🚀 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/playground/api.ts

@@ -0,0 +1,59 @@
+import { api } from '@/lib/api'
+import { API_ENDPOINTS } from './constants'
+import type {
+  ChatCompletionRequest,
+  ChatCompletionResponse,
+  ModelOption,
+  GroupOption,
+} from './types'
+
+/**
+ * Send chat completion request (non-streaming)
+ */
+export async function sendChatCompletion(
+  payload: ChatCompletionRequest
+): Promise<ChatCompletionResponse> {
+  const res = await api.post(API_ENDPOINTS.CHAT_COMPLETIONS, payload, {
+    skipErrorHandler: true,
+  } as Record<string, unknown>)
+  return res.data
+}
+
+/**
+ * Get user available models
+ */
+export async function getUserModels(): Promise<ModelOption[]> {
+  const res = await api.get(API_ENDPOINTS.USER_MODELS)
+  const { data } = res
+
+  if (!data.success || !Array.isArray(data.data)) {
+    return []
+  }
+
+  return data.data.map((model: string) => ({
+    label: model,
+    value: model,
+  }))
+}
+
+/**
+ * Get user groups
+ */
+export async function getUserGroups(): Promise<GroupOption[]> {
+  const res = await api.get(API_ENDPOINTS.USER_GROUPS)
+  const { data } = res
+
+  if (!data.success || !data.data) {
+    return []
+  }
+
+  const groupData = data.data as Record<string, { desc: string; ratio: number }>
+
+  // label is for button display (name only); desc is for dropdown content
+  return Object.entries(groupData).map(([group, info]) => ({
+    label: group,
+    value: group,
+    ratio: info.ratio,
+    desc: info.desc,
+  }))
+}

web/default/src/features/playground/components/message-action-button.tsx

@@ -0,0 +1,51 @@
+import type { LucideIcon } from 'lucide-react'
+import { Button } from '@/components/ui/button'
+import {
+  Tooltip,
+  TooltipContent,
+  TooltipTrigger,
+} from '@/components/ui/tooltip'
+import { MESSAGE_ACTION_BUTTON_STYLES } from '../constants'
+
+interface MessageActionButtonProps {
+  icon: LucideIcon
+  label: string
+  onClick: () => void
+  disabled?: boolean
+  className?: string
+  variant?: 'default' | 'destructive'
+}
+
+export function MessageActionButton({
+  icon: Icon,
+  label,
+  onClick,
+  disabled = false,
+  className = '',
+  variant = 'default',
+}: MessageActionButtonProps) {
+  const baseStyle =
+    variant === 'destructive'
+      ? MESSAGE_ACTION_BUTTON_STYLES.DELETE
+      : MESSAGE_ACTION_BUTTON_STYLES.BASE
+
+  return (
+    <Tooltip>
+      <TooltipTrigger asChild>
+        <Button
+          variant='ghost'
+          size='icon'
+          className={`${baseStyle} ${className}`}
+          onClick={onClick}
+          disabled={disabled}
+          aria-label={label}
+        >
+          <Icon className={MESSAGE_ACTION_BUTTON_STYLES.ICON} />
+        </Button>
+      </TooltipTrigger>
+      <TooltipContent>
+        <p>{label}</p>
+      </TooltipContent>
+    </Tooltip>
+  )
+}

web/default/src/features/playground/components/message-actions.tsx

@@ -0,0 +1,110 @@
+import { Copy, Check, RefreshCw, Edit, Trash2 } from 'lucide-react'
+import { toast } from 'sonner'
+import { useCopyToClipboard } from '@/hooks/use-copy-to-clipboard'
+import { TooltipProvider } from '@/components/ui/tooltip'
+import { MESSAGE_ACTION_LABELS } from '../constants'
+import { useMessageActionGuard } from '../hooks/use-message-action-guard'
+import type { Message } from '../types'
+import { MessageActionButton } from './message-action-button'
+
+interface MessageActionsProps {
+  message: Message
+  onCopy?: (message: Message) => void
+  onRegenerate?: (message: Message) => void
+  onEdit?: (message: Message) => void
+  onDelete?: (message: Message) => void
+  isGenerating?: boolean
+  alwaysVisible?: boolean
+  className?: string
+}
+
+export function MessageActions({
+  message,
+  onCopy,
+  onRegenerate,
+  onEdit,
+  onDelete,
+  isGenerating = false,
+  alwaysVisible = false,
+  className = '',
+}: MessageActionsProps) {
+  const { copiedText, copyToClipboard } = useCopyToClipboard()
+  const { guardAction } = useMessageActionGuard(isGenerating)
+
+  const isAssistant = message.from === 'assistant'
+  const hasContent = message.versions.some((v) => v.content)
+  const isLoading =
+    message.status === 'loading' || message.status === 'streaming'
+  const content = message.versions[0]?.content || ''
+  const isCopied = copiedText === content
+
+  const handleCopy = () => {
+    if (!content) {
+      toast.warning(MESSAGE_ACTION_LABELS.NO_CONTENT)
+      return
+    }
+    copyToClipboard(content)
+    onCopy?.(message)
+  }
+
+  const handleRegenerate = guardAction(() => onRegenerate?.(message))
+  const handleEdit = guardAction(() => onEdit?.(message))
+  const handleDelete = guardAction(() => onDelete?.(message))
+
+  const visibilityClass = alwaysVisible
+    ? 'opacity-100'
+    : 'opacity-0 group-hover:opacity-100 max-md:opacity-100'
+
+  return (
+    <TooltipProvider delayDuration={300}>
+      <div
+        className={`flex items-center gap-0.5 transition-opacity ${visibilityClass} ${className}`}
+      >
+        {/* Copy */}
+        {hasContent && (
+          <MessageActionButton
+            icon={isCopied ? Check : Copy}
+            label={
+              isCopied
+                ? MESSAGE_ACTION_LABELS.COPIED
+                : MESSAGE_ACTION_LABELS.COPY
+            }
+            onClick={handleCopy}
+            className={isCopied ? 'text-green-600' : ''}
+          />
+        )}
+
+        {/* Regenerate - only for assistant messages */}
+        {isAssistant && !isLoading && onRegenerate && (
+          <MessageActionButton
+            icon={RefreshCw}
+            label={MESSAGE_ACTION_LABELS.REGENERATE}
+            onClick={handleRegenerate}
+            disabled={isGenerating}
+          />
+        )}
+
+        {/* Edit */}
+        {hasContent && onEdit && (
+          <MessageActionButton
+            icon={Edit}
+            label={MESSAGE_ACTION_LABELS.EDIT}
+            onClick={handleEdit}
+            disabled={isGenerating}
+          />
+        )}
+
+        {/* Delete */}
+        {onDelete && (
+          <MessageActionButton
+            icon={Trash2}
+            label={MESSAGE_ACTION_LABELS.DELETE}
+            onClick={handleDelete}
+            disabled={isGenerating}
+            variant='destructive'
+          />
+        )}
+      </div>
+    </TooltipProvider>
+  )
+}

web/default/src/features/playground/components/message-error.tsx

@@ -0,0 +1,61 @@
+import { AlertCircle, AlertTriangle, Settings } from 'lucide-react'
+import { useTranslation } from 'react-i18next'
+import { useAuthStore } from '@/stores/auth-store'
+import { Alert, AlertDescription, AlertTitle } from '@/components/ui/alert'
+import { Button } from '@/components/ui/button'
+import { MESSAGE_STATUS } from '../constants'
+import type { Message } from '../types'
+
+interface MessageErrorProps {
+  message: Message
+  className?: string
+}
+
+/**
+ * Display error messages using Alert component
+ * Following ai-elements pattern for error handling
+ */
+export function MessageError({ message, className = '' }: MessageErrorProps) {
+  const { t } = useTranslation()
+  const user = useAuthStore((s) => s.auth.user)
+  const isAdmin = user?.role != null && user.role >= 10
+
+  if (message.status !== MESSAGE_STATUS.ERROR) {
+    return null
+  }
+
+  const errorContent =
+    message.versions[0]?.content || 'An unknown error occurred'
+
+  if (message.errorCode === 'model_price_error') {
+    return (
+      <Alert variant='default' className={className}>
+        <AlertTriangle className='text-orange-500' />
+        <AlertTitle>{t('Model Price Not Configured')}</AlertTitle>
+        <AlertDescription className='space-y-2'>
+          <p>{errorContent}</p>
+          {isAdmin && (
+            <Button
+              variant='outline'
+              size='sm'
+              onClick={() =>
+                window.open('/console/setting?tab=ratio', '_blank')
+              }
+            >
+              <Settings className='mr-1 h-3.5 w-3.5' />
+              {t('Go to Settings')}
+            </Button>
+          )}
+        </AlertDescription>
+      </Alert>
+    )
+  }
+
+  return (
+    <Alert variant='destructive' className={className}>
+      <AlertCircle />
+      <AlertTitle>{t('Error')}</AlertTitle>
+      <AlertDescription>{errorContent}</AlertDescription>
+    </Alert>
+  )
+}

web/default/src/features/playground/components/playground-chat.tsx

@@ -0,0 +1,273 @@
+import { useEffect, useMemo, useState } from 'react'
+import { cn } from '@/lib/utils'
+import { Button } from '@/components/ui/button'
+import { Textarea } from '@/components/ui/textarea'
+import {
+  Branch,
+  BranchMessages,
+  BranchNext,
+  BranchPage,
+  BranchPrevious,
+  BranchSelector,
+} from '@/components/ai-elements/branch'
+import {
+  Conversation,
+  ConversationContent,
+  ConversationScrollButton,
+} from '@/components/ai-elements/conversation'
+import { Loader } from '@/components/ai-elements/loader'
+import { Message, MessageContent } from '@/components/ai-elements/message'
+import {
+  Reasoning,
+  ReasoningContent,
+  ReasoningTrigger,
+} from '@/components/ai-elements/reasoning'
+import { Response } from '@/components/ai-elements/response'
+import { Shimmer } from '@/components/ai-elements/shimmer'
+import {
+  Source,
+  Sources,
+  SourcesContent,
+  SourcesTrigger,
+} from '@/components/ai-elements/sources'
+import { MESSAGE_ROLES } from '../constants'
+import { getMessageContentStyles } from '../lib/message-styles'
+import { parseThinkTags } from '../lib/message-utils'
+import type { Message as MessageType } from '../types'
+import { MessageActions } from './message-actions'
+import { MessageError } from './message-error'
+
+interface PlaygroundChatProps {
+  messages: MessageType[]
+  onCopyMessage?: (message: MessageType) => void
+  onRegenerateMessage?: (message: MessageType) => void
+  onEditMessage?: (message: MessageType) => void
+  onDeleteMessage?: (message: MessageType) => void
+  isGenerating?: boolean
+  editingKey?: string | null
+  onSaveEdit?: (newContent: string) => void
+  onCancelEdit?: (open: boolean) => void
+  onSaveEditAndSubmit?: (newContent: string) => void
+}
+
+export function PlaygroundChat({
+  messages,
+  onCopyMessage,
+  onRegenerateMessage,
+  onEditMessage,
+  onDeleteMessage,
+  isGenerating = false,
+  editingKey,
+  onSaveEdit,
+  onCancelEdit,
+  onSaveEditAndSubmit,
+}: PlaygroundChatProps) {
+  const [editText, setEditText] = useState('')
+  const [originalText, setOriginalText] = useState('')
+
+  useEffect(() => {
+    if (!editingKey) return
+    const message = messages.find((m) => m.key === editingKey)
+    const content = message?.versions?.[0]?.content || ''
+    // eslint-disable-next-line react-hooks/set-state-in-effect
+    setEditText(content)
+
+    setOriginalText(content)
+  }, [editingKey, messages])
+
+  const isEditing = (key: string) => editingKey === key
+  const isEmpty = useMemo(() => !editText.trim(), [editText])
+  const isChanged = useMemo(
+    () => editText !== originalText,
+    [editText, originalText]
+  )
+  return (
+    <Conversation>
+      {/* Remove outer padding; apply padding to inner centered container to align with input */}
+      <ConversationContent className='p-0'>
+        <div className='mx-auto w-full max-w-4xl px-4 py-4'>
+          {messages.map((message, messageIndex) => {
+            const { versions = [] } = message
+            const isLastAssistantMessage =
+              messageIndex === messages.length - 1 &&
+              message.from === MESSAGE_ROLES.ASSISTANT
+            return (
+              <Branch defaultBranch={0} key={message.key}>
+                <BranchMessages>
+                  {versions.map((version, versionIndex) => (
+                    <Message
+                      className='group flex-row-reverse'
+                      from={message.from}
+                      key={`${message.key}-${version.id}-${versionIndex}`}
+                    >
+                      <div className='w-full min-w-0 flex-1 basis-full py-1'>
+                        {isEditing(message.key) ? (
+                          <div className='space-y-2'>
+                            <Textarea
+                              value={editText}
+                              onChange={(e) => setEditText(e.target.value)}
+                              className='font-mono text-sm'
+                              rows={8}
+                            />
+                            <div className='flex gap-2'>
+                              {/* Save & Submit only makes sense for user messages */}
+                              {message.from === MESSAGE_ROLES.USER && (
+                                <Button
+                                  size='sm'
+                                  onClick={() =>
+                                    onSaveEditAndSubmit?.(editText)
+                                  }
+                                  disabled={isEmpty || !isChanged}
+                                >
+                                  Save & Submit
+                                </Button>
+                              )}
+                              <Button
+                                size='sm'
+                                onClick={() => onSaveEdit?.(editText)}
+                                disabled={isEmpty || !isChanged}
+                              >
+                                Save
+                              </Button>
+                              <Button
+                                size='sm'
+                                variant='outline'
+                                onClick={() => onCancelEdit?.(false)}
+                              >
+                                Cancel
+                              </Button>
+                            </div>
+                          </div>
+                        ) : (
+                          <>
+                            {(() => {
+                              const isAssistant =
+                                message.from === MESSAGE_ROLES.ASSISTANT
+                              const hasSources = !!message.sources?.length
+                              const showReasoning =
+                                isAssistant && !!message.reasoning?.content
+                              const showLoader =
+                                isAssistant &&
+                                !message.isReasoningStreaming &&
+                                (message.status === 'loading' ||
+                                  (message.status === 'streaming' &&
+                                    !version.content))
+                              const showMessageContent =
+                                (message.from === MESSAGE_ROLES.USER ||
+                                  !message.isReasoningStreaming) &&
+                                !!version.content
+
+                              // Extract visible content (remove <think> tags for assistant messages)
+                              const displayContent = isAssistant
+                                ? parseThinkTags(version.content).visibleContent
+                                : version.content
+
+                              const actions = (
+                                <MessageActions
+                                  message={message}
+                                  onCopy={onCopyMessage}
+                                  onRegenerate={onRegenerateMessage}
+                                  onEdit={onEditMessage}
+                                  onDelete={onDeleteMessage}
+                                  isGenerating={isGenerating}
+                                  alwaysVisible={isLastAssistantMessage}
+                                  className='mt-1'
+                                />
+                              )
+
+                              return (
+                                <>
+                                  {/* Sources */}
+                                  {hasSources && (
+                                    <Sources>
+                                      <SourcesTrigger
+                                        count={message.sources!.length}
+                                      />
+                                      <SourcesContent>
+                                        {message.sources!.map(
+                                          (source, sourceIndex) => (
+                                            <Source
+                                              href={source.href}
+                                              key={`${message.key}-source-${sourceIndex}`}
+                                              title={source.title}
+                                            />
+                                          )
+                                        )}
+                                      </SourcesContent>
+                                    </Sources>
+                                  )}
+
+                                  {/* Reasoning */}
+                                  {showReasoning && (
+                                    <Reasoning
+                                      defaultOpen={true}
+                                      isStreaming={message.isReasoningStreaming}
+                                    >
+                                      <ReasoningTrigger />
+                                      <ReasoningContent>
+                                        {message.reasoning!.content}
+                                      </ReasoningContent>
+                                    </Reasoning>
+                                  )}
+
+                                  {/* Loader */}
+                                  {showLoader && (
+                                    <div className='flex items-center gap-2 py-2'>
+                                      <Loader />
+                                      <Shimmer className='text-sm' duration={1}>
+                                        Responding...
+                                      </Shimmer>
+                                    </div>
+                                  )}
+
+                                  {/* Error or Content */}
+                                  {message.status === 'error' ? (
+                                    <>
+                                      <MessageError
+                                        message={message}
+                                        className='mb-2'
+                                      />
+                                      {actions}
+                                    </>
+                                  ) : (
+                                    showMessageContent && (
+                                      <>
+                                        <MessageContent
+                                          variant='flat'
+                                          className={cn(
+                                            getMessageContentStyles()
+                                          )}
+                                        >
+                                          <Response>{displayContent}</Response>
+                                        </MessageContent>
+                                        {actions}
+                                      </>
+                                    )
+                                  )}
+                                </>
+                              )
+                            })()}
+                          </>
+                        )}
+                      </div>
+                    </Message>
+                  ))}
+                </BranchMessages>
+
+                {/* Branch selector for multiple versions */}
+                {versions.length > 1 && (
+                  <BranchSelector className='px-0' from={message.from}>
+                    <BranchPrevious />
+                    <BranchPage />
+                    <BranchNext />
+                  </BranchSelector>
+                )}
+              </Branch>
+            )
+          })}
+        </div>
+      </ConversationContent>
+      <ConversationScrollButton />
+    </Conversation>
+  )
+}

web/default/src/features/playground/components/playground-input.tsx

@@ -0,0 +1,222 @@
+import { useState } from 'react'
+import {
+  PaperclipIcon,
+  FileIcon,
+  ImageIcon,
+  ScreenShareIcon,
+  CameraIcon,
+  GlobeIcon,
+  SendIcon,
+  SquareIcon,
+  BarChartIcon,
+  BoxIcon,
+  NotepadTextIcon,
+  CodeSquareIcon,
+  GraduationCapIcon,
+} from 'lucide-react'
+import { useTranslation } from 'react-i18next'
+import { toast } from 'sonner'
+import {
+  DropdownMenu,
+  DropdownMenuContent,
+  DropdownMenuItem,
+  DropdownMenuTrigger,
+} from '@/components/ui/dropdown-menu'
+import {
+  PromptInput,
+  PromptInputButton,
+  PromptInputFooter,
+  PromptInputTextarea,
+  PromptInputTools,
+  type PromptInputMessage,
+} from '@/components/ai-elements/prompt-input'
+import { Suggestion, Suggestions } from '@/components/ai-elements/suggestion'
+import { ModelGroupSelector } from '@/components/model-group-selector'
+import type { ModelOption, GroupOption } from '../types'
+
+interface PlaygroundInputProps {
+  onSubmit: (text: string) => void
+  onStop?: () => void
+  disabled?: boolean
+  isGenerating?: boolean
+  models: ModelOption[]
+  modelValue: string
+  onModelChange: (value: string) => void
+  isModelLoading?: boolean
+  groups: GroupOption[]
+  groupValue: string
+  onGroupChange: (value: string) => void
+}
+
+const suggestions = [
+  { icon: BarChartIcon, text: 'Analyze data', color: '#76d0eb' },
+  { icon: BoxIcon, text: 'Surprise me', color: '#76d0eb' },
+  { icon: NotepadTextIcon, text: 'Summarize text', color: '#ea8444' },
+  { icon: CodeSquareIcon, text: 'Code', color: '#6c71ff' },
+  { icon: GraduationCapIcon, text: 'Get advice', color: '#76d0eb' },
+  { icon: null, text: 'More' },
+]
+
+export function PlaygroundInput({
+  onSubmit,
+  onStop,
+  disabled,
+  isGenerating,
+  models,
+  modelValue,
+  onModelChange,
+  isModelLoading = false,
+  groups,
+  groupValue,
+  onGroupChange,
+}: PlaygroundInputProps) {
+  const { t } = useTranslation()
+  const [text, setText] = useState('')
+
+  const isModelSelectDisabled =
+    disabled || isModelLoading || models.length === 0
+  const isGroupSelectDisabled = disabled || groups.length === 0
+
+  const handleSubmit = (message: PromptInputMessage) => {
+    if (!message.text?.trim() || disabled) return
+    onSubmit(message.text)
+    setText('')
+  }
+
+  const handleFileAction = (action: string) => {
+    toast.info(t('Feature in development'), {
+      description: action,
+    })
+  }
+
+  const handleSuggestionClick = (suggestion: string) => {
+    onSubmit(suggestion)
+  }
+
+  return (
+    <div className='grid shrink-0 gap-4 px-1 md:pb-4'>
+      <PromptInput
+        groupClassName='rounded-[20px] [--radius:20px]'
+        onSubmit={handleSubmit}
+      >
+        <PromptInputTextarea
+          autoComplete='off'
+          autoCorrect='off'
+          autoCapitalize='off'
+          spellCheck={false}
+          className='px-5 md:text-base'
+          disabled={disabled}
+          onChange={(event) => setText(event.target.value)}
+          placeholder={t('Ask anything')}
+          value={text}
+        />
+
+        <PromptInputFooter className='p-2.5'>
+          <PromptInputTools>
+            <DropdownMenu>
+              <DropdownMenuTrigger asChild>
+                <PromptInputButton
+                  className='!rounded-full border font-medium'
+                  disabled={disabled}
+                  variant='outline'
+                >
+                  <PaperclipIcon size={16} />
+                  <span className='hidden sm:inline'>{t('Attach')}</span>
+                  <span className='sr-only sm:hidden'>{t('Attach')}</span>
+                </PromptInputButton>
+              </DropdownMenuTrigger>
+              <DropdownMenuContent align='start'>
+                <DropdownMenuItem
+                  onClick={() => handleFileAction('upload-file')}
+                >
+                  <FileIcon className='mr-2' size={16} />
+                  {t('Upload file')}
+                </DropdownMenuItem>
+                <DropdownMenuItem
+                  onClick={() => handleFileAction('upload-photo')}
+                >
+                  <ImageIcon className='mr-2' size={16} />
+                  {t('Upload photo')}
+                </DropdownMenuItem>
+                <DropdownMenuItem
+                  onClick={() => handleFileAction('take-screenshot')}
+                >
+                  <ScreenShareIcon className='mr-2' size={16} />
+                  {t('Take screenshot')}
+                </DropdownMenuItem>
+                <DropdownMenuItem
+                  onClick={() => handleFileAction('take-photo')}
+                >
+                  <CameraIcon className='mr-2' size={16} />
+                  {t('Take photo')}
+                </DropdownMenuItem>
+              </DropdownMenuContent>
+            </DropdownMenu>
+
+            <PromptInputButton
+              className='rounded-full border font-medium'
+              disabled={disabled}
+              onClick={() => toast.info(t('Search feature in development'))}
+              variant='outline'
+            >
+              <GlobeIcon size={16} />
+              <span className='hidden sm:inline'>{t('Search')}</span>
+              <span className='sr-only sm:hidden'>{t('Search')}</span>
+            </PromptInputButton>
+          </PromptInputTools>
+
+          <div className='flex items-center gap-1.5 md:gap-2'>
+            <ModelGroupSelector
+              selectedModel={modelValue}
+              models={models}
+              onModelChange={onModelChange}
+              selectedGroup={groupValue}
+              groups={groups}
+              onGroupChange={onGroupChange}
+              disabled={isModelSelectDisabled || isGroupSelectDisabled}
+            />
+
+            {isGenerating && onStop ? (
+              <PromptInputButton
+                className='text-foreground rounded-full font-medium'
+                onClick={onStop}
+                variant='secondary'
+              >
+                <SquareIcon className='fill-current' size={16} />
+                <span className='hidden sm:inline'>{t('Stop')}</span>
+                <span className='sr-only sm:hidden'>{t('Stop')}</span>
+              </PromptInputButton>
+            ) : (
+              <PromptInputButton
+                className='text-foreground rounded-full font-medium'
+                disabled={disabled || !text.trim()}
+                type='submit'
+                variant='secondary'
+              >
+                <SendIcon size={16} />
+                <span className='hidden sm:inline'>{t('Send')}</span>
+                <span className='sr-only sm:hidden'>{t('Send')}</span>
+              </PromptInputButton>
+            )}
+          </div>
+        </PromptInputFooter>
+      </PromptInput>
+
+      <Suggestions>
+        {suggestions.map(({ icon: Icon, text, color }) => (
+          <Suggestion
+            className={`text-xs font-normal sm:text-sm ${
+              text === 'More' ? 'hidden sm:flex' : ''
+            }`}
+            key={text}
+            onClick={() => handleSuggestionClick(text)}
+            suggestion={text}
+          >
+            {Icon && <Icon size={16} style={{ color }} />}
+            {text}
+          </Suggestion>
+        ))}
+      </Suggestions>
+    </div>
+  )
+}

web/default/src/features/playground/constants.ts

@@ -0,0 +1,82 @@
+import type { PlaygroundConfig, ParameterEnabled } from './types'
+
+// Message constants
+export const MESSAGE_ROLES = {
+  USER: 'user',
+  ASSISTANT: 'assistant',
+  SYSTEM: 'system',
+} as const
+
+export const MESSAGE_STATUS = {
+  LOADING: 'loading',
+  STREAMING: 'streaming',
+  COMPLETE: 'complete',
+  ERROR: 'error',
+} as const
+
+// API endpoints
+export const API_ENDPOINTS = {
+  CHAT_COMPLETIONS: '/pg/chat/completions',
+  USER_MODELS: '/api/user/models',
+  USER_GROUPS: '/api/user/self/groups',
+} as const
+
+// Default group
+export const DEFAULT_GROUP = 'auto' as const
+
+// Default configuration
+export const DEFAULT_CONFIG: PlaygroundConfig = {
+  model: 'gpt-4o',
+  group: DEFAULT_GROUP,
+  temperature: 0.7,
+  top_p: 1,
+  max_tokens: 4096,
+  frequency_penalty: 0,
+  presence_penalty: 0,
+  seed: null,
+  stream: true,
+}
+
+export const DEFAULT_PARAMETER_ENABLED: ParameterEnabled = {
+  temperature: true,
+  top_p: true,
+  max_tokens: false,
+  frequency_penalty: true,
+  presence_penalty: true,
+  seed: false,
+}
+
+// Storage keys
+export const STORAGE_KEYS = {
+  CONFIG: 'playground_config',
+  MESSAGES: 'playground_messages',
+  PARAMETER_ENABLED: 'playground_parameter_enabled',
+} as const
+
+// Error messages
+export const ERROR_MESSAGES = {
+  API_REQUEST_ERROR: 'Request error occurred',
+  NETWORK_ERROR: 'Network connection failed or server not responding',
+  PARSE_ERROR: 'Error parsing response data',
+  STREAM_START_ERROR: 'Error establishing connection',
+  CONNECTION_CLOSED: 'Connection closed',
+  INTERRUPTED: 'Generation was interrupted',
+} as const
+
+// Message action button styles
+export const MESSAGE_ACTION_BUTTON_STYLES = {
+  BASE: 'size-7 text-muted-foreground hover:text-foreground',
+  DELETE: 'size-7 text-muted-foreground hover:text-destructive',
+  ICON: 'size-4',
+} as const
+
+// Message action labels
+export const MESSAGE_ACTION_LABELS = {
+  COPY: 'Copy',
+  COPIED: 'Copied!',
+  REGENERATE: 'Regenerate',
+  EDIT: 'Edit',
+  DELETE: 'Delete',
+  NO_CONTENT: 'No content to copy',
+  WAIT_GENERATION: 'Please wait for the current generation to complete',
+} as const

web/default/src/features/playground/hooks/index.ts

@@ -0,0 +1,4 @@
+export * from './use-playground-state'
+export * from './use-stream-request'
+export * from './use-chat-handler'
+export * from './use-message-action-guard'

web/default/src/features/playground/hooks/use-chat-handler.ts

@@ -0,0 +1,189 @@
+import { useCallback } from 'react'
+import { toast } from 'sonner'
+import { sendChatCompletion } from '../api'
+import { MESSAGE_STATUS, ERROR_MESSAGES } from '../constants'
+import {
+  buildChatCompletionPayload,
+  updateAssistantMessageWithError,
+  updateLastAssistantMessage,
+  processStreamingContent,
+  finalizeMessage,
+} from '../lib'
+import type { Message, PlaygroundConfig, ParameterEnabled } from '../types'
+import { useStreamRequest } from './use-stream-request'
+
+interface UseChatHandlerOptions {
+  config: PlaygroundConfig
+  parameterEnabled: ParameterEnabled
+  onMessageUpdate: (updater: (prev: Message[]) => Message[]) => void
+}
+
+/**
+ * Hook for handling chat message sending and receiving
+ */
+export function useChatHandler({
+  config,
+  parameterEnabled,
+  onMessageUpdate,
+}: UseChatHandlerOptions) {
+  const { sendStreamRequest, stopStream, isStreaming } = useStreamRequest()
+
+  // Handle stream update
+  const handleStreamUpdate = useCallback(
+    (type: 'reasoning' | 'content', chunk: string) => {
+      onMessageUpdate((prev) =>
+        updateLastAssistantMessage(prev, (message) => {
+          if (message.status === MESSAGE_STATUS.ERROR) return message
+
+          if (type === 'reasoning') {
+            // Direct API reasoning_content
+            return {
+              ...message,
+              reasoning: {
+                content: (message.reasoning?.content || '') + chunk,
+                duration: 0,
+              },
+              isReasoningStreaming: true,
+              status: MESSAGE_STATUS.STREAMING,
+            }
+          }
+
+          // Content streaming: handle <think> tags
+          return {
+            ...processStreamingContent(message, chunk),
+            status: MESSAGE_STATUS.STREAMING,
+          }
+        })
+      )
+    },
+    [onMessageUpdate]
+  )
+
+  // Handle stream complete
+  const handleStreamComplete = useCallback(() => {
+    onMessageUpdate((prev) =>
+      updateLastAssistantMessage(prev, (message) =>
+        message.status === MESSAGE_STATUS.COMPLETE ||
+        message.status === MESSAGE_STATUS.ERROR
+          ? message
+          : { ...finalizeMessage(message), status: MESSAGE_STATUS.COMPLETE }
+      )
+    )
+  }, [onMessageUpdate])
+
+  // Handle stream error
+  const handleStreamError = useCallback(
+    (error: string, errorCode?: string) => {
+      toast.error(error)
+      onMessageUpdate((prev) =>
+        updateAssistantMessageWithError(prev, error, errorCode)
+      )
+    },
+    [onMessageUpdate]
+  )
+
+  // Send streaming chat request
+  const sendStreamingChat = useCallback(
+    (messages: Message[]) => {
+      const payload = buildChatCompletionPayload(
+        messages,
+        config,
+        parameterEnabled
+      )
+      sendStreamRequest(
+        payload,
+        handleStreamUpdate,
+        handleStreamComplete,
+        handleStreamError
+      )
+    },
+    [
+      config,
+      parameterEnabled,
+      sendStreamRequest,
+      handleStreamUpdate,
+      handleStreamComplete,
+      handleStreamError,
+    ]
+  )
+
+  // Send non-streaming chat request
+  const sendNonStreamingChat = useCallback(
+    async (messages: Message[]) => {
+      const payload = buildChatCompletionPayload(
+        messages,
+        config,
+        parameterEnabled
+      )
+
+      try {
+        const response = await sendChatCompletion(payload)
+        const choice = response.choices?.[0]
+        if (!choice) return
+
+        onMessageUpdate((prev) =>
+          updateLastAssistantMessage(prev, (message) => ({
+            ...finalizeMessage(
+              {
+                ...message,
+                versions: [
+                  {
+                    ...message.versions[0],
+                    content: choice.message?.content || '',
+                  },
+                ],
+              },
+              choice.message?.reasoning_content
+            ),
+            status: MESSAGE_STATUS.COMPLETE,
+          }))
+        )
+      } catch (error: unknown) {
+        const err = error as {
+          response?: {
+            data?: { message?: string; error?: { code?: string } }
+          }
+          message?: string
+        }
+        handleStreamError(
+          err?.response?.data?.message ||
+            err?.message ||
+            ERROR_MESSAGES.API_REQUEST_ERROR,
+          err?.response?.data?.error?.code || undefined
+        )
+      }
+    },
+    [config, parameterEnabled, onMessageUpdate, handleStreamError]
+  )
+
+  // Send chat request (stream or non-stream based on config)
+  const sendChat = useCallback(
+    (messages: Message[]) => {
+      if (config.stream) {
+        sendStreamingChat(messages)
+      } else {
+        sendNonStreamingChat(messages)
+      }
+    },
+    [config.stream, sendStreamingChat, sendNonStreamingChat]
+  )
+
+  // Stop generation
+  const stopGeneration = useCallback(() => {
+    stopStream()
+    onMessageUpdate((prev) =>
+      updateLastAssistantMessage(prev, (message) =>
+        message.status === MESSAGE_STATUS.LOADING ||
+        message.status === MESSAGE_STATUS.STREAMING
+          ? { ...finalizeMessage(message), status: MESSAGE_STATUS.COMPLETE }
+          : message
+      )
+    )
+  }, [stopStream, onMessageUpdate])
+
+  return {
+    sendChat,
+    stopGeneration,
+    isGenerating: isStreaming,
+  }
+}

web/default/src/features/playground/hooks/use-message-action-guard.ts

@@ -0,0 +1,24 @@
+import { useCallback } from 'react'
+import { toast } from 'sonner'
+import { MESSAGE_ACTION_LABELS } from '../constants'
+
+/**
+ * Hook to guard message actions when generation is in progress
+ * Provides a wrapper that checks if generation is active before executing
+ */
+export function useMessageActionGuard(isGenerating: boolean) {
+  const guardAction = useCallback(
+    (action: () => void) => {
+      return () => {
+        if (isGenerating) {
+          toast.warning(MESSAGE_ACTION_LABELS.WAIT_GENERATION)
+          return
+        }
+        action()
+      }
+    },
+    [isGenerating]
+  )
+
+  return { guardAction }
+}

web/default/src/features/playground/hooks/use-playground-state.ts

@@ -0,0 +1,112 @@
+import { useState, useCallback } from 'react'
+import { DEFAULT_CONFIG, DEFAULT_PARAMETER_ENABLED } from '../constants'
+import {
+  loadConfig,
+  saveConfig,
+  loadParameterEnabled,
+  saveParameterEnabled,
+  loadMessages,
+  saveMessages,
+} from '../lib'
+import type {
+  Message,
+  PlaygroundConfig,
+  ParameterEnabled,
+  ModelOption,
+  GroupOption,
+} from '../types'
+
+/**
+ * Main state management hook for playground
+ */
+export function usePlaygroundState() {
+  // Load initial state from localStorage
+  const [config, setConfig] = useState<PlaygroundConfig>(() => {
+    const savedConfig = loadConfig()
+    return { ...DEFAULT_CONFIG, ...savedConfig }
+  })
+
+  const [parameterEnabled, setParameterEnabled] = useState<ParameterEnabled>(
+    () => {
+      const saved = loadParameterEnabled()
+      return { ...DEFAULT_PARAMETER_ENABLED, ...saved }
+    }
+  )
+
+  const [messages, setMessages] = useState<Message[]>(() => {
+    return loadMessages() || []
+  })
+
+  const [models, setModels] = useState<ModelOption[]>([])
+  const [groups, setGroups] = useState<GroupOption[]>([])
+
+  // Update config with automatic save
+  const updateConfig = useCallback(
+    <K extends keyof PlaygroundConfig>(key: K, value: PlaygroundConfig[K]) => {
+      setConfig((prev) => {
+        const updated = { ...prev, [key]: value }
+        saveConfig(updated)
+        return updated
+      })
+    },
+    []
+  )
+
+  // Update parameter enabled with automatic save
+  const updateParameterEnabled = useCallback(
+    (key: keyof ParameterEnabled, value: boolean) => {
+      setParameterEnabled((prev) => {
+        const updated = { ...prev, [key]: value }
+        saveParameterEnabled(updated)
+        return updated
+      })
+    },
+    []
+  )
+
+  // Update messages with automatic save
+  const updateMessages = useCallback(
+    (updater: Message[] | ((prev: Message[]) => Message[])) => {
+      setMessages((prev) => {
+        const newMessages =
+          typeof updater === 'function' ? updater(prev) : updater
+        saveMessages(newMessages)
+        return newMessages
+      })
+    },
+    []
+  )
+
+  // Clear all messages
+  const clearMessages = useCallback(() => {
+    updateMessages([])
+  }, [updateMessages])
+
+  // Reset config to defaults
+  const resetConfig = useCallback(() => {
+    setConfig(DEFAULT_CONFIG)
+    setParameterEnabled(DEFAULT_PARAMETER_ENABLED)
+    saveConfig(DEFAULT_CONFIG)
+    saveParameterEnabled(DEFAULT_PARAMETER_ENABLED)
+  }, [])
+
+  return {
+    // State
+    config,
+    parameterEnabled,
+    messages,
+    models,
+    groups,
+
+    // Setters
+    setModels,
+    setGroups,
+
+    // Actions
+    updateConfig,
+    updateParameterEnabled,
+    updateMessages,
+    clearMessages,
+    resetConfig,
+  }
+}

web/default/src/features/playground/hooks/use-stream-request.ts

@@ -0,0 +1,136 @@
+import { useCallback, useRef } from 'react'
+import { SSE } from 'sse.js'
+import { getCommonHeaders } from '@/lib/api'
+import { API_ENDPOINTS, ERROR_MESSAGES } from '../constants'
+import type { ChatCompletionRequest, ChatCompletionChunk } from '../types'
+
+/**
+ * Hook for handling streaming chat completion requests
+ */
+export function useStreamRequest() {
+  const sseSourceRef = useRef<SSE | null>(null)
+  const isStreamCompleteRef = useRef(false)
+
+  const sendStreamRequest = useCallback(
+    (
+      payload: ChatCompletionRequest,
+      onUpdate: (type: 'reasoning' | 'content', chunk: string) => void,
+      onComplete: () => void,
+      onError: (error: string, errorCode?: string) => void
+    ) => {
+      const source = new SSE(API_ENDPOINTS.CHAT_COMPLETIONS, {
+        headers: getCommonHeaders(),
+        method: 'POST',
+        payload: JSON.stringify(payload),
+      })
+
+      sseSourceRef.current = source
+      isStreamCompleteRef.current = false
+
+      const closeSource = () => {
+        source.close()
+        sseSourceRef.current = null
+      }
+
+      const handleError = (errorMessage: string, errorCode?: string) => {
+        if (!isStreamCompleteRef.current) {
+          onError(errorMessage, errorCode)
+          closeSource()
+        }
+      }
+
+      source.addEventListener('message', (e: MessageEvent) => {
+        if (e.data === '[DONE]') {
+          isStreamCompleteRef.current = true
+          closeSource()
+          onComplete()
+          return
+        }
+
+        try {
+          const chunk: ChatCompletionChunk = JSON.parse(e.data)
+          const delta = chunk.choices?.[0]?.delta
+
+          if (delta) {
+            if (delta.reasoning_content) {
+              onUpdate('reasoning', delta.reasoning_content)
+            }
+            if (delta.content) {
+              onUpdate('content', delta.content)
+            }
+          }
+        } catch (error) {
+          // eslint-disable-next-line no-console
+          console.error('Failed to parse SSE message:', error)
+          handleError(ERROR_MESSAGES.PARSE_ERROR)
+        }
+      })
+
+      source.addEventListener('error', (e: Event & { data?: string }) => {
+        // Only handle errors if stream didn't complete normally
+        if (source.readyState !== 2) {
+          // eslint-disable-next-line no-console
+          console.error('SSE Error:', e)
+          let errorMessage = e.data || ERROR_MESSAGES.API_REQUEST_ERROR
+          let errorCode: string | undefined
+          if (e.data) {
+            try {
+              const parsed = JSON.parse(e.data) as {
+                error?: { message?: string; code?: string }
+              }
+              if (parsed?.error) {
+                errorMessage = parsed.error.message || errorMessage
+                errorCode = parsed.error.code || undefined
+              }
+            } catch {
+              // not JSON, use raw string
+            }
+          }
+          handleError(errorMessage, errorCode)
+        }
+      })
+
+      source.addEventListener(
+        'readystatechange',
+        (e: Event & { readyState?: number }) => {
+          const status = (source as unknown as { status?: number }).status
+          if (
+            e.readyState !== undefined &&
+            e.readyState >= 2 &&
+            status !== undefined &&
+            status !== 200
+          ) {
+            handleError(`HTTP ${status}: ${ERROR_MESSAGES.CONNECTION_CLOSED}`)
+          }
+        }
+      )
+
+      try {
+        source.stream()
+      } catch (error: unknown) {
+        // eslint-disable-next-line no-console
+        console.error('Failed to start SSE stream:', error)
+        onError(ERROR_MESSAGES.STREAM_START_ERROR)
+        sseSourceRef.current = null
+      }
+    },
+    []
+  )
+
+  const stopStream = useCallback(() => {
+    if (sseSourceRef.current) {
+      sseSourceRef.current.close()
+      sseSourceRef.current = null
+    }
+  }, [])
+
+  // eslint-disable-next-line react-hooks/refs
+  const isStreaming = sseSourceRef.current !== null
+
+  return {
+    sendStreamRequest,
+    stopStream,
+    // eslint-disable-next-line react-hooks/refs
+    isStreaming,
+  }
+}

web/default/src/features/playground/index.tsx

@@ -0,0 +1,191 @@
+import { useCallback, useEffect, useState } from 'react'
+import { useQuery } from '@tanstack/react-query'
+import { getUserModels, getUserGroups } from './api'
+import { PlaygroundChat } from './components/playground-chat'
+import { PlaygroundInput } from './components/playground-input'
+import { DEFAULT_GROUP } from './constants'
+import { usePlaygroundState, useChatHandler } from './hooks'
+import { createUserMessage, createLoadingAssistantMessage } from './lib'
+import type { Message as MessageType } from './types'
+
+export function Playground() {
+  const {
+    config,
+    parameterEnabled,
+    messages,
+    models,
+    groups,
+    updateMessages,
+    setModels,
+    setGroups,
+    updateConfig,
+  } = usePlaygroundState()
+
+  const { sendChat, stopGeneration, isGenerating } = useChatHandler({
+    config,
+    parameterEnabled,
+    onMessageUpdate: updateMessages,
+  })
+
+  // Edit dialog state
+  const [editingMessageKey, setEditingMessageKey] = useState<string | null>(
+    null
+  )
+
+  // Load models
+  const { data: modelsData, isLoading: isLoadingModels } = useQuery({
+    queryKey: ['playground-models'],
+    queryFn: getUserModels,
+  })
+
+  // Load groups
+  const { data: groupsData } = useQuery({
+    queryKey: ['playground-groups'],
+    queryFn: getUserGroups,
+  })
+
+  // Update models when data changes
+  useEffect(() => {
+    if (!modelsData) return
+
+    setModels(modelsData)
+
+    // Set default model if current model is not available
+    const isCurrentModelValid = modelsData.some((m) => m.value === config.model)
+    if (modelsData.length > 0 && !isCurrentModelValid) {
+      updateConfig('model', modelsData[0].value)
+    }
+  }, [modelsData, config.model, setModels, updateConfig])
+
+  // Update groups when data changes
+  useEffect(() => {
+    if (!groupsData) return
+
+    // Add auto group if not present
+    const hasAutoGroup = groupsData.some((g) => g.value === DEFAULT_GROUP)
+    const processedGroups = hasAutoGroup
+      ? groupsData
+      : [
+          {
+            value: DEFAULT_GROUP,
+            label: 'Auto',
+            ratio: 1,
+            desc: 'Circuit Breaker',
+          },
+          ...groupsData,
+        ]
+
+    setGroups(processedGroups)
+  }, [groupsData, setGroups])
+
+  const handleSendMessage = (text: string) => {
+    const userMessage = createUserMessage(text)
+    const assistantMessage = createLoadingAssistantMessage()
+
+    const newMessages = [...messages, userMessage, assistantMessage]
+    updateMessages(newMessages)
+
+    // Send chat request
+    sendChat(newMessages)
+  }
+
+  const handleCopyMessage = (message: MessageType) => {
+    // Copy is handled in MessageActions component
+    // eslint-disable-next-line no-console
+    console.log('Message copied:', message.key)
+  }
+
+  const handleRegenerateMessage = (message: MessageType) => {
+    // Find the message index and regenerate from there
+    const messageIndex = messages.findIndex((m) => m.key === message.key)
+    if (messageIndex === -1) return
+
+    // Remove messages after this one and regenerate
+    const messagesUpToHere = messages.slice(0, messageIndex)
+    const loadingMessage = createLoadingAssistantMessage()
+    const newMessages = [...messagesUpToHere, loadingMessage]
+
+    updateMessages(newMessages)
+    sendChat(newMessages)
+  }
+
+  const handleEditMessage = useCallback((message: MessageType) => {
+    setEditingMessageKey(message.key)
+  }, [])
+
+  const handleEditOpenChange = useCallback((open: boolean) => {
+    if (!open) setEditingMessageKey(null)
+  }, [])
+
+  // Apply edit and optionally re-submit from the edited user message
+  const applyEdit = useCallback(
+    (newContent: string, submit: boolean) => {
+      if (!editingMessageKey) return
+      const index = messages.findIndex((m) => m.key === editingMessageKey)
+      if (index === -1) return
+
+      const updated = messages.map((m) =>
+        m.key === editingMessageKey
+          ? { ...m, versions: [{ ...m.versions[0], content: newContent }] }
+          : m
+      )
+
+      setEditingMessageKey(null)
+
+      if (!submit || updated[index].from !== 'user') {
+        updateMessages(updated)
+        return
+      }
+
+      const toSubmit = [
+        ...updated.slice(0, index + 1),
+        createLoadingAssistantMessage(),
+      ]
+      updateMessages(toSubmit)
+      sendChat(toSubmit)
+    },
+    [editingMessageKey, messages, updateMessages, sendChat]
+  )
+
+  const handleDeleteMessage = (message: MessageType) => {
+    const newMessages = messages.filter((m) => m.key !== message.key)
+    updateMessages(newMessages)
+  }
+
+  return (
+    <div className='relative flex size-full flex-col overflow-hidden'>
+      {/* Full-width scroll container: scrolling works even over side whitespace */}
+      <div className='flex flex-1 flex-col overflow-hidden'>
+        <PlaygroundChat
+          messages={messages}
+          onCopyMessage={handleCopyMessage}
+          onRegenerateMessage={handleRegenerateMessage}
+          onEditMessage={handleEditMessage}
+          onDeleteMessage={handleDeleteMessage}
+          isGenerating={isGenerating}
+          editingKey={editingMessageKey}
+          onCancelEdit={handleEditOpenChange}
+          onSaveEdit={(newContent) => applyEdit(newContent, false)}
+          onSaveEditAndSubmit={(newContent) => applyEdit(newContent, true)}
+        />
+      </div>
+
+      {/* Input area: center content and constrain to the same container width */}
+      <div className='mx-auto w-full max-w-4xl'>
+        <PlaygroundInput
+          disabled={isGenerating}
+          groups={groups}
+          groupValue={config.group}
+          isGenerating={isGenerating}
+          isModelLoading={isLoadingModels}
+          modelValue={config.model}
+          models={models}
+          onGroupChange={(value) => updateConfig('group', value)}
+          onModelChange={(value) => updateConfig('model', value)}
+          onStop={stopGeneration}
+          onSubmit={handleSendMessage}
+        />
+      </div>
+    </div>
+  )
+}

web/default/src/features/playground/lib/index.ts

@@ -0,0 +1,4 @@
+export * from './message-utils'
+export * from './payload-builder'
+export * from './storage'
+export * from './message-styles'

web/default/src/features/playground/lib/message-styles.ts

@@ -0,0 +1,32 @@
+/**
+ * Get message content styles based on role
+ * Encapsulates styling logic for user and assistant messages
+ */
+export function getMessageContentStyles() {
+  return [
+    // Assistant content fills the row; user bubble auto-width
+    'group-[.is-assistant]:w-full',
+    'group-[.is-assistant]:max-w-none',
+    'group-[.is-user]:w-fit',
+    // User bubble: rounded and themed background
+    'group-[.is-user]:text-foreground',
+    'group-[.is-user]:bg-secondary',
+    'dark:group-[.is-user]:bg-muted',
+    'group-[.is-user]:rounded-[24px]',
+    // Assistant bubble: flat serif style (one-sided style)
+    'group-[.is-assistant]:text-foreground',
+    'group-[.is-assistant]:bg-transparent',
+    'group-[.is-assistant]:p-0',
+    'group-[.is-assistant]:font-serif',
+    // Preferred readable widths and wrapping
+    'leading-relaxed',
+    'break-words',
+    'whitespace-pre-wrap',
+    'sm:leading-7',
+    // Cap user bubble width so it does not look like a banner
+    'group-[.is-user]:max-w-[85%]',
+    'sm:group-[.is-user]:max-w-[62ch]',
+    'md:group-[.is-user]:max-w-[68ch]',
+    'lg:group-[.is-user]:max-w-[72ch]',
+  ].join(' ')
+}

web/default/src/features/playground/lib/message-utils.ts

@@ -0,0 +1,337 @@
+import { nanoid } from 'nanoid'
+import { MESSAGE_ROLES, MESSAGE_STATUS, ERROR_MESSAGES } from '../constants'
+import type {
+  Message,
+  MessageVersion,
+  ChatCompletionMessage,
+  ContentPart,
+} from '../types'
+
+/**
+ * Create a new message version
+ */
+export function createMessageVersion(content: string): MessageVersion {
+  return {
+    id: nanoid(),
+    content,
+  }
+}
+
+/**
+ * Get current version from message (always returns the first version)
+ */
+export function getCurrentVersion(message: Message): MessageVersion {
+  return message.versions[0] || { id: 'default', content: '' }
+}
+
+/**
+ * Update current version content in message
+ */
+export function updateCurrentVersionContent(
+  message: Message,
+  content: string
+): Message {
+  const currentVersion = getCurrentVersion(message)
+  return {
+    ...message,
+    versions: [{ ...currentVersion, content }],
+  }
+}
+
+/**
+ * Create a user message
+ */
+export function createUserMessage(content: string): Message {
+  return {
+    key: nanoid(),
+    from: MESSAGE_ROLES.USER,
+    versions: [createMessageVersion(content)],
+  }
+}
+
+/**
+ * Create a loading assistant message
+ */
+export function createLoadingAssistantMessage(): Message {
+  return {
+    key: nanoid(),
+    from: MESSAGE_ROLES.ASSISTANT,
+    versions: [createMessageVersion('')],
+    reasoning: undefined,
+    isReasoningComplete: false,
+    isContentComplete: false,
+    isReasoningStreaming: false,
+    status: MESSAGE_STATUS.LOADING,
+  }
+}
+
+/**
+ * Build message content with optional images
+ */
+export function buildMessageContent(
+  text: string,
+  imageUrls: string[] = []
+): string | ContentPart[] {
+  const validImages = imageUrls.filter((url) => url.trim() !== '')
+
+  if (validImages.length === 0) {
+    return text
+  }
+
+  const parts: ContentPart[] = [
+    {
+      type: 'text',
+      text: text || '',
+    },
+    ...validImages.map((url) => ({
+      type: 'image_url' as const,
+      image_url: { url: url.trim() },
+    })),
+  ]
+
+  return parts
+}
+
+/**
+ * Extract text content from message content
+ */
+export function getTextContent(content: string | ContentPart[]): string {
+  if (typeof content === 'string') {
+    return content
+  }
+
+  if (Array.isArray(content)) {
+    const textPart = content.find((part) => part.type === 'text')
+    return textPart?.text || ''
+  }
+
+  return ''
+}
+
+/**
+ * Format message for API request
+ */
+export function formatMessageForAPI(message: Message): ChatCompletionMessage {
+  const currentVersion = getCurrentVersion(message)
+  return {
+    role: message.from,
+    content: currentVersion.content,
+  }
+}
+
+/**
+ * Check if message is valid for API request
+ * Excludes loading/streaming assistant messages and empty content
+ */
+export function isValidMessage(message: Message): boolean {
+  if (!message || !message.from || !message.versions.length) return false
+
+  const content = message.versions[0]?.content
+  if (content === undefined) return false
+
+  // Exclude empty assistant messages (loading/streaming placeholders)
+  if (message.from === 'assistant' && !content.trim()) return false
+
+  return true
+}
+
+/**
+ * Parse content to separate thinking from visible text
+ * Handles both complete and incomplete <think> tags
+ */
+export function parseThinkTags(content: string): {
+  visibleContent: string
+  reasoning: string
+  hasUnclosedTag: boolean
+} {
+  if (!content.includes('<think>')) {
+    return { visibleContent: content, reasoning: '', hasUnclosedTag: false }
+  }
+
+  const visibleParts: string[] = []
+  const reasoningParts: string[] = []
+  let currentPos = 0
+  let hasUnclosed = false
+
+  while (true) {
+    // Find next <think> tag
+    const openPos = content.indexOf('<think>', currentPos)
+
+    if (openPos === -1) {
+      // No more think tags, add remaining content
+      if (currentPos < content.length) {
+        visibleParts.push(content.substring(currentPos))
+      }
+      break
+    }
+
+    // Add visible content before this tag
+    if (openPos > currentPos) {
+      visibleParts.push(content.substring(currentPos, openPos))
+    }
+
+    // Look for matching </think> tag
+    const closePos = content.indexOf('</think>', openPos + 7)
+
+    if (closePos === -1) {
+      // Unclosed tag: rest is reasoning buffer
+      reasoningParts.push(content.substring(openPos + 7))
+      hasUnclosed = true
+      break
+    }
+
+    // Extract reasoning content between tags
+    reasoningParts.push(content.substring(openPos + 7, closePos))
+    currentPos = closePos + 8
+  }
+
+  return {
+    visibleContent: visibleParts.join('').trim(),
+    reasoning: reasoningParts.join('\n\n').trim(),
+    hasUnclosedTag: hasUnclosed,
+  }
+}
+
+/**
+ * Update the last assistant message with an error
+ * @param messages - Current messages array
+ * @param errorMessage - Error message to display
+ * @returns Updated messages array
+ */
+export function updateAssistantMessageWithError(
+  messages: Message[],
+  errorMessage: string,
+  errorCode?: string
+): Message[] {
+  return updateLastAssistantMessage(messages, (message) => {
+    const updatedMessage = updateCurrentVersionContent(
+      message,
+      `${ERROR_MESSAGES.API_REQUEST_ERROR}: ${errorMessage}`
+    )
+    return {
+      ...updatedMessage,
+      status: MESSAGE_STATUS.ERROR,
+      isReasoningStreaming: false,
+      errorCode: errorCode || null,
+    }
+  })
+}
+
+/**
+ * Helper function to update the last assistant message
+ * @param messages - Current messages array
+ * @param updater - Function to update the message
+ * @returns Updated messages array or original if no assistant message found
+ */
+export function updateLastAssistantMessage(
+  messages: Message[],
+  updater: (message: Message) => Message
+): Message[] {
+  if (messages.length === 0) return messages
+  const last = messages[messages.length - 1]
+  if (!last || last.from !== MESSAGE_ROLES.ASSISTANT) return messages
+
+  const updated = [...messages]
+  updated[updated.length - 1] = updater(last)
+  return updated
+}
+
+/**
+ * Process content chunk during streaming
+ * Separates <think> reasoning from visible content in real-time
+ * Note: versions[0].content keeps the full raw content (with tags) during streaming
+ */
+export function processStreamingContent(
+  message: Message,
+  contentChunk?: string
+): Message {
+  const currentVersion = getCurrentVersion(message)
+  const fullContent = contentChunk
+    ? currentVersion.content + contentChunk
+    : currentVersion.content
+
+  const { reasoning, hasUnclosedTag } = parseThinkTags(fullContent)
+
+  // Preserve existing reasoning if no think tags found (e.g., from API reasoning_content)
+  const finalReasoning = reasoning
+    ? { content: reasoning, duration: 0 }
+    : message.reasoning
+
+  return {
+    ...updateCurrentVersionContent(message, fullContent),
+    reasoning: finalReasoning,
+    isReasoningStreaming: hasUnclosedTag,
+  }
+}
+
+/**
+ * Finalize message after streaming completes
+ * Cleans content and consolidates reasoning from all sources
+ */
+export function finalizeMessage(
+  message: Message,
+  apiReasoningContent?: string
+): Message {
+  const currentVersion = getCurrentVersion(message)
+  const { visibleContent, reasoning } = parseThinkTags(currentVersion.content)
+
+  // Priority:
+  // 1. API reasoning_content passed as parameter (non-streaming response)
+  // 2. Existing message.reasoning (from streaming reasoning_content)
+  // 3. Extracted think tags from content
+  const finalReasoning =
+    apiReasoningContent || message.reasoning?.content || reasoning || ''
+
+  return {
+    ...updateCurrentVersionContent(message, visibleContent),
+    reasoning: finalReasoning
+      ? { content: finalReasoning, duration: message.reasoning?.duration || 0 }
+      : undefined,
+    isReasoningStreaming: false,
+  }
+}
+
+/**
+ * Sanitize messages loaded from storage
+ * Converts stuck loading/streaming messages to stable state
+ */
+export function sanitizeMessagesOnLoad(messages: Message[]): Message[] {
+  let targetIndex = -1
+  for (let i = messages.length - 1; i >= 0; i--) {
+    const m = messages[i]
+    if (
+      m?.from === MESSAGE_ROLES.ASSISTANT &&
+      (m?.status === MESSAGE_STATUS.LOADING ||
+        m?.status === MESSAGE_STATUS.STREAMING)
+    ) {
+      targetIndex = i
+      break
+    }
+  }
+
+  if (targetIndex === -1) return messages
+
+  const finalized = finalizeMessage(messages[targetIndex])
+  const hasContent = finalized.versions?.[0]?.content?.trim()
+  const hasReasoning = finalized.reasoning?.content?.trim()
+
+  const sanitized: Message =
+    hasContent || hasReasoning
+      ? {
+          ...finalized,
+          status: MESSAGE_STATUS.COMPLETE,
+          isReasoningStreaming: false,
+        }
+      : {
+          ...updateCurrentVersionContent(
+            finalized,
+            `${ERROR_MESSAGES.API_REQUEST_ERROR}: ${ERROR_MESSAGES.INTERRUPTED}`
+          ),
+          status: MESSAGE_STATUS.ERROR,
+          isReasoningStreaming: false,
+        }
+
+  const result = [...messages]
+  result[targetIndex] = sanitized
+  return result
+}

web/default/src/features/playground/lib/payload-builder.ts

@@ -0,0 +1,49 @@
+import type {
+  ChatCompletionRequest,
+  Message,
+  PlaygroundConfig,
+  ParameterEnabled,
+} from '../types'
+import { formatMessageForAPI, isValidMessage } from './message-utils'
+
+/**
+ * Build API request payload from messages and config
+ */
+export function buildChatCompletionPayload(
+  messages: Message[],
+  config: PlaygroundConfig,
+  parameterEnabled: ParameterEnabled
+): ChatCompletionRequest {
+  // Filter and format valid messages
+  const processedMessages = messages
+    .filter(isValidMessage)
+    .map(formatMessageForAPI)
+
+  const payload: ChatCompletionRequest = {
+    model: config.model,
+    group: config.group,
+    messages: processedMessages,
+    stream: config.stream,
+  }
+
+  // Add enabled parameters
+  const parameterKeys: Array<keyof ParameterEnabled> = [
+    'temperature',
+    'top_p',
+    'max_tokens',
+    'frequency_penalty',
+    'presence_penalty',
+    'seed',
+  ]
+
+  parameterKeys.forEach((key) => {
+    if (parameterEnabled[key]) {
+      const value = config[key as keyof PlaygroundConfig]
+      if (value !== undefined && value !== null) {
+        ;(payload as unknown as Record<string, unknown>)[key] = value
+      }
+    }
+  })
+
+  return payload
+}

web/default/src/features/playground/lib/storage.ts

@@ -0,0 +1,112 @@
+import { STORAGE_KEYS } from '../constants'
+import type { PlaygroundConfig, ParameterEnabled, Message } from '../types'
+import { sanitizeMessagesOnLoad } from './message-utils'
+
+/**
+ * Load playground config from localStorage
+ */
+export function loadConfig(): Partial<PlaygroundConfig> {
+  try {
+    const saved = localStorage.getItem(STORAGE_KEYS.CONFIG)
+    if (saved) {
+      return JSON.parse(saved)
+    }
+  } catch (error) {
+    // eslint-disable-next-line no-console
+    console.error('Failed to load config:', error)
+  }
+  return {}
+}
+
+/**
+ * Save playground config to localStorage
+ */
+export function saveConfig(config: Partial<PlaygroundConfig>): void {
+  try {
+    localStorage.setItem(STORAGE_KEYS.CONFIG, JSON.stringify(config))
+  } catch (error) {
+    // eslint-disable-next-line no-console
+    console.error('Failed to save config:', error)
+  }
+}
+
+/**
+ * Load parameter enabled state from localStorage
+ */
+export function loadParameterEnabled(): Partial<ParameterEnabled> {
+  try {
+    const saved = localStorage.getItem(STORAGE_KEYS.PARAMETER_ENABLED)
+    if (saved) {
+      return JSON.parse(saved)
+    }
+  } catch (error) {
+    // eslint-disable-next-line no-console
+    console.error('Failed to load parameter enabled:', error)
+  }
+  return {}
+}
+
+/**
+ * Save parameter enabled state to localStorage
+ */
+export function saveParameterEnabled(
+  parameterEnabled: Partial<ParameterEnabled>
+): void {
+  try {
+    localStorage.setItem(
+      STORAGE_KEYS.PARAMETER_ENABLED,
+      JSON.stringify(parameterEnabled)
+    )
+  } catch (error) {
+    // eslint-disable-next-line no-console
+    console.error('Failed to save parameter enabled:', error)
+  }
+}
+
+/**
+ * Load messages from localStorage
+ */
+export function loadMessages(): Message[] | null {
+  try {
+    const saved = localStorage.getItem(STORAGE_KEYS.MESSAGES)
+    if (saved) {
+      const parsed: Message[] = JSON.parse(saved)
+      const sanitized = sanitizeMessagesOnLoad(parsed)
+      // Persist sanitized result to avoid re-sanitizing on subsequent loads
+      if (sanitized !== parsed) {
+        saveMessages(sanitized)
+      }
+      return sanitized
+    }
+  } catch (error) {
+    // eslint-disable-next-line no-console
+    console.error('Failed to load messages:', error)
+  }
+  return null
+}
+
+/**
+ * Save messages to localStorage
+ */
+export function saveMessages(messages: Message[]): void {
+  try {
+    localStorage.setItem(STORAGE_KEYS.MESSAGES, JSON.stringify(messages))
+  } catch (error) {
+    // eslint-disable-next-line no-console
+    console.error('Failed to save messages:', error)
+  }
+}
+
+/**
+ * Clear all playground data
+ */
+export function clearPlaygroundData(): void {
+  try {
+    localStorage.removeItem(STORAGE_KEYS.CONFIG)
+    localStorage.removeItem(STORAGE_KEYS.PARAMETER_ENABLED)
+    localStorage.removeItem(STORAGE_KEYS.MESSAGES)
+  } catch (error) {
+    // eslint-disable-next-line no-console
+    console.error('Failed to clear playground data:', error)
+  }
+}

web/default/src/features/playground/types.ts

@@ -0,0 +1,124 @@
+// Message types
+export type MessageRole = 'user' | 'assistant' | 'system'
+
+export type MessageStatus = 'loading' | 'streaming' | 'complete' | 'error'
+
+export interface MessageVersion {
+  id: string
+  content: string
+}
+
+export interface Message {
+  key: string
+  from: MessageRole
+  versions: MessageVersion[]
+  sources?: { href: string; title: string }[]
+  reasoning?: {
+    content: string
+    duration: number
+  }
+  isReasoningStreaming?: boolean
+  isReasoningComplete?: boolean
+  isContentComplete?: boolean
+  status?: MessageStatus
+  errorCode?: string | null
+}
+
+// API payload types
+export interface ChatCompletionMessage {
+  role: MessageRole
+  content: string | ContentPart[]
+}
+
+export interface ContentPart {
+  type: 'text' | 'image_url'
+  text?: string
+  image_url?: {
+    url: string
+  }
+}
+
+export interface ChatCompletionRequest {
+  model: string
+  group?: string
+  messages: ChatCompletionMessage[]
+  stream: boolean
+  temperature?: number
+  top_p?: number
+  max_tokens?: number
+  frequency_penalty?: number
+  presence_penalty?: number
+  seed?: number
+}
+
+export interface ChatCompletionChunk {
+  id: string
+  object: string
+  created: number
+  model: string
+  choices: Array<{
+    index: number
+    delta: {
+      role?: MessageRole
+      content?: string
+      reasoning_content?: string
+    }
+    finish_reason: string | null
+  }>
+}
+
+export interface ChatCompletionResponse {
+  id: string
+  object: string
+  created: number
+  model: string
+  choices: Array<{
+    index: number
+    message: {
+      role: MessageRole
+      content: string
+      reasoning_content?: string
+    }
+    finish_reason: string
+  }>
+  usage?: {
+    prompt_tokens: number
+    completion_tokens: number
+    total_tokens: number
+  }
+}
+
+// Configuration types
+export interface PlaygroundConfig {
+  model: string
+  group: string
+  temperature: number
+  top_p: number
+  max_tokens: number
+  frequency_penalty: number
+  presence_penalty: number
+  seed: number | null
+  stream: boolean
+}
+
+export interface ParameterEnabled {
+  temperature: boolean
+  top_p: boolean
+  max_tokens: boolean
+  frequency_penalty: boolean
+  presence_penalty: boolean
+  seed: boolean
+}
+
+// Model and group options
+export interface ModelOption {
+  label: string
+  value: string
+}
+
+export interface GroupOption {
+  label: string
+  value: string
+  ratio: number
+  desc?: string
+}