align build config

- upgrade @rsbuild/core, @rsbuild/plugin-react, and Rspack 2 transitives; bump TanStack Router packages and refresh bun.lock.
- replace deprecated performance.chunkSplit with top-level splitChunks cache groups for react, radix, and tanstack vendors.
- factor dev server proxy into devProxy; set legalComments to none in prod; enable performance.buildCache keyed by VITE_REACT_APP_VERSION.
- TanStack Router plugin: enable autoCodeSplitting only in production for faster dev navigation and HMR.

* fix(i18n): update translations for API keys and Waffo Pancake settings

- Corrected translations for "API Private Key" and "Merchant ID is required" across multiple languages.
- Added new translation for "Configure Waffo Pancake hosted checkout integration for USD-priced top-ups."
- Updated various existing translations to ensure consistency and clarity in user interface text.

* refactor(code-block): simplify code highlighting and improve theme handling

- Updated the highlightCode function to support dual themes in a single call, reducing complexity.
- Removed unnecessary state management for dark theme HTML, streamlining the component.
- Enhanced CSS for Shiki themes to ensure proper token color application in dark mode.

* refactor(wallet): use isWaffoPancakePayment for pancake payment dispatch

- replace the waffo_pancake string literal with the shared helper for consistency with use-payment and PAYMENT_TYPES.
- centralize the value so a constant change does not require hunting for typos in multiple call sites.

* fix(wallet): validate waffo pancake checkout url and safe open

- allow only parseable http/https redirect targets from the backend, rejecting dangerous schemes.
- pass noopener and noreferrer in window.open to reduce reverse tabnabbing.
- show a toast and abort on invalid URLs; add i18n entries across locales.

* fix(wallet): harden payment icon image URLs

- add normalizeHttpIconUrl to allow only http(s) after resolution and reject userinfo in URLs.
- set referrerPolicy, lazy loading, and async decode on the icon <img> to cut referrer leakage.
- fall back to built-in icons on invalid URLs, same as when iconUrl is missing.

* fix(pricing): label param() conditions as body param in dynamic pricing

- non-header request rules map to `param()`, not query strings.
- align with tiered pricing editor by using the existing `Body param` string.

* fix(rsbuild): update legalComments handling in build config

- Rely on Rsbuild's default legalComments setting in all modes to ensure compliance with open-source licensing requirements.
- Clarified comments to explain the implications of omitting legalComments in production.

* fix(i18n): correct billing and codex UI strings in locale files

- restore ~83 en.json values to English (tool pricing, audit text, alipay label, etc.).
- add proper fr/ru/vi/ja strings so those locales no longer copy zh.
- change five locale files only; zh.json unchanged.

* fix(i18n): update locale files for improved translations and sync report

- Added missing translations and corrected existing strings in English, French, Japanese, Russian, Vietnamese, and Chinese locale files.
- Updated the sync report to reflect zero missing translations across multiple locales.
- Enhanced the untranslated count for Japanese locale to ensure completeness.
- Changed the base locale from zh.json to en.json for better alignment.

* chore(agents): add i18n-translate agent skill

- add `.agents/skills/i18n-translate/SKILL.md` documenting locale layout under `web/default` and
  `bun run i18n:sync` usage.
- capture a repeatable maintainer workflow with embedded script examples to find missing keys
  and untranslated values.
- give agents a clear path to complete and verify translations across en, zh, fr, ja, ru, and vi.

* feat(settings): hide frontend theme setting (#25)

* feat(settings): hide frontend theme setting

- add a local hidden feature flag with window.newapiUnlock support.
- hide the frontend theme option by default and reveal it immediately after unlock.

* feat(settings): support click unlock for frontend theme setting

- add a shared hidden click unlock hook for repeated-click gated UI.
- reveal the frontend theme option after triple-clicking the system information title.
- preserve the Doubao API address ten-click unlock behavior and remove global unlock functions.

* feat(sync 59337e9): Sync classic tiered billing, upstream price synchronization, and model management features to web/default (#26)

* feat(skill): add classic-to-default-sync skill for auditing and syncing web/classic changes to web/default

- Introduced a new skill to inspect a given commit's changes in web/classic and synchronize features and fixes to web/default.
- Documented workflow steps for extracting diffs, mapping changes, triaging, implementing, and reporting on the synchronization process.
- Emphasized quality standards and internationalization considerations for new user-visible strings.

* feat(web/default): sync billing and model management features from classic

- add `len` condition variable (total input context length); introduce
  BILLING_PRICING_VARS / BILLING_CONDITION_VARS to separate pricing vars
  from condition-only vars; fix tier condition regex to accept `len`.
- rewrite upstream ratio sync components to support per-model grouped
  rows and new ratio types (create_cache, image, audio, billing_expr).
- add LlmPromptHelper component; update tiered presets to use `len` for
  conditions; add GLM-4.5 Air, Doubao Seed 1.8, Qwen3 Omni Flash, and
  weekend-discount presets.
- add created_at / last_login_at columns to users table; add "Removed
  Models" tab to FetchModelsDialog for mapping source keys not in the
  models list.
- add extractMappingSourceModels helper; update dynamic-pricing-breakdown
  to use system currency settings; add 19 i18n keys across all locales.

* ✨ feat(default): surface tiered billing in usage logs and gate Passkey ops behind 2FA

Continues the classic-to-default sync (commit 1be6cdb) by porting the
remaining audit-log, pricing-hint, and Passkey lifecycle features from
web/classic to web/default using the default frontend's component
patterns (Radix UI, Tailwind, shadcn-style dialogs).

* feat(usage-logs): show tiered_expr breakdown and matched tier in details

  - Extend `LogOtherData` with `billing_mode`, `expr_b64`, and
    `matched_tier` fields populated by the backend for tiered logs.
  - Add `decodeBillingExprB64`, `resolveMatchedTier`, and
    `getTieredBillingSummary` helpers in `usage-logs/lib/format.ts` that
    centralise tiered-billing parsing on top of the canonical
    `parseTiersFromExpr` / `BILLING_PRICING_VARS` from the pricing
    feature, instead of duplicating the classic-frontend renderer.
  - Render `<DynamicPricingBreakdown>` inside the consume-log details
    dialog with the matched tier row highlighted in emerald and tagged
    "Matched"; suppress the legacy claude/audio/image cost rows when a
    tiered expression is in effect.
  - Surface per-tier prices and the matched tier label in log row
    segments and the billing breakdown table.

* feat(pricing): show tier-count, time-based, and request-based hints in model list

  - Add `summarizeTieredExpr` that derives compact dynamic-pricing
    metadata (tier count + presence of time/request conditions) from a
    `tiered_expr` model, computed once per render via `useMemo`, so
    users can tell *what kind* of dynamic pricing applies before
    drilling into the model details.
  - Render the hints alongside the existing "Dynamic Pricing" badge in
    `<ModelRow>`.
  - Extend `<DynamicPricingBreakdown>` with a `matchedTierLabel` prop so
    the same component can be reused from the usage-log details dialog
    to highlight the tier that actually fired.

* feat(profile): require Security Verification for Passkey register/remove

  - Wire `usePasskeyManagement` through `useSecureVerification` and
    `<SecureVerificationDialog>` in `<PasskeyCard>`.
  - Registration prompts for 2FA before issuing the Passkey credential
    (only when 2FA is already enabled — otherwise the browser-level
    Passkey prompt itself acts as proof of presence and we register
    directly).
  - Removal prompts for 2FA or Passkey, whichever the account has
    enabled, with informative toasts when neither method is available
    or the device lacks Passkey support.
  - Scope the dialog method set to the required factor so users cannot
    fall back to a weaker method, and propagate cancellation cleanly.

* refactor: tighten upstream-ratio-sync and fix tier editor narrowing

  - Drop the unused `hasSynced` state and dead `getOrderedRatioTypes` /
    `isSelectableUpstreamValue` imports from `upstream-ratio-sync.tsx`.
  - In the cost estimator, narrow `BILLING_EXTRA_VARS` entries with a
    null-`field` guard to silence the type checker and make the
    "pricing variables only" contract explicit.
  - Apply Prettier-consistent formatting to the upstream-ratio-sync
    table/columns, channel mutate drawer, system info section,
    tier-expr, and wallet helpers (no behaviour change).

* i18n: add 9 keys across en/zh/fr/ja/ru/vi

  - `{{count}} tiers`, `Billing Process`, `Matched`, `Matched Tier`,
    `Request-based`, `Security verification`, `Time-based`, plus the
    two new Passkey verification description strings.

* 🔧 refactor(default): align upstream price sync, tiered billing, and fetch-models with classic 59337e9

Port and optimize the remaining web/classic features from commit 59337e9 to web/default,
covering upstream price synchronization, tiered billing expressions, model fetching, and
channel preset detection. Improve component architecture, memoization, and i18n coverage.

Upstream Price Sync
- Extend sync to all ratio fields: CacheRatio, CreateCacheRatio, ImageRatio,
  AudioRatio, AudioCompletionRatio in addition to ModelRatio / CompletionRatio
  / ModelPrice
- Add tiered billing sync (billing_mode + billing_expr) with auto-pairing so
  selecting one upstream tier value populates the other from the same source
- Bulk select / unselect per upstream column with indeterminate checkbox state
  reflecting partial selection
- Confidence indicators warn when an upstream entry is heuristically derived
- Conflict confirm dialog gains loading state and disables actions during sync
- Default endpoint per channel: /api/pricing for official preset,
  /api/models.dev for the models.dev preset, /api/v1/models for OpenRouter,
  with the rest falling back to the global default
- Rename tab label from "Upstream sync" to "Upstream price sync" for clarity

Tiered Pricing Editor
- Add `len` (full input length, including cache hits) as a tier-condition
  variable to avoid mis-routing when cache hits reduce `p`
- When inserting a new tier, automatically convert the previous catch-all into
  a bounded tier with a `len <= X` upper bound
- Cap each tier at 0~2 conditions and disable the add-condition button at the
  limit, with an Alert explaining the recommended `len` usage
- Extend presets with Multimodal (img / img_o / ai / ao), Request rule
  (header/param matching), and Time-based (hour / weekday) entries
- Embed an LLM prompt helper that copies a model-aware template for designing
  expressions with ChatGPT / Claude

Fetch Models Dialog
- Add a "Removed Models" tab listing models still in the local selection but
  no longer returned by the upstream listing
- Exclude `model_mapping` source keys from the removed view so aliases never
  appear as missing entries
- Force-remount tab content on tab switch via `key` prop to clear stale state
- Switch count placeholders to `{{count}}` interpolation across "Existing
  Models", "New Models", and "Removed Models" labels

Channel Selector & Constants
- Recognize the models.dev preset (id, base_url, name) alongside the existing
  official-channel preset detection
- Add MODELS_DEV_PRESET_* and OPENROUTER_* constants and reorder
  ENDPOINT_OPTIONS so `pricing` is preferred over `ratio_config`
- Expose the new ratio types in RATIO_TYPE_OPTIONS for the sync filter

Types
- Add optional `type` field to UpstreamChannel for endpoint inference
- Extend RatioType union with create_cache_ratio, image_ratio, audio_ratio,
  audio_completion_ratio, billing_mode, and billing_expr

Code Quality & Performance
- Extract upstream-ratio-sync-helpers.ts to host shared types
  (RatioDifferenceEntry, ModelRow, ResolutionsMap), field ordering
  (RATIO_SYNC_FIELDS, SYNC_FIELD_ORDER, NUMERIC_SYNC_FIELDS), and selection
  logic (getPreferredSyncField, isSelectableUpstreamValue, getSyncFieldLabel)
- Memoize the column definitions in useUpstreamRatioSyncColumns and pull the
  per-cell rendering into a renderUpstreamValue helper to remove inline IIFEs
- Wrap handleBulkSelect / handleBulkUnselect in useCallback for stable refs;
  rename the misleading `_upstream` parameter to `upstream`
- Convert parsedRatios from useCallback (returning a function) to useMemo
  (returning the value) and update all call sites to read it as a value
- Memoize the channels list with useMemo so the endpoint-init effect no
  longer fires on every render due to a fresh `?? []` reference

i18n
- Add and translate new keys ("Upstream price sync", "Audio Ratio", "Audio
  Completion Ratio", "Cache Create Ratio", "Image Ratio", "Expression
  Billing", "Fixed Price", "{{n}} model(s) selected", tier guidance, etc.)
  across en, zh, fr, ja, ru, vi
- Fix truncated keys ("Existing Models (", "New Models (", "Removed Models (")
  to proper {{count}} interpolated forms in every locale
- bun run i18n:sync reports 0 missing and 0 extra keys for every locale

Verification
- bun run typecheck: pass
- bun run lint: pass
- bun run i18n:sync: pass (0 missing / 0 extras across all locales)

* 🐛 fix(default): port classic 73e5557 tiered-billing fixes and dedupe Title-Case ratio i18n keys

Sync the web/classic frontend fixes from upstream merge 73e5557 to
web/default, and clean up duplicated Title-Case ratio labels in the
upstream sync UI that were shadowing the canonical sentence-case i18n
keys.

Cache-token filter for tiered model price (port of 9f8a4ec05)
- The matched-tier breakdown shown in the usage-log details dialog
  and in the log table previously listed every cache-related price
  (Cache Read, Cache Write, Cache Write 1h) regardless of whether
  the request actually consumed cache tokens.
- `getTieredBillingSummary` in `usage-logs/lib/format.ts` now skips
  `cache`-group vars when none of `cache_tokens`,
  `cache_creation_tokens`, `cache_creation_tokens_5m`, or
  `cache_creation_tokens_1h` are positive, mirroring the classic
  `renderTieredModelPrice` / `renderTieredModelPriceSimple` logic.
- Extract `hasAnyCacheTokens(other)` as an exported helper so the
  predicate is defined once.
- Add a `hideCacheColumns?: boolean` prop to
  `DynamicPricingBreakdown` and wire it up from the log details
  dialog so the full tier table hides cache columns under the same
  condition. `model-details.tsx` keeps the default (show all
  configured prices), since that view represents the model's
  pricing structure rather than a specific call.

`tiered_expr` ratio/price fallback during sync delays (port of bee339d27)
- When saving a model in tiered-expression mode, the visual editor
  used to delete every ratio/price map entry for the model and only
  write `billing_setting.billing_mode` /
  `billing_setting.billing_expr`. In multi-instance deployments,
  instances that had not yet observed the billing_setting update
  fell back to ratios that no longer existed, breaking pricing.
- `model-ratio-dialog.tsx`: `handleSubmit` always passes every form
  field (`price`, `ratio`, `cacheRatio`, `createCacheRatio`,
  `completionRatio`, `imageRatio`, `audioRatio`,
  `audioCompletionRatio`) into the data object regardless of
  `pricingMode`, so a switch from per-token to tiered_expr no
  longer drops the previously entered ratios.
- `model-ratio-visual-editor.tsx`:
  - The row builder now also surfaces ratio/price values for
    `tiered_expr` rows, so they survive the edit-and-save round
    trip and the next save.
  - `handleSave` factors out a `setIfPresent` helper and persists
    ratio/price entries for `tiered_expr` models alongside
    billing_mode / billing_expr. These act purely as fallback
    because the backend's `ModelPriceHelper` checks `billing_mode`
    first.
  - Cell rendering mutes ratio/price values whenever the row is
    `tiered_expr` (in addition to the existing per-request
    muting), making it visually clear the values are fallback,
    not the active pricing source.

i18n: dedupe Title-Case ratio labels in upstream sync
- `upstream-ratio-sync` `RATIO_TYPE_OPTIONS` previously used
  Title-Case labels (`Model Ratio`, `Cache Ratio`, `Audio
  Completion Ratio`, …) that were rendered through `t()` but never
  existed as canonical keys in the catalogue. The form-field side
  has used sentence-case (`Model ratio`, `Cache ratio`, …) for
  some time, leaving two parallel translation entries per ratio
  type and causing the upstream sync UI to fall back to the
  English source string in zh/ja/ru/fr/vi.
- Rewrite `RATIO_TYPE_OPTIONS` in
  `system-settings/models/constants.ts` and the conflict-detection
  labels in `upstream-ratio-sync.tsx` to reuse the sentence-case
  keys.
- Drop the duplicate Title-Case entries from every locale and
  promote the better translations onto the surviving sentence-case
  keys (e.g. zh `Image ratio` keeps "图片倍率", `Audio completion
  ratio` keeps "音频补全倍率").
- Add a comment to `RATIO_TYPE_OPTIONS` warning future
  contributors not to switch back to Title Case without updating
  the catalogue.

Note on backend fix 4e93148d9
- The backend portion of the merge (allocating a fresh map in
  `updateConfigFromMap` so removed keys are properly cleared) is
  already on HEAD; no additional change is needed.

Verification
- `bun run typecheck`: pass
- `bun run lint`: pass
- `bun run i18n:sync`: 0 missing / 0 extras across
  en / zh / fr / ja / ru / vi

---------

Co-authored-by: Seefs <40468931+seefs001@users.noreply.github.com>
Co-authored-by: Seefs <i@seefs.me>
Co-authored-by: feitianbubu <feitianbubu@qq.com>
Co-authored-by: Calcium-Ion <i@caion.me>
Co-authored-by: Xyfacai <xyfacai@gmail.com>
Co-authored-by: xiangsx <1984871009@qq.com>
Co-authored-by: 郑伯涛 <351175318@qq.com>
Co-authored-by: RedwindA <austinaosid@gmail.com>
Co-authored-by: dean <1006393151@qq.com>
Co-authored-by: QuentinHsu <xuquentinyang@gmail.com>
Co-authored-by: Bliod <bliod@bliod.lan>
Co-authored-by: Apple\Apple <zeraturing@foxmail.com>

web/default/src/features/system-settings/general/channel-affinity/api.ts

@@ -0,0 +1,45 @@
+import { api } from '@/lib/api'
+import type { CacheStats } from './types'
+
+export async function getCacheStats(): Promise<{
+  success: boolean
+  message?: string
+  data?: CacheStats
+}> {
+  const res = await api.get('/api/option/channel_affinity_cache', {
+    disableDuplicate: true,
+  } as Record<string, unknown>)
+  return res.data
+}
+
+export async function clearAllCache(): Promise<{
+  success: boolean
+  message?: string
+}> {
+  const res = await api.delete('/api/option/channel_affinity_cache', {
+    params: { all: true },
+  })
+  return res.data
+}
+
+export async function clearRuleCache(
+  ruleName: string
+): Promise<{ success: boolean; message?: string }> {
+  const res = await api.delete('/api/option/channel_affinity_cache', {
+    params: { rule_name: ruleName },
+  })
+  return res.data
+}
+
+export async function getAffinityUsageCache(params: {
+  rule_name: string
+  using_group: string
+  key_hint: string
+  key_fp: string
+}): Promise<{ success: boolean; message?: string; data?: unknown }> {
+  const res = await api.get('/api/log/channel_affinity_usage_cache', {
+    params,
+    disableDuplicate: true,
+  } as Record<string, unknown>)
+  return res.data
+}

web/default/src/features/system-settings/general/channel-affinity/cache-stats-dialog.tsx

@@ -0,0 +1,154 @@
+import { useEffect, useMemo, useState, useRef } from 'react'
+import { useTranslation } from 'react-i18next'
+import { toast } from 'sonner'
+import { formatTimestampToDate } from '@/lib/format'
+import {
+  Dialog,
+  DialogContent,
+  DialogHeader,
+  DialogTitle,
+} from '@/components/ui/dialog'
+import { getAffinityUsageCache } from './api'
+
+function formatRate(hit: number, total: number): string {
+  if (!total || total <= 0) return '-'
+  const r = (hit / total) * 100
+  if (!Number.isFinite(r)) return '-'
+  return `${r.toFixed(2)}%`
+}
+
+interface Props {
+  open: boolean
+  onOpenChange: (open: boolean) => void
+  target: {
+    rule_name: string
+    using_group: string
+    key_hint: string
+    key_fp: string
+  } | null
+}
+
+export function CacheStatsDialog(props: Props) {
+  const { t } = useTranslation()
+  const [loading, setLoading] = useState(false)
+  const [stats, setStats] = useState<Record<string, unknown> | null>(null)
+  const seqRef = useRef(0)
+
+  useEffect(() => {
+    if (!props.open || !props.target?.rule_name || !props.target?.key_fp) {
+      // eslint-disable-next-line react-hooks/set-state-in-effect
+      setStats(null)
+      return
+    }
+
+    const seq = ++seqRef.current
+
+    setLoading(true)
+
+    setStats(null)
+
+    getAffinityUsageCache(props.target)
+      .then((res) => {
+        if (seq !== seqRef.current) return
+        if (res.success) setStats((res.data as Record<string, unknown>) || {})
+        else toast.error(res.message || t('Request failed'))
+      })
+      .catch(() => {
+        if (seq !== seqRef.current) return
+        toast.error(t('Request failed'))
+      })
+      .finally(() => {
+        if (seq !== seqRef.current) return
+        setLoading(false)
+      })
+  }, [props.open, props.target, t])
+
+  const rows = useMemo(() => {
+    if (!stats) return []
+    const s = stats
+    const data: { key: string; value: string | number }[] = []
+    const hit = Number(s.hit || 0)
+    const total = Number(s.total || 0)
+
+    if (s.rule_name || props.target?.rule_name)
+      data.push({
+        key: t('Rule'),
+        value: (s.rule_name || props.target?.rule_name || '') as string,
+      })
+    if (s.using_group || props.target?.using_group)
+      data.push({
+        key: t('Group'),
+        value: (s.using_group || props.target?.using_group || '') as string,
+      })
+    if (props.target?.key_hint)
+      data.push({ key: t('Key Summary'), value: props.target.key_hint })
+    if (s.key_fp || props.target?.key_fp)
+      data.push({
+        key: t('Key Fingerprint'),
+        value: (s.key_fp || props.target?.key_fp || '') as string,
+      })
+    if (Number(s.window_seconds || 0) > 0)
+      data.push({ key: t('TTL (seconds)'), value: s.window_seconds as number })
+    if (total > 0)
+      data.push({
+        key: t('Hit Rate'),
+        value: `${hit}/${total} (${formatRate(hit, total)})`,
+      })
+    if (Number(s.last_seen_at || 0) > 0)
+      data.push({
+        key: t('Last Seen'),
+        value: formatTimestampToDate(s.last_seen_at as number | undefined),
+      })
+
+    const promptTokens = Number(s.prompt_tokens || 0)
+    const cachedTokens = Number(s.cached_tokens || 0)
+    const completionTokens = Number(s.completion_tokens || 0)
+    const totalTokens = Number(s.total_tokens || 0)
+
+    if (promptTokens > 0)
+      data.push({ key: 'Prompt tokens', value: promptTokens })
+    if (cachedTokens > 0)
+      data.push({ key: 'Cached tokens', value: cachedTokens })
+    if (completionTokens > 0)
+      data.push({ key: 'Completion tokens', value: completionTokens })
+    if (totalTokens > 0) data.push({ key: 'Total tokens', value: totalTokens })
+
+    return data
+  }, [stats, props.target, t])
+
+  return (
+    <Dialog open={props.open} onOpenChange={props.onOpenChange}>
+      <DialogContent className='sm:max-w-lg'>
+        <DialogHeader>
+          <DialogTitle>{t('Channel Affinity: Upstream Cache Hit')}</DialogTitle>
+        </DialogHeader>
+        <p className='text-muted-foreground text-xs'>
+          {t(
+            'Hit criteria: If cached tokens exist in usage, it counts as a hit.'
+          )}
+        </p>
+        {loading ? (
+          <div className='text-muted-foreground py-8 text-center text-sm'>
+            {t('Loading...')}
+          </div>
+        ) : rows.length > 0 ? (
+          <div className='space-y-2'>
+            {rows.map((row) => (
+              <div
+                key={row.key}
+                className='flex justify-between border-b pb-1 text-sm'
+              >
+                <span className='text-muted-foreground'>{row.key}</span>
+                <span className='font-medium'>{row.value}</span>
+              </div>
+            ))}
+          </div>
+        ) : (
+          <div className='text-muted-foreground py-8 text-center text-sm'>
+            {t('No data available')}
+          </div>
+        )}
+      </DialogContent>
+    </Dialog>
+  )
+}

web/default/src/features/system-settings/general/channel-affinity/constants.ts

@@ -0,0 +1,89 @@
+import type { AffinityRule } from './types'
+
+const CODEX_CLI_HEADER_PASSTHROUGH_HEADERS = [
+  'Originator',
+  'Session_id',
+  'User-Agent',
+  'X-Codex-Beta-Features',
+  'X-Codex-Turn-Metadata',
+]
+
+const CLAUDE_CLI_HEADER_PASSTHROUGH_HEADERS = [
+  'X-Stainless-Arch',
+  'X-Stainless-Lang',
+  'X-Stainless-Os',
+  'X-Stainless-Package-Version',
+  'X-Stainless-Retry-Count',
+  'X-Stainless-Runtime',
+  'X-Stainless-Runtime-Version',
+  'X-Stainless-Timeout',
+  'User-Agent',
+  'X-App',
+  'Anthropic-Beta',
+  'Anthropic-Dangerous-Direct-Browser-Access',
+  'Anthropic-Version',
+]
+
+function buildPassHeadersTemplate(headers: string[]) {
+  return {
+    operations: [
+      {
+        mode: 'pass_headers',
+        value: [...headers],
+        keep_origin: true,
+      },
+    ],
+  }
+}
+
+export type RuleTemplate = Omit<AffinityRule, 'id'>
+
+export const RULE_TEMPLATES: Record<string, RuleTemplate> = {
+  codexCli: {
+    name: 'codex cli trace',
+    model_regex: ['^gpt-.*$'],
+    path_regex: ['/v1/responses'],
+    key_sources: [{ type: 'gjson', path: 'prompt_cache_key' }],
+    param_override_template: buildPassHeadersTemplate(
+      CODEX_CLI_HEADER_PASSTHROUGH_HEADERS
+    ),
+    value_regex: '',
+    ttl_seconds: 0,
+    skip_retry_on_failure: true,
+    include_using_group: true,
+    include_model_name: false,
+    include_rule_name: true,
+  },
+  claudeCli: {
+    name: 'claude cli trace',
+    model_regex: ['^claude-.*$'],
+    path_regex: ['/v1/messages'],
+    key_sources: [{ type: 'gjson', path: 'metadata.user_id' }],
+    param_override_template: buildPassHeadersTemplate(
+      CLAUDE_CLI_HEADER_PASSTHROUGH_HEADERS
+    ),
+    value_regex: '',
+    ttl_seconds: 0,
+    skip_retry_on_failure: true,
+    include_using_group: true,
+    include_model_name: false,
+    include_rule_name: true,
+  },
+}
+
+export function makeUniqueName(
+  existingNames: Set<string>,
+  baseName: string
+): string {
+  const base = (baseName || '').trim() || 'rule'
+  if (!existingNames.has(base)) return base
+  for (let i = 2; i < 1000; i++) {
+    const n = `${base}-${i}`
+    if (!existingNames.has(n)) return n
+  }
+  return `${base}-${Date.now()}`
+}
+
+export function cloneTemplate<T>(template: T): T {
+  return JSON.parse(JSON.stringify(template))
+}

web/default/src/features/system-settings/general/channel-affinity/index.tsx

@@ -0,0 +1,696 @@
+import { useCallback, useEffect, useState } from 'react'
+import { Edit, FileText, Plus, RefreshCw, Trash2, X } from 'lucide-react'
+import { useTranslation } from 'react-i18next'
+import { toast } from 'sonner'
+import { Alert, AlertDescription } from '@/components/ui/alert'
+import { Button } from '@/components/ui/button'
+import {
+  DropdownMenu,
+  DropdownMenuContent,
+  DropdownMenuItem,
+  DropdownMenuTrigger,
+} from '@/components/ui/dropdown-menu'
+import { Input } from '@/components/ui/input'
+import { Label } from '@/components/ui/label'
+import { Separator } from '@/components/ui/separator'
+import { Switch } from '@/components/ui/switch'
+import {
+  Table,
+  TableBody,
+  TableCell,
+  TableHead,
+  TableHeader,
+  TableRow,
+} from '@/components/ui/table'
+import { Textarea } from '@/components/ui/textarea'
+import { ConfirmDialog } from '@/components/confirm-dialog'
+import { StatusBadge } from '@/components/status-badge'
+import { SettingsSection } from '../../components/settings-section'
+import { useUpdateOption } from '../../hooks/use-update-option'
+import { getCacheStats, clearAllCache, clearRuleCache } from './api'
+import { RULE_TEMPLATES, cloneTemplate, makeUniqueName } from './constants'
+import { RuleEditorDialog } from './rule-editor-dialog'
+import type { AffinityRule, CacheStats, ChannelAffinitySettings } from './types'
+
+function parseRules(jsonStr: string): AffinityRule[] {
+  try {
+    const arr = JSON.parse(jsonStr || '[]')
+    if (!Array.isArray(arr)) return []
+    return arr.map(
+      (r: Record<string, unknown>, i: number) =>
+        ({ id: i, ...r }) as AffinityRule
+    )
+  } catch {
+    return []
+  }
+}
+
+function serializeRules(rules: AffinityRule[]): string {
+  return JSON.stringify(rules.map(({ id: _, ...rest }) => rest))
+}
+
+interface Props {
+  defaultValues: ChannelAffinitySettings
+}
+
+export function ChannelAffinitySection(props: Props) {
+  const { t } = useTranslation()
+  const updateOption = useUpdateOption()
+
+  const [enabled, setEnabled] = useState(
+    props.defaultValues['channel_affinity_setting.enabled']
+  )
+  const [switchOnSuccess, setSwitchOnSuccess] = useState(
+    props.defaultValues['channel_affinity_setting.switch_on_success']
+  )
+  const [maxEntries, setMaxEntries] = useState(
+    props.defaultValues['channel_affinity_setting.max_entries']
+  )
+  const [defaultTtl, setDefaultTtl] = useState(
+    props.defaultValues['channel_affinity_setting.default_ttl_seconds']
+  )
+  const [rules, setRules] = useState<AffinityRule[]>(() =>
+    parseRules(props.defaultValues['channel_affinity_setting.rules'])
+  )
+
+  const [editMode, setEditMode] = useState<'visual' | 'json'>('visual')
+  const [jsonText, setJsonText] = useState(() =>
+    JSON.stringify(
+      parseRules(props.defaultValues['channel_affinity_setting.rules']).map(
+        ({ id: _, ...r }) => r
+      ),
+      null,
+      2
+    )
+  )
+  const [cacheStats, setCacheStats] = useState<CacheStats | null>(null)
+  const [cacheLoading, setCacheLoading] = useState(false)
+  const [saving, setSaving] = useState(false)
+
+  const [ruleEditorOpen, setRuleEditorOpen] = useState(false)
+  const [editingRule, setEditingRule] = useState<AffinityRule | null>(null)
+  const [ruleTemplateKey, setRuleTemplateKey] = useState<string | null>(null)
+  const [clearAllDialogOpen, setClearAllDialogOpen] = useState(false)
+  const [clearRuleName, setClearRuleName] = useState<string | null>(null)
+  const [fillTemplateDialogOpen, setFillTemplateDialogOpen] = useState(false)
+
+  useEffect(() => {
+    setEnabled(props.defaultValues['channel_affinity_setting.enabled'])
+    setSwitchOnSuccess(
+      props.defaultValues['channel_affinity_setting.switch_on_success']
+    )
+    setMaxEntries(props.defaultValues['channel_affinity_setting.max_entries'])
+    setDefaultTtl(
+      props.defaultValues['channel_affinity_setting.default_ttl_seconds']
+    )
+    const parsed = parseRules(
+      props.defaultValues['channel_affinity_setting.rules']
+    )
+    setRules(parsed)
+    setJsonText(
+      JSON.stringify(
+        parsed.map(({ id: _, ...r }) => r),
+        null,
+        2
+      )
+    )
+  }, [props.defaultValues])
+
+  const refreshCache = useCallback(async () => {
+    setCacheLoading(true)
+    try {
+      const res = await getCacheStats()
+      if (res.success) setCacheStats(res.data || null)
+    } catch {
+      toast.error(t('Failed to refresh cache stats'))
+    } finally {
+      setCacheLoading(false)
+    }
+  }, [t])
+
+  useEffect(() => {
+    refreshCache()
+  }, [refreshCache])
+
+  const appendCliTemplates = () => {
+    const existingNames = new Set(
+      rules.map((r) => (r.name || '').trim()).filter((x) => x.length > 0)
+    )
+
+    const templates = Object.values(RULE_TEMPLATES).map((tpl) => {
+      const base = cloneTemplate(tpl)
+      const name = makeUniqueName(existingNames, tpl.name)
+      existingNames.add(name)
+      return { ...base, name }
+    })
+
+    setRules((prev) =>
+      [...prev, ...templates].map((r, idx) => ({ ...r, id: idx }))
+    )
+    toast.success(t('Templates appended'))
+    setFillTemplateDialogOpen(false)
+  }
+
+  const handleFillTemplates = () => {
+    if (rules.length === 0) {
+      appendCliTemplates()
+    } else {
+      setFillTemplateDialogOpen(true)
+    }
+  }
+
+  const handleSave = async () => {
+    let rulesJson: string
+    if (editMode === 'json') {
+      try {
+        const parsed = JSON.parse(jsonText)
+        if (!Array.isArray(parsed)) {
+          toast.error(t('Rules JSON must be an array'))
+          return
+        }
+        rulesJson = JSON.stringify(parsed)
+      } catch {
+        toast.error(t('Invalid rules JSON format'))
+        return
+      }
+    } else {
+      rulesJson = serializeRules(rules)
+    }
+
+    setSaving(true)
+    try {
+      const updates: { key: string; value: string }[] = []
+
+      if (enabled !== props.defaultValues['channel_affinity_setting.enabled'])
+        updates.push({
+          key: 'channel_affinity_setting.enabled',
+          value: String(enabled),
+        })
+      if (
+        switchOnSuccess !==
+        props.defaultValues['channel_affinity_setting.switch_on_success']
+      )
+        updates.push({
+          key: 'channel_affinity_setting.switch_on_success',
+          value: String(switchOnSuccess),
+        })
+      if (
+        maxEntries !==
+        props.defaultValues['channel_affinity_setting.max_entries']
+      )
+        updates.push({
+          key: 'channel_affinity_setting.max_entries',
+          value: String(maxEntries),
+        })
+      if (
+        defaultTtl !==
+        props.defaultValues['channel_affinity_setting.default_ttl_seconds']
+      )
+        updates.push({
+          key: 'channel_affinity_setting.default_ttl_seconds',
+          value: String(defaultTtl),
+        })
+
+      const origRules = props.defaultValues['channel_affinity_setting.rules']
+      const origSerialized = (() => {
+        try {
+          return JSON.stringify(JSON.parse(origRules || '[]'))
+        } catch {
+          return '[]'
+        }
+      })()
+      if (rulesJson !== origSerialized) {
+        updates.push({
+          key: 'channel_affinity_setting.rules',
+          value: rulesJson,
+        })
+      }
+
+      if (updates.length === 0) {
+        toast.info(t('No changes'))
+        return
+      }
+
+      for (const u of updates) {
+        await updateOption.mutateAsync(u)
+      }
+      toast.success(t('Saved successfully'))
+    } catch {
+      toast.error(t('Failed to save'))
+    } finally {
+      setSaving(false)
+    }
+  }
+
+  const handleRuleSave = (rule: AffinityRule) => {
+    setRules((prev) => {
+      const existIdx = prev.findIndex(
+        (r) => r.id === rule.id || (rule.name && r.name === editingRule?.name)
+      )
+      if (existIdx >= 0) {
+        const next = [...prev]
+        next[existIdx] = { ...rule, id: existIdx }
+        return next
+      }
+      return [...prev, { ...rule, id: prev.length }]
+    })
+    setEditingRule(null)
+  }
+
+  const handleDeleteRule = (idx: number) => {
+    setRules((prev) =>
+      prev.filter((_, i) => i !== idx).map((r, i) => ({ ...r, id: i }))
+    )
+    toast.success(t('Deleted successfully'))
+  }
+
+  const handleClearAll = async () => {
+    const res = await clearAllCache()
+    if (res.success) {
+      toast.success(t('Cleared'))
+      refreshCache()
+    }
+    setClearAllDialogOpen(false)
+  }
+
+  const handleClearRule = async () => {
+    if (!clearRuleName) return
+    const res = await clearRuleCache(clearRuleName)
+    if (res.success) {
+      toast.success(t('Cleared'))
+      refreshCache()
+    }
+    setClearRuleName(null)
+  }
+
+  const switchToJsonMode = () => {
+    setJsonText(
+      JSON.stringify(
+        rules.map(({ id: _, ...r }) => r),
+        null,
+        2
+      )
+    )
+    setEditMode('json')
+  }
+
+  const switchToVisualMode = () => {
+    try {
+      const parsed = JSON.parse(jsonText)
+      if (!Array.isArray(parsed)) {
+        toast.error(t('Rules JSON must be an array'))
+        return
+      }
+      setRules(
+        parsed.map(
+          (r: Record<string, unknown>, i: number) =>
+            ({ id: i, ...r }) as AffinityRule
+        )
+      )
+      setEditMode('visual')
+    } catch {
+      toast.error(t('Invalid rules JSON format'))
+    }
+  }
+
+  return (
+    <>
+      <SettingsSection
+        title={t('Channel Affinity')}
+        description={t(
+          'Prioritize reusing the last successful channel based on keys extracted from request context (sticky routing)'
+        )}
+      >
+        <Alert>
+          <AlertDescription className='text-xs'>
+            {t(
+              'Channel affinity reuses the last successful channel based on keys extracted from the request context or JSON body.'
+            )}
+          </AlertDescription>
+        </Alert>
+
+        {/* Basic Settings */}
+        <div className='grid grid-cols-1 gap-4 md:grid-cols-3'>
+          <div className='flex items-center gap-2'>
+            <Switch checked={enabled} onCheckedChange={setEnabled} />
+            <Label>{t('Enable')}</Label>
+          </div>
+          <div className='grid gap-1.5'>
+            <Label>{t('Max Entries')}</Label>
+            <Input
+              type='number'
+              min={0}
+              value={maxEntries}
+              onChange={(e) => setMaxEntries(Number(e.target.value))}
+            />
+          </div>
+          <div className='grid gap-1.5'>
+            <Label>{t('Default TTL (seconds)')}</Label>
+            <Input
+              type='number'
+              min={0}
+              value={defaultTtl}
+              onChange={(e) => setDefaultTtl(Number(e.target.value))}
+            />
+          </div>
+        </div>
+
+        <div className='flex items-center gap-2'>
+          <Switch
+            checked={switchOnSuccess}
+            onCheckedChange={setSwitchOnSuccess}
+          />
+          <Label>{t('Switch affinity on success')}</Label>
+          <span className='text-muted-foreground text-xs'>
+            {t(
+              'If the affinity channel fails and retry succeeds on another channel, update affinity to the successful channel.'
+            )}
+          </span>
+        </div>
+
+        <Separator />
+
+        {/* Toolbar */}
+        <div className='flex flex-wrap items-center gap-2'>
+          <Button
+            variant={editMode === 'visual' ? 'default' : 'outline'}
+            size='sm'
+            onClick={editMode === 'json' ? switchToVisualMode : undefined}
+          >
+            {t('Visual')}
+          </Button>
+          <Button
+            variant={editMode === 'json' ? 'default' : 'outline'}
+            size='sm'
+            onClick={editMode === 'visual' ? switchToJsonMode : undefined}
+          >
+            JSON
+          </Button>
+          <DropdownMenu>
+            <DropdownMenuTrigger asChild>
+              <Button variant='outline' size='sm'>
+                <Plus className='mr-1 h-3 w-3' />
+                {t('Add Rule')}
+              </Button>
+            </DropdownMenuTrigger>
+            <DropdownMenuContent>
+              <DropdownMenuItem
+                onClick={() => {
+                  setEditingRule(null)
+                  setRuleTemplateKey(null)
+                  setRuleEditorOpen(true)
+                }}
+              >
+                {t('Blank Rule')}
+              </DropdownMenuItem>
+              <DropdownMenuItem
+                onClick={() => {
+                  setEditingRule(null)
+                  setRuleTemplateKey('codexCli')
+                  setRuleEditorOpen(true)
+                }}
+              >
+                Codex CLI
+              </DropdownMenuItem>
+              <DropdownMenuItem
+                onClick={() => {
+                  setEditingRule(null)
+                  setRuleTemplateKey('claudeCli')
+                  setRuleEditorOpen(true)
+                }}
+              >
+                Claude CLI
+              </DropdownMenuItem>
+            </DropdownMenuContent>
+          </DropdownMenu>
+          <Button variant='outline' size='sm' onClick={handleFillTemplates}>
+            <FileText className='mr-1 h-3 w-3' />
+            {t('Fill Templates')}
+          </Button>
+          <Button size='sm' onClick={handleSave} disabled={saving}>
+            {saving ? t('Saving...') : t('Save')}
+          </Button>
+          <Button
+            variant='outline'
+            size='sm'
+            onClick={refreshCache}
+            disabled={cacheLoading}
+          >
+            <RefreshCw
+              className={`mr-1 h-3 w-3 ${cacheLoading ? 'animate-spin' : ''}`}
+            />
+            {t('Refresh Cache')}
+          </Button>
+          <Button
+            variant='destructive'
+            size='sm'
+            onClick={() => setClearAllDialogOpen(true)}
+          >
+            {t('Clear All Cache')}
+          </Button>
+          {cacheStats && (
+            <span className='text-muted-foreground text-xs'>
+              {t('Cache Entries')}: {cacheStats.total} /{' '}
+              {cacheStats.cache_capacity}
+            </span>
+          )}
+        </div>
+
+        {/* Rules Table or JSON Editor */}
+        {editMode === 'visual' ? (
+          <div className='overflow-x-auto rounded-md border'>
+            <Table>
+              <TableHeader>
+                <TableRow>
+                  <TableHead>{t('Name')}</TableHead>
+                  <TableHead>{t('Model Regex')}</TableHead>
+                  <TableHead>{t('Key Sources')}</TableHead>
+                  <TableHead>{t('TTL')}</TableHead>
+                  <TableHead>{t('Retry')}</TableHead>
+                  <TableHead>{t('Scope')}</TableHead>
+                  <TableHead>{t('Cache')}</TableHead>
+                  <TableHead className='text-right'>{t('Actions')}</TableHead>
+                </TableRow>
+              </TableHeader>
+              <TableBody>
+                {rules.length === 0 ? (
+                  <TableRow>
+                    <TableCell
+                      colSpan={8}
+                      className='text-muted-foreground py-8 text-center'
+                    >
+                      {t('No rules yet')}
+                    </TableCell>
+                  </TableRow>
+                ) : (
+                  rules.map((rule, idx) => (
+                    <TableRow key={idx}>
+                      <TableCell className='font-medium'>
+                        {rule.name || '-'}
+                      </TableCell>
+                      <TableCell>
+                        <div className='text-muted-foreground flex items-center gap-1.5 text-xs font-medium'>
+                          {(rule.model_regex || []).length > 0 && (
+                            <span
+                              className='size-1.5 shrink-0 rounded-full bg-slate-400'
+                              aria-hidden='true'
+                            />
+                          )}
+                          {(rule.model_regex || [])
+                            .slice(0, 2)
+                            .map((r, i, arr) => (
+                              <span
+                                key={i}
+                                className='flex items-center gap-1.5'
+                              >
+                                {r}
+                                {i < arr.length - 1 && (
+                                  <span className='text-muted-foreground/30'>
+                                    ·
+                                  </span>
+                                )}
+                              </span>
+                            ))}
+                          {(rule.model_regex || []).length > 2 && (
+                            <span className='text-muted-foreground/50'>
+                              +{(rule.model_regex || []).length - 2}
+                            </span>
+                          )}
+                        </div>
+                      </TableCell>
+                      <TableCell>
+                        <div className='text-muted-foreground flex items-center gap-1.5 text-xs font-medium'>
+                          {(rule.key_sources || []).length > 0 && (
+                            <span
+                              className='size-1.5 shrink-0 rounded-full bg-slate-400'
+                              aria-hidden='true'
+                            />
+                          )}
+                          {(rule.key_sources || [])
+                            .slice(0, 2)
+                            .map((src, i, arr) => (
+                              <span
+                                key={i}
+                                className='flex items-center gap-1.5'
+                              >
+                                {src.type}:
+                                {src.type === 'gjson' ? src.path : src.key}
+                                {i < arr.length - 1 && (
+                                  <span className='text-muted-foreground/30'>
+                                    ·
+                                  </span>
+                                )}
+                              </span>
+                            ))}
+                          {(rule.key_sources || []).length > 2 && (
+                            <span className='text-muted-foreground/50'>
+                              +{(rule.key_sources || []).length - 2}
+                            </span>
+                          )}
+                        </div>
+                      </TableCell>
+                      <TableCell>{rule.ttl_seconds || '-'}</TableCell>
+                      <TableCell>
+                        <StatusBadge
+                          label={
+                            rule.skip_retry_on_failure
+                              ? t('No Retry')
+                              : t('Retry')
+                          }
+                          variant={
+                            rule.skip_retry_on_failure ? 'danger' : 'neutral'
+                          }
+                          copyable={false}
+                        />
+                      </TableCell>
+                      <TableCell>
+                        {(() => {
+                          const scopeItems = [
+                            rule.include_using_group && t('Group'),
+                            rule.include_model_name && t('Model'),
+                            rule.include_rule_name && t('Rule'),
+                          ].filter(Boolean) as string[]
+                          if (scopeItems.length === 0) return '-'
+                          return (
+                            <div className='text-muted-foreground flex items-center gap-1.5 text-xs font-medium'>
+                              <span
+                                className='size-1.5 shrink-0 rounded-full bg-slate-400'
+                                aria-hidden='true'
+                              />
+                              {scopeItems.map((item, idx, arr) => (
+                                <span
+                                  key={idx}
+                                  className='flex items-center gap-1.5'
+                                >
+                                  {item}
+                                  {idx < arr.length - 1 && (
+                                    <span className='text-muted-foreground/30'>
+                                      ·
+                                    </span>
+                                  )}
+                                </span>
+                              ))}
+                            </div>
+                          )
+                        })()}
+                      </TableCell>
+                      <TableCell>
+                        {rule.include_rule_name && cacheStats?.by_rule_name
+                          ? cacheStats.by_rule_name[rule.name] || 0
+                          : 'N/A'}
+                      </TableCell>
+                      <TableCell className='text-right'>
+                        <div className='flex justify-end gap-1'>
+                          {rule.include_rule_name && (
+                            <Button
+                              variant='ghost'
+                              size='icon'
+                              className='h-7 w-7'
+                              onClick={() => setClearRuleName(rule.name)}
+                              title={t('Clear cache for this rule')}
+                            >
+                              <X className='h-3 w-3' />
+                            </Button>
+                          )}
+                          <Button
+                            variant='ghost'
+                            size='icon'
+                            className='h-7 w-7'
+                            onClick={() => {
+                              setEditingRule(rule)
+                              setRuleTemplateKey(null)
+                              setRuleEditorOpen(true)
+                            }}
+                          >
+                            <Edit className='h-3 w-3' />
+                          </Button>
+                          <Button
+                            variant='ghost'
+                            size='icon'
+                            className='h-7 w-7'
+                            onClick={() => handleDeleteRule(idx)}
+                          >
+                            <Trash2 className='h-3 w-3' />
+                          </Button>
+                        </div>
+                      </TableCell>
+                    </TableRow>
+                  ))
+                )}
+              </TableBody>
+            </Table>
+          </div>
+        ) : (
+          <div className='grid gap-1.5'>
+            <Label>{t('Rules JSON')}</Label>
+            <Textarea
+              className='min-h-[300px] font-mono text-xs'
+              value={jsonText}
+              onChange={(e) => setJsonText(e.target.value)}
+            />
+          </div>
+        )}
+      </SettingsSection>
+
+      <RuleEditorDialog
+        open={ruleEditorOpen}
+        onOpenChange={setRuleEditorOpen}
+        rule={editingRule}
+        onSave={handleRuleSave}
+        templateKey={ruleTemplateKey}
+      />
+
+      <ConfirmDialog
+        open={clearAllDialogOpen}
+        onOpenChange={setClearAllDialogOpen}
+        title={t('Confirm clearing all channel affinity cache')}
+        desc={t(
+          'This will delete all channel affinity cache entries still in memory.'
+        )}
+        handleConfirm={handleClearAll}
+        destructive
+      />
+
+      {clearRuleName !== null && (
+        <ConfirmDialog
+          open
+          onOpenChange={(v) => !v && setClearRuleName(null)}
+          title={t('Confirm clearing cache for this rule')}
+          desc={`${t('Rule')}: ${clearRuleName}`}
+          handleConfirm={handleClearRule}
+          destructive
+        />
+      )}
+
+      <ConfirmDialog
+        open={fillTemplateDialogOpen}
+        onOpenChange={setFillTemplateDialogOpen}
+        title={t('Fill Codex CLI / Claude CLI Templates')}
+        desc={t(
+          'This will append 2 template rules (Codex CLI and Claude CLI) to the existing rule list.'
+        )}
+        handleConfirm={appendCliTemplates}
+      />
+    </>
+  )
+}

web/default/src/features/system-settings/general/channel-affinity/rule-editor-dialog.tsx

@@ -0,0 +1,429 @@
+import { useEffect, useState } from 'react'
+import { useForm } from 'react-hook-form'
+import { Plus, Trash2 } from 'lucide-react'
+import { useTranslation } from 'react-i18next'
+import { toast } from 'sonner'
+import { Button } from '@/components/ui/button'
+import {
+  Collapsible,
+  CollapsibleContent,
+  CollapsibleTrigger,
+} from '@/components/ui/collapsible'
+import {
+  Dialog,
+  DialogContent,
+  DialogFooter,
+  DialogHeader,
+  DialogTitle,
+} from '@/components/ui/dialog'
+import { Input } from '@/components/ui/input'
+import { Label } from '@/components/ui/label'
+import {
+  Select,
+  SelectContent,
+  SelectItem,
+  SelectTrigger,
+  SelectValue,
+} from '@/components/ui/select'
+import { Separator } from '@/components/ui/separator'
+import { Switch } from '@/components/ui/switch'
+import { Textarea } from '@/components/ui/textarea'
+import { RULE_TEMPLATES } from './constants'
+import type { AffinityRule, KeySource } from './types'
+
+const KEY_SOURCE_TYPES = ['context_int', 'context_string', 'gjson'] as const
+
+const CONTEXT_KEY_PRESETS = [
+  'id',
+  'token_id',
+  'token_key',
+  'token_group',
+  'group',
+  'username',
+  'user_group',
+  'user_email',
+  'specific_channel_id',
+]
+
+interface RuleFormValues {
+  name: string
+  model_regex_text: string
+  path_regex_text: string
+  user_agent_include_text: string
+  value_regex: string
+  ttl_seconds: number
+  skip_retry_on_failure: boolean
+  include_using_group: boolean
+  include_model_name: boolean
+  include_rule_name: boolean
+  param_override_template_json: string
+}
+
+function normalizeStringList(text: string): string[] {
+  return text
+    .split('\n')
+    .map((s) => s.trim())
+    .filter((s) => s.length > 0)
+}
+
+function normalizeKeySource(src: Partial<KeySource>): KeySource {
+  const type = (src?.type || 'gjson') as KeySource['type']
+  if (type === 'gjson') return { type, key: '', path: src?.path || '' }
+  return { type, key: src?.key || '', path: '' }
+}
+
+interface Props {
+  open: boolean
+  onOpenChange: (open: boolean) => void
+  rule: AffinityRule | null
+  onSave: (rule: AffinityRule) => void
+  templateKey?: string | null
+}
+
+export function RuleEditorDialog(props: Props) {
+  const { t } = useTranslation()
+  const isEdit = !!props.rule?.name
+  const [keySources, setKeySources] = useState<KeySource[]>([
+    { type: 'gjson', path: '' },
+  ])
+  const [advancedOpen, setAdvancedOpen] = useState(false)
+
+  const form = useForm<RuleFormValues>({
+    defaultValues: {
+      name: '',
+      model_regex_text: '',
+      path_regex_text: '',
+      user_agent_include_text: '',
+      value_regex: '',
+      ttl_seconds: 0,
+      skip_retry_on_failure: false,
+      include_using_group: true,
+      include_model_name: false,
+      include_rule_name: true,
+      param_override_template_json: '',
+    },
+  })
+
+  const resetFromRule = (r: Partial<AffinityRule>) => {
+    form.reset({
+      name: r.name || '',
+      model_regex_text: (r.model_regex || []).join('\n'),
+      path_regex_text: (r.path_regex || []).join('\n'),
+      user_agent_include_text: (r.user_agent_include || []).join('\n'),
+      value_regex: r.value_regex || '',
+      ttl_seconds: r.ttl_seconds || 0,
+      skip_retry_on_failure: !!r.skip_retry_on_failure,
+      include_using_group: r.include_using_group ?? true,
+      include_model_name: !!r.include_model_name,
+      include_rule_name: r.include_rule_name ?? true,
+      param_override_template_json: r.param_override_template
+        ? JSON.stringify(r.param_override_template, null, 2)
+        : '',
+    })
+    const sources = (r.key_sources || []).map(normalizeKeySource)
+    setKeySources(sources.length > 0 ? sources : [{ type: 'gjson', path: '' }])
+    if (r.param_override_template) setAdvancedOpen(true)
+  }
+
+  useEffect(() => {
+    if (!props.open) return
+
+    if (props.rule) {
+      resetFromRule(props.rule)
+    } else if (props.templateKey && RULE_TEMPLATES[props.templateKey]) {
+      resetFromRule(RULE_TEMPLATES[props.templateKey])
+    } else {
+      form.reset({
+        name: '',
+        model_regex_text: '',
+        path_regex_text: '',
+        user_agent_include_text: '',
+        value_regex: '',
+        ttl_seconds: 0,
+        skip_retry_on_failure: false,
+        include_using_group: true,
+        include_model_name: false,
+        include_rule_name: true,
+        param_override_template_json: '',
+      })
+      setKeySources([{ type: 'gjson', path: '' }])
+    }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [props.open, props.rule, props.templateKey])
+
+  const handleSave = (values: RuleFormValues) => {
+    const modelRegex = normalizeStringList(values.model_regex_text)
+    if (modelRegex.length === 0) {
+      toast.error(t('At least one model regex pattern is required'))
+      return
+    }
+
+    const validKeySources = keySources
+      .map(normalizeKeySource)
+      .filter((s) => s.type && (s.type === 'gjson' ? s.path : s.key))
+    if (validKeySources.length === 0) {
+      toast.error(t('At least one valid key source is required'))
+      return
+    }
+
+    let paramTemplate: Record<string, unknown> | null = null
+    if (values.param_override_template_json.trim()) {
+      try {
+        const parsed = JSON.parse(values.param_override_template_json)
+        if (
+          typeof parsed !== 'object' ||
+          Array.isArray(parsed) ||
+          parsed === null
+        ) {
+          toast.error(t('Parameter override template must be a JSON object'))
+          return
+        }
+        paramTemplate = parsed
+      } catch {
+        toast.error(t('Invalid JSON in parameter override template'))
+        return
+      }
+    }
+
+    const rule: AffinityRule = {
+      id: props.rule?.id,
+      name: values.name.trim(),
+      model_regex: modelRegex,
+      path_regex: normalizeStringList(values.path_regex_text),
+      user_agent_include: normalizeStringList(values.user_agent_include_text),
+      key_sources: validKeySources,
+      value_regex: values.value_regex.trim(),
+      ttl_seconds: Number(values.ttl_seconds || 0),
+      skip_retry_on_failure: values.skip_retry_on_failure,
+      include_using_group: values.include_using_group,
+      include_model_name: values.include_model_name,
+      include_rule_name: values.include_rule_name,
+      param_override_template: paramTemplate,
+    }
+
+    props.onSave(rule)
+    props.onOpenChange(false)
+  }
+
+  return (
+    <Dialog open={props.open} onOpenChange={props.onOpenChange}>
+      <DialogContent className='max-h-[85vh] max-w-2xl overflow-y-auto'>
+        <DialogHeader>
+          <DialogTitle>{isEdit ? t('Edit Rule') : t('Add Rule')}</DialogTitle>
+        </DialogHeader>
+
+        <form onSubmit={form.handleSubmit(handleSave)} className='space-y-4'>
+          <div className='grid gap-1.5'>
+            <Label>{t('Name')} *</Label>
+            <Input
+              placeholder='prefer-by-conversation-id'
+              {...form.register('name', { required: true })}
+            />
+          </div>
+
+          <div className='grid grid-cols-2 gap-3'>
+            <div className='grid gap-1.5'>
+              <Label>{t('Model Regex (one per line)')} *</Label>
+              <Textarea
+                rows={4}
+                placeholder={'^gpt-4o.*$\n^claude-3.*$'}
+                {...form.register('model_regex_text', { required: true })}
+              />
+            </div>
+            <div className='grid gap-1.5'>
+              <Label>{t('Path Regex (one per line)')}</Label>
+              <Textarea
+                rows={4}
+                placeholder='/v1/chat/completions'
+                {...form.register('path_regex_text')}
+              />
+            </div>
+          </div>
+
+          <div className='flex items-center gap-2'>
+            <Switch
+              checked={form.watch('skip_retry_on_failure')}
+              onCheckedChange={(v) => form.setValue('skip_retry_on_failure', v)}
+            />
+            <Label>{t('Skip retry on failure')}</Label>
+          </div>
+
+          <Separator />
+
+          {/* Key Sources */}
+          <div>
+            <div className='mb-2 flex items-center justify-between'>
+              <Label>{t('Key Sources')}</Label>
+              <Button
+                type='button'
+                variant='outline'
+                size='sm'
+                onClick={() =>
+                  setKeySources((prev) => [
+                    ...prev,
+                    { type: 'gjson', path: '' },
+                  ])
+                }
+              >
+                <Plus className='mr-1 h-3 w-3' />
+                {t('Add')}
+              </Button>
+            </div>
+            <p className='text-muted-foreground mb-2 text-xs'>
+              {t('Common Keys')}: {CONTEXT_KEY_PRESETS.join(', ')}
+            </p>
+            <div className='space-y-2'>
+              {keySources.map((src, idx) => (
+                <div key={idx} className='flex items-center gap-2'>
+                  <Select
+                    value={src.type}
+                    onValueChange={(v: KeySource['type']) => {
+                      const next = [...keySources]
+                      next[idx] = normalizeKeySource({ ...src, type: v })
+                      setKeySources(next)
+                    }}
+                  >
+                    <SelectTrigger className='w-[160px]'>
+                      <SelectValue />
+                    </SelectTrigger>
+                    <SelectContent>
+                      {KEY_SOURCE_TYPES.map((t) => (
+                        <SelectItem key={t} value={t}>
+                          {t}
+                        </SelectItem>
+                      ))}
+                    </SelectContent>
+                  </Select>
+                  <Input
+                    className='flex-1'
+                    placeholder={
+                      src.type === 'gjson'
+                        ? 'metadata.conversation_id'
+                        : 'user_id'
+                    }
+                    value={
+                      src.type === 'gjson' ? src.path || '' : src.key || ''
+                    }
+                    onChange={(e) => {
+                      const next = [...keySources]
+                      if (src.type === 'gjson') {
+                        next[idx] = { ...src, path: e.target.value }
+                      } else {
+                        next[idx] = { ...src, key: e.target.value }
+                      }
+                      setKeySources(next)
+                    }}
+                  />
+                  <Button
+                    type='button'
+                    variant='ghost'
+                    size='icon'
+                    onClick={() =>
+                      setKeySources((prev) => prev.filter((_, i) => i !== idx))
+                    }
+                  >
+                    <Trash2 className='h-4 w-4' />
+                  </Button>
+                </div>
+              ))}
+            </div>
+          </div>
+
+          <Separator />
+
+          {/* Advanced */}
+          <Collapsible open={advancedOpen} onOpenChange={setAdvancedOpen}>
+            <CollapsibleTrigger asChild>
+              <Button
+                type='button'
+                variant='ghost'
+                className='w-full justify-start'
+              >
+                {advancedOpen ? '▼' : '▶'} {t('Advanced Settings')}
+              </Button>
+            </CollapsibleTrigger>
+            <CollapsibleContent className='space-y-3 pt-2'>
+              <div className='grid gap-1.5'>
+                <Label>{t('User-Agent include (one per line)')}</Label>
+                <Textarea
+                  rows={3}
+                  placeholder='curl&#10;PostmanRuntime'
+                  {...form.register('user_agent_include_text')}
+                />
+              </div>
+
+              <div className='grid grid-cols-2 gap-3'>
+                <div className='grid gap-1.5'>
+                  <Label>{t('Value Regex')}</Label>
+                  <Input
+                    placeholder='^[-0-9A-Za-z._:]{1,128}$'
+                    {...form.register('value_regex')}
+                  />
+                </div>
+                <div className='grid gap-1.5'>
+                  <Label>{t('TTL (seconds, 0 = default)')}</Label>
+                  <Input
+                    type='number'
+                    min={0}
+                    {...form.register('ttl_seconds')}
+                  />
+                </div>
+              </div>
+
+              <div className='grid gap-1.5'>
+                <Label>{t('Parameter Override Template (JSON)')}</Label>
+                <Textarea
+                  rows={5}
+                  placeholder='{"operations": [...]}'
+                  {...form.register('param_override_template_json')}
+                  className='font-mono text-xs'
+                />
+              </div>
+
+              <div className='grid grid-cols-3 gap-3'>
+                <div className='flex items-center gap-2'>
+                  <Switch
+                    checked={form.watch('include_using_group')}
+                    onCheckedChange={(v) =>
+                      form.setValue('include_using_group', v)
+                    }
+                  />
+                  <Label className='text-xs'>{t('Include Group')}</Label>
+                </div>
+                <div className='flex items-center gap-2'>
+                  <Switch
+                    checked={form.watch('include_model_name')}
+                    onCheckedChange={(v) =>
+                      form.setValue('include_model_name', v)
+                    }
+                  />
+                  <Label className='text-xs'>{t('Include Model')}</Label>
+                </div>
+                <div className='flex items-center gap-2'>
+                  <Switch
+                    checked={form.watch('include_rule_name')}
+                    onCheckedChange={(v) =>
+                      form.setValue('include_rule_name', v)
+                    }
+                  />
+                  <Label className='text-xs'>{t('Include Rule Name')}</Label>
+                </div>
+              </div>
+            </CollapsibleContent>
+          </Collapsible>
+
+          <DialogFooter>
+            <Button
+              type='button'
+              variant='outline'
+              onClick={() => props.onOpenChange(false)}
+            >
+              {t('Cancel')}
+            </Button>
+            <Button type='submit'>{t('Save')}</Button>
+          </DialogFooter>
+        </form>
+      </DialogContent>
+    </Dialog>
+  )
+}

web/default/src/features/system-settings/general/channel-affinity/types.ts

@@ -0,0 +1,38 @@
+export interface KeySource {
+  type: 'context_int' | 'context_string' | 'gjson'
+  key?: string
+  path?: string
+}
+
+export interface AffinityRule {
+  id?: number
+  name: string
+  model_regex: string[]
+  path_regex: string[]
+  user_agent_include?: string[]
+  key_sources: KeySource[]
+  value_regex?: string
+  ttl_seconds: number
+  skip_retry_on_failure: boolean
+  include_using_group: boolean
+  include_model_name: boolean
+  include_rule_name: boolean
+  param_override_template?: Record<string, unknown> | null
+}
+
+export interface CacheStats {
+  enabled: boolean
+  total: number
+  unknown: number
+  by_rule_name: Record<string, number>
+  cache_capacity: number
+  cache_algo: string
+}
+
+export interface ChannelAffinitySettings {
+  'channel_affinity_setting.enabled': boolean
+  'channel_affinity_setting.switch_on_success': boolean
+  'channel_affinity_setting.max_entries': number
+  'channel_affinity_setting.default_ttl_seconds': number
+  'channel_affinity_setting.rules': string
+}