github.com/DEEIX-AI/DEEIX-Chat

✨ Feature

• Message deletion: Delete individual user or assistant messages while preserving subsequent messages and reconnecting the conversation; the first message and messages being generated are protected.
• OpenRouter image support: Added text-to-image generation and reference-image editing, with image model discovery and streaming responses where supported.
• Tiered pricing import: Import OpenRouter pricing tiers, including input, output, cache-read, and cache-write rates.
• Signed MCP user context: MCP calls can optionally carry signed user, conversation, and request information, enabling services to verify caller context.
• MCP header presets: Added quick-fill request headers, JSON validation, and protection against overwriting conflicting values.
• Embedding compatibility: Added an option to omit the dimensions parameter for embedding providers that do not accept it.

⚡ Optimization

• Dark theme: Adjusted background, border, and foreground contrast for clearer visual hierarchy.
• Per-device appearance: Light, dark, and system modes are now saved independently on each device instead of being overwritten by account synchronization.
• Chat navigation: Improved message-position and response-outline navigation, including heading alignment, active-section tracking, and hover targets.
• Markdown tables: Refined column widths, long-text wrapping, line spacing, and explicit line-break spacing.
• Admin scrolling: Pages can now scroll when the pointer is over the sidebar or page margins.
• Pricing configuration: Added clearer price units, tier ranges, and import warnings; tier limits now use full token counts instead of thousands.
• Upstream model status: Clearly marks models that have disappeared from upstream catalogs and explains route availability and automatic recovery.
• File indexing: Added a distinct “No text” state and count, plus an option to include these files when manually rebuilding indexes.

🐛 Fixed

• Fixed tool-call steps disappearing during long reasoning streams and improved preservation of reasoning content across rounds.
• Fixed rendering problems caused by nested code fences.
• Fixed excessive escaping when pasting rich content as Markdown.
• Fixed model submenus shifting unexpectedly on hover or close.
• Fixed table content bleeding above sticky headers.
• Fixed light-theme flashes when loading a page with dark mode selected.
• Fixed locked accounts being reported as invalid credentials; login errors now show the remaining lock duration.
• Fixed compatibility with upstreams that only accept a single leading system message, while preserving explicit prompt-cache behavior.
• Fixed repeated extraction and reindexing of files with no readable text.
• Fixed imported cache-write pricing calculations and aligned displayed prices with billing.
• Fixed raw JSON error output during upstream model synchronization.
• Improved error handling when deleting messages or replying to a message that has already been deleted.

✨ Features

• Introduced a tabbed command palette for chat @ and / actions.
• Added on-demand expansion for complete tool call results.
• Added usage balance, lifetime spending, and yearly activity insights to subscription settings.
• Added per-user yearly activity heatmaps.
• Added project default models and inherited MCP tool settings.
• Added manual file vectorization.
• Added an admin switch to disable knowledge base capabilities.
• Added safer upstream model catalog synchronization.
• Improved temporary conversation support for tools, files, and retrieval features.

⚡ Improvements

• Improved chat composer focus behavior and keyboard navigation.
• Preserved saved drafts when creating a new conversation.
• Clearly separated attachment upload and file-processing states.
• Improved file preview, file list, and knowledge base detail pages.
• Unified preview layouts for sidebar search and starred conversations.
• Refined MCP and knowledge base popovers and dialogs.
• Improved empty states across announcements, files, models, logs, and settings.
• Improved code block and HTML block rendering across streamed content.
• Improved table scrolling, column sizing, and Firefox compatibility.
• Added better scroll-fade fallbacks for browsers without scroll-driven animation support.

🐛 Fixes

• Fixed cached prompt billing to use observed usage.
• Fixed stale conversation run indicators after generation.
• Fixed blank scrolling before model output.
• Fixed excess blank space after generation.
• Fixed long reasoning content being lost during stream recovery.
• Fixed reasoning snapshot API inconsistencies.
• Fixed multiline web text pasting into the composer.
• Fixed untagged code fences being parsed as Markdown.
• Fixed active session location handling.
• Fixed admin redemption filters and model selector styling.
• Improved SMTP delivery, authentication, and generation lifecycle stability.

✨ Feature

• Added privacy-preserving Temporary Chats that are excluded from normal history, cleared after leaving the page, and support model options, text-based MCP tools, Skills, knowledge bases, moderation, and Agent traces.
• Added xAI video extension with single-MP4 input, prompts, duration controls, compatible model routing, and attachment validation.
• Redesigned Agent traces to interleave reasoning and tool calls in their real execution order.
• Added structured tool cards with arguments, results, errors, source information, exit codes, copy actions, and duration.
• Added Agent run summaries showing tool-call count, reasoning rounds, and total execution time.
• Added independent preferences for automatically expanding active reasoning and active tool calls.
• Added model-aware context windows with automatic detection, presets, custom values from 4,096 to 16,000,000 tokens, and output-limit validation.
• Added bulk context-window updates with conflict detection, unchanged-value checks, and partial-failure feedback.
• Added live conversation run indicators in the sidebar for responses continuing in the background.
• Added MCP tool per-call pricing with search, filtering, selection, and bulk-pricing controls.
• Added paid-tool indicators in chat, including per-call prices and successful-call-only billing rules.
• Added MCP usage snapshots and detailed billing items. Self-use mode records usage without charging, while priced MCP calls are billed separately from model usage.
• Added administrator redemption records with filtering by user, redemption reference, code, and reward type.
• Redemption details now include code status, reward content, subscription duration, and balances before and after redemption.
• Enhanced Markdown tables and Mermaid diagrams with CSV/Markdown downloads, copy, fullscreen, zoom, reset, and multiple export formats.
• Added smoother hover expansion for the collapsed application sidebar.

⚡ Optimization

• Context compression now uses each model’s effective input budget, with configurable fallback windows and percentage-based trigger thresholds.
• Context compression now exposes pending, completed, and failed states and can retry unsuccessful background summaries.
• Token-budget protection now works independently from context compression, allowing oversized history to be trimmed even when rolling summaries are disabled.
• Optimized long-conversation screenshots and limited captures to the most recent or selected 100 messages.
• Improved sidebar navigation, conversation switching, search previews, recent conversations, starred conversations, and project navigation responsiveness.
• Preserved the application sidebar while navigating into the administration console, avoiding unnecessary layout and state resets.
• Added pagination and “Load more” workflows while reducing repeated capability checks in project catalogs and knowledge-base selectors.
• Added batched file-processing status queries to reduce one-request-per-file polling.
• Reduced redundant knowledge-base and file-processing refreshes by sharing processing-state updates.
• Bounded file-processing queues and background concurrency for better stability under heavy upload workloads.
• Improved chat submission coordination across queued messages, media tasks, branching, cancellation, and streaming completion.
• Added periodic authoritative run snapshots so clients can recover when stream start or finish events are missed.
• Isolated upstream 429 backoff by model route so one limited route does not block other healthy routes.
• Improved media playback with more reliable seeking, fullscreen behavior, inline controls, and progress updates.
• Improved streamed Markdown rendering, table layout, footnote navigation, and code-copy actions.
• Added graceful shutdown draining with readiness switching, active-request completion time, and clean SSE disconnection.
• Improved background-task supervision and write-pressure controls for more reliable high-load operation.

🐛 Fixed

• Fixed streamed Agent traces losing, duplicating, reordering, or flickering reasoning and tool-call records.
• Fixed historical conversations displaying duplicate reasoning blocks after trace-format migrations.
• Fixed reasoning timers and execution durations resetting or disappearing during streaming and after page refreshes.
• Fixed inconsistent duration formatting across reasoning steps, tool calls, message metadata, and Agent summaries.
• Fixed chat-setting changes not immediately synchronizing with active trace rendering.
• Fixed processing and execution traces being lost after editing assistant messages.
• Fixed forked conversations losing traces copied from the source conversation.
• Fixed conversation forks being created from user messages or while source history was incomplete.
• Fixed message-position navigation selecting or reporting the wrong message.
• Fixed response-outline navigation targeting stale or incorrect response sections.
• Fixed automatic scrolling becoming unstable while streamed response blocks changed height.
• Fixed queued messages continuing after their required parent response failed.
• Fixed queue completion not consistently refreshing message-position and response-outline controls.
• Fixed sidebar run indicators remaining stuck after server-side generation had finished.
• Fixed sidebar, recent, starred, project, and response navigation becoming inconsistent after rapid route changes.
• Fixed media playback state being reset during surrounding interface updates.
• Fixed project-dialog footer actions becoming inaccessible when prompts or option lists were too long.
• Fixed manually selected code losing raw characters or formatting when pasted into chat.
• Fixed text copied from web pages being reconstructed as unwanted HTML instead of preserving literal visible content.
• Fixed unsafe or malformed tags being reconstructed while normalizing tool output.
• Fixed weak-network upload retries creating duplicate file records.
• Fixed interrupted uploads leaving processing records in an incorrect state.
• Fixed file-processing polling continuing after a file completed, failed, or was removed.
• Fixed stale model and processing state affecting attachments, file mentions, file pages, and knowledge bases.
• Fixed cancelled or timed-out file-processing requests remaining active.
• Fixed inconsistent file states across chat attachments, the file manager, and knowledge-base pages.
• Fixed knowledge-base processing updates requiring a full-page refresh before appearing.

✨ Feature

• Added personal and built-in knowledge bases with managed files and RAG retrieval.
• Added knowledge-base selection in chat, project-level defaults, retrieval status, and source citations in responses.
• Added message forking to create a new conversation from any completed user or assistant message while preserving its preceding context and attachments.
• Added synchronized Markdown response outlines for quickly navigating long answers.
• Added Mermaid diagram downloads in SVG, PNG, and editable MMD formats.
• Added secure SVG artifact preview and download support.
• Added Mistral OCR with configurable endpoint, API key, model, and timeout.
• Added native tool and thinking-summary support for Gemini Interactions.
• Added managed custom model icons with PNG, JPEG, and WebP uploads, deduplication, reusable icon libraries, and safe cleanup.
• Added safe deletion of unused custom model vendors.
• Added a global switch for automatic model circuit breaking while retaining per-model and per-upstream policies.

⚡ Optimization

• Expanded embedding support to dimensions up to 4096 across PostgreSQL and SQLite vector storage.
• Improved model protocol and upstream-binding management so route configurations remain isolated and intact during synchronization or editing.
• Improved Gemini Interactions compatibility with the official request, streaming, native-tool, reasoning, and event schemas.
• Preserved partial assistant output when reconnecting to an in-progress generation.
• Synchronized user-setting cache invalidation across multiple backend instances so changes take effect immediately.
• Improved knowledge-base file health, indexing status, availability diagnostics, and retrieval-source visibility.
• Improved content moderation configuration validation, event filtering, interrupted-generation recovery, and shared-conversation displays.
• Improved model icon validation and lifecycle management, including reference checks and delayed cleanup of unused uploads.
• Improved administration dialog consistency and permission-group table presentation.

🐛 Fixed

• Fixed message forks carrying invalid state, incomplete history, or inconsistent parent relationships.
• Fixed forked conversations appearing incorrectly in the sidebar or being available before message generation completed.
• Fixed upstream model binding edits losing protocol or route configuration.
• Fixed Gemini Interactions native tools, thinking summaries, streamed events, and large step indexes being parsed incorrectly.
• Fixed Mistral OCR losing Markdown structure or incorrectly buffering streamed request bodies.
• Fixed possible allocation overflow when processing large Mistral OCR requests.
• Fixed stale user preferences remaining active after settings were changed.
• Fixed EPay checkout URL construction, gateway validation, signing, error reporting, and redirect diagnostics.
• Fixed unexpected chat scrolling when interacting with or streaming around code blocks.
• Fixed interrupted chat generations failing to resume correctly after reconnecting.
• Fixed content-moderation recovery states and numeric configuration validation.

⚠️ Note

This update includes a migration to fix the vector store. Please back up your database and schedule a maintenance window before upgrading. For PostgreSQL, ensure pgvector is version 0.8.0 or higher. After the upgrade, please check the admin dashboard and rebuild any invalid vectors.

✨ Feature

• Added model vendor management with custom stable keys, display names, and icons.
• Added optional model display groups that organize the model picker independently from technical vendors.
• Added bulk model-group assignment and drag-and-drop ordering for groups and models.
• Added an OAuth authorization bridge with PKCE for web, desktop, and other public clients while preserving the existing 2FA and session flows.
• Added a recommended server-side OAuth callback URL alongside the legacy web callback.
• Added native xAI image and video generation with duration and resolution controls.
• Added per-second pricing for video models, with duration billing applied only after successful generation.
• Added MCP image attachment processors that convert uploaded images into text context before invoking the main model.
• Added configurable MCP image and prompt argument mappings with Data URL and raw Base64 transfer modes.
• Added project and conversation presets for MCP image processors, limited to one active processor per request.
• Added configurable SSRF hostname and CIDR allowlists for trusted private integrations.

⚡ Optimization

• Added support for trusted private identity-provider and model-upstream endpoints without disabling global SSRF protection.
• Revalidated every outbound redirect to prevent trusted endpoints from redirecting requests to unsafe metadata or link-local destinations.
• Secured generated-media retrieval with controlled redirects, response validation, download limits, and safer artifact persistence.
• Added persistent OpenRouter pricing caching with a 24-hour refresh window and stale-cache fallback when the upstream catalog is unavailable.
• Improved model picker organization with custom groups while keeping unassigned models grouped by their technical vendor.
• Added schema-aware MCP processor validation that detects missing image fields and unmapped required arguments.
• Improved model call logs with separate input, output, reasoning, cache, image-count, and video-duration usage details.
• Improved video-generation configuration by exposing only duration options supported by the active model.
• Updated Next.js and related frontend dependencies to address known security advisories.
• Improved redemption-code validation responses with clearer field and reason information.

🐛 Fixed

• Fixed image attachments bypassing their configured MCP processors.
• Fixed MCP image processors receiving incorrectly encoded images or incomplete prompt arguments.
• Fixed historical RAG evidence being recalled from unrelated conversation branches.
• Fixed retried messages inheriting token usage from their original attempts.
• Fixed Android browser speech recognition repeatedly restarting, becoming stuck, or failing without a useful error.
• Added clearer voice-input errors for microphone permissions, unavailable devices, unsupported languages, network failures, and missing speech.
• Fixed xAI video polling instability and invalid duration submissions.
• Fixed integer overflow when allocating large video debug payloads.
• Fixed generated video usage and pricing information being displayed inconsistently in model-call logs.

✨ Feature

• Added full-text conversation search across titles, labels, projects, metadata, and message content, with matched-message previews and paginated results.
• Added manual conversation label management, including the ability to add, remove, and clear labels.
• Added a user preference to disable automatic conversation label generation, reducing auxiliary model calls and token usage.
• Added bulk project moves for selected conversations from the Recent page.
• Added Markdown preview to the message composer.
• Added rich-text paste conversion that preserves Markdown formatting, including links, lists, tables, and code.
• Added click-to-copy support for inline code.
• Added configurable OpenAI prompt caching with Auto, Enabled, and Disabled policies; implicit and explicit cache modes; TTL; and retention controls.
• Added dynamic upstream header values for conversation IDs, session keys, request IDs, and OpenAI diagnostic request IDs, with quick-fill presets for common providers.
• Added post-settlement balances to user and administrator usage logs.
• Added detailed conversation-event inspection for administrators, including payload size information.
• Added administrative cleanup of trace events for selected conversation runs while preserving conversations, messages, attachments, calls, and billing records.

⚡ Optimization

• Added automatic request-route failover for eligible timeouts, rate limits, server errors, and network failures, with support for trying up to three routes before returning an error.
• Preserved historical images across multi-turn conversations so follow-up requests retain the intended visual context.
• Expanded reasoning-context compatibility for DeepSeek, Moonshot, Zhipu, Xiaomi MiMo, Alibaba Qwen, MiniMax, and OpenRouter-compatible routes.
• Added the required reasoning-history request options for providers such as Qwen when previous reasoning content is present.
• Improved MCP tool synchronization with an explicit choice between preserving locally edited names and descriptions or replacing them with server-provided metadata.
• Standardized administrator-facing storage, attachment, image, document, and full-context size settings in MB, including alignment of the default full-context limit to 2 MB.
• Prevented large file bodies from inflating conversation traces; oversized historical payloads are now safely omitted from the log viewer.
• Deferred new-conversation creation until the first message is sent, preventing empty conversations from appearing in history.
• Improved the visual selected state in announcement dialogs.
• Improved code-block styling and readability in rendered Markdown.
• Improved chat scrolling after sending messages and stabilized the viewport when streaming finishes.
• Improved desktop model-menu sizing based on available viewport space.
• Improved generation-resume cleanup so canceled or replaced resume streams cannot apply stale updates.
• Added clear validation for conversation search queries that exceed the supported 200-character limit.

🐛 Fixed

• Fixed failed conversation turns disappearing when the user has insufficient balance.
• Fixed interrupted generations losing reasoning content or producing incomplete usage settlement.
• Fixed stopped or failed first generations overwriting the conversation’s fallback title.
• Fixed concurrent generations and queued messages leaking between different conversation branches.
• Fixed negative account balances being clamped or displayed incorrectly in user and administrator views.
• Fixed Gemini text-only responses being handled incorrectly by media-generation flows.
• Fixed Gemini MCP tool declarations losing referenced JSON Schema definitions.
• Fixed reasoning content being omitted when cloning shared conversations.
• Fixed OpenAI prompt-cache breakpoints not rolling correctly through conversation history.
• Fixed explicit OpenAI cache history being dropped from outbound requests.
• Fixed OpenAI cache affinity and cache breakpoints not persisting across turns.
• Fixed cache-boundary handling for OpenAI-compatible relay services.
• Fixed prompt-cache affinity headers being applied to models or routes without the required capability.
• Fixed mobile Markdown tables overflowing or rendering with unusable column layouts.
• Fixed wide Markdown tables lacking accessible horizontal-scroll guidance.
• Fixed chat layout shifts during streaming and scroll-position changes after streaming completed.
• Fixed desktop model menus overflowing in constrained viewports.
• Fixed installed PWA sessions overriding the device’s system rotation lock.
• Fixed empty Markdown links being incorrectly shown as blocked links.
• Fixed copied code blocks containing an unwanted trailing newline.
• Fixed resumed generation streams retaining stale cursors or replay state after cancellation.

✨ Feature

• Added an admin Usage Statistics dashboard with date range, user, permission group, model, and billing-scope filters, plus trend charts and model/user rankings.
• Added project-level defaults for MCP tools and Skills, so new conversations inside a project can start with the right tool and Skill set automatically.
• Added cancellation handling for chat message submissions, keeping the stream connected long enough to preserve final partial content, usage, and settlement state.
• Added public branding APIs and a dynamic web app manifest, allowing deployed branding to drive frontend metadata, logos, icons, and PWA details.
• Added a generated @deeix/api-contract package to keep backend Swagger contracts and frontend TypeScript API types aligned.

⚡ Optimization

• Improved assistant retry behavior so successful retry branches preserve usable conversation context.
• Improved tool-result handling by estimating tool calls/results in context budgets, trimming oversized outputs, and omitting opaque payloads from model context.
• Improved HTML Visual Prompt handling with theme-aware CSS variables, stricter safe-tag/style constraints, and better artifact preview theme matching.
• Improved project editing UI with dedicated controls for default MCP tools and Skills.
• Improved branding loading across app shell, login, settings, share pages, notifications, and screenshots.
• Simplified HTTP security middleware by removing app-level HTTPS redirects, leaving TLS and redirect policy to the deployment layer.
• Migrated frontend workspace tooling to a monorepo flow with shared contract generation and quality checks.

🐛 Fixed

• Fixed context loss after successful assistant retry.
• Fixed invalid recovered assistant retry lineage being accepted into conversation context.
• Fixed stale PWA caches by retiring the old service worker and clearing legacy cache entries.
• Fixed branding configuration not being consistently exposed through public frontend/runtime endpoints.
• Removed the identity-provider logo proxy path and simplified custom provider logo handling to avoid unsafe logo rendering flows.

✨ Feature

• Added configurable branding for app title, Logo, Favicon, PWA icons, manifest metadata, notifications, screenshots, login page, sidebar, and public share pages.
• Added safer usage billing authorization with per-request budget reservation, reservation renewal, and idempotent settlement for paid model calls.
• Added OpenAI Responses background mode support, with cancellation handling, retry fallback, and usage recovery for interrupted runs.
• Added improved OpenAI Responses image generation handling, including partial previews, final image persistence, and safer trace output.
• Expanded MinerU parsing with configurable support for PDF, Word, Presentation, and Excel.
• Added richer admin conversation logs with upstream, model route, protocol, binding code, usage, and tool details.

⚡ Optimization

• Improved the Files page and preview experience across PDF, sheets, documents, media, images, text, and code, including loading, zoom, fullscreen, and detail views.
• Standardized workspace and sidebar navigation across recent conversations, projects, starred items, search, sharing, exporting, deletion, and bulk actions.
• Improved chat generation state handling for long streams, retry/edit branches, pending messages, and concurrent generation limits.
• Improved billing settings wording by changing prepaid amount into a clearer per-request reservation concept.
• Improved announcement dialogs, Skills/Prompt lists, recent lists, and file lists with more stable infinite loading and dialog display.
• Improved model source configuration with clearer priority and weight descriptions.
• Improved MCP request timeout handling with shared defaults and clearer behavior.

🐛 Fixed

• Fixed usage settlement when the final bill exceeds available balance, so completed paid requests are still recorded and settled correctly.
• Fixed duplicate or retried billing settlement from creating repeated charges for the same usage request.
• Fixed paid generation cancellation/failure paths to release, renew, or mark billing reservations for reconciliation correctly.
• Fixed hardcoded branding across sidebar, login, screenshots, notifications, generated media placeholders, and public-facing views.
• Fixed MinerU file type support, presentation document handling, and related MIME guidance.
• Fixed unstable dialog content during close animations across admin, settings, files, recent, Skills, and Prompt views.

✨ Feature

• Added video generation support for models that can create video content.
• Added cancellation support and usage metrics for image / video generation tasks.
• Added user-side export for all personal conversations, messages, and metadata.
• Added permission groups for model access control and billing rate multipliers.
• Added subscription-plan binding to permission groups, allowing plans to grant model access automatically.
• Added model access rules by model, vendor, protocol, upstream, and related scopes.
• Added OpenRouter / models.dev official pricing fetch and import for faster model pricing setup.
• Added configurable default model for new conversations.
• Added time-based grouping for recent conversations in the sidebar.
• Added drag-and-drop sorting for projects in the sidebar.
• Added reasoning content passback for preserving and reusing upstream reasoning content.
• Enhanced Gemini interactions for video generation and multi-turn workflows.

⚡ Optimization

• Improved desktop model selection menu positioning, scrolling, and dynamic height.
• Improved new-conversation model selection so the user’s chosen model is preserved before a conversation ID exists.
• Improved MCP tool interaction handling to reduce accidental selection behavior.
• Improved the chat side preview rail with dynamic line width based on interaction and distance.
• Improved user message reuse and branch handling for edit, retry, and continuation flows.
• Improved attachment context handling for images and historical conversation data.
• Improved upstream thinking trace display with smoother live updates and flushing.
• Improved media generation task states for queued, running, and result-saving phases.
• Improved permission group error handling and configuration clarity.
• Improved billing page structure by separating model pricing, plans, redemption codes, and tool billing.
• Improved model pricing import with multipliers, remote refresh, and validation.
• Improved sortable list interactions across admin pages.

🐛 Fixed

• Fixed selected model being reset in new-conversation flows.
• Fixed overly frequent upstream thinking trace updates causing unnecessary UI pressure.
• Fixed incomplete permission group error handling.
• Fixed project creation button styling across responsive breakpoints.
• Fixed user export review issues and improved export file handling stability.

✨ Feature

• Added conversation screenshots with full-conversation capture, selected-message capture, preview, copy, and download.
• Added a chat side preview rail for quickly navigating long conversations and previewing nearby Q&A snippets.
• Added admin export for all conversations, messages, and metadata as JSONL.
• Added log cleanup by log type and cutoff date across audit, auth, usage, order, conversation event, and system logs.
• Added model circuit-breaker and source-level governance with failure thresholds, cooldown duration, sliding windows, and source-level scope.
• Added clearer model and upstream availability states, including disabled models, disabled upstreams, disabled upstream models, missing sources, and circuit-open status.
• Added MCP display order management for servers and tools in the conversation picker.
• Added default MCP tools, allowing a single tool or an entire MCP server to be enabled by default for new conversations.
• Added customizable MCP Tool and Skills prompts for admins.
• Added per-conversation memory for selected Skills and MCP tools.
• Added OpenRouter Chat Completions support with updated routing and model option handling.
• Enhanced Gemini tool calling with thought signatures and server-side tool invocation support.

⚡ Optimization

• Improved conversation rendering with stronger Markdown, inline formatting, math, HTML styling, color validation, and URL normalization.
• Improved model picker cards with clearer model details, pricing, free-model labels, and native tool capability display.
• Improved model configuration panels for native tools, reasoning/thinking options, web search, and related controls.
• Improved chat scrolling with a message scroller, position rail, and fade effects for long conversations.
• Improved attachment and message layouts for long content.
• Improved thinking and tool-call displays by separating model reasoning, MCP tool calls, and tool results.
• Improved tool-call flow with better persistence, output handling, tracing, and final display.
• Improved screenshot generation with long-conversation limits, cleaner message metadata, and branding in captures.
• Improved announcement loading and display states with an entry point, empty state, and failure messaging.
• Improved localized timestamp display across chat messages and navigation search.
• Improved admin logs UI with clearer log types, impact descriptions, and high-risk operation warnings.
• Improved model management and upstream source pages with clearer source status, protocol, priority, weight, and circuit-breaker actions.
• Improved Prompt / Skill table rendering stability.
• Improved AuthGuard token handling for more reliable session state.

🐛 Fixed

• Fixed incorrect first-login initial username requirement detection.
• Fixed unstable Turnstile validation behavior during registration.
• Fixed conversation title auto-generation across locales.
• Fixed inconsistent conversation title localization.
• Fixed mobile navigation when starting a new chat from a shared conversation page.
• Fixed Prompt / Skill list rendering issues.
• Fixed model table state typing issues that could affect display or builds.
• Fixed Docker builds missing the default FRONTEND_DIST_DIR.
• Fixed inconsistent submenu styling in dropdown menus.
• Fixed performance and allocation issues when deduplicating tool output strings.

✨ Feature

• Added chat message queuing, allowing users to type and queue the next message while a response is still generating.
• Added user message editing, enabling edited messages to continue as a new response branch.
• Added a “reuse model options” setting so new conversations can reuse the last parameters used with the same model.
• Added improved conversation default model resolution, restoring the right model more accurately when reopening existing conversations.
• Added display currency configuration, allowing admins to show user-facing costs in USD or CNY.
• Added filtering support to the admin user list for more flexible account management.
• Added public share file URL rewriting for more reliable file and attachment access on shared pages.
• Improved compaction summary logic to preserve long-conversation context more accurately.

⚡ Optimization

• Improved chat attachment type handling for code files, special extensions, and MIME compatibility.
• Aligned frontend attachment MIME prechecks with backend upload policy.
• Improved billing and subscription displays, clearly marking free models as not billed.
• Improved payment order handling and CNY conversion consistency across payment flows.
• Improved cost display in the model picker and message metadata.
• Improved MCP client SSE handling for larger event lines.
• Improved mobile sidebar navigation after refresh across recent, starred, and project conversations.
• Improved admin billing refresh behavior to preserve balance and period subscription data.

🐛 Fixed

• Fixed mobile sidebar conversation switching after page refresh.
• Fixed project navigation not using the mobile sidebar navigation behavior.
• Fixed admin billing balance display disappearing after refresh.
• Fixed period-mode subscription data missing from admin user views.
• Fixed inaccurate MIME inference for some uploaded code files.
• Fixed MCP client failures when handling very large SSE lines.

✨ Feature

• Added Skills management, allowing admins to maintain built-in Skills and users to create custom Skills that can be loaded into conversations with /.
• Added a unified Skills and Prompts page for managing reusable conversation context.
• Added / selection in the chat composer for inserting Prompts and loading Skills.
• Added collapsible sidebar sections for recent conversations, starred conversations, and projects.
• Added persistent sidebar state, so collapsed sections are remembered locally.
• Added user activity tracking, with “last active” visibility and sorting in the admin user list.
• Added OpenAI Responses support for instructions and previous response IDs to improve stateful response continuity.
• Added locked model option paths, allowing admins to prevent key model parameters from being overridden in chat.
• Added improved model option filtering and input generation for background text tasks such as titles and summaries.
• Added automatic avatar saving after upload.

⚡ Optimization

• Improved chat composer resizing for smoother long-message input.
• Improved message metadata visibility on touch devices such as tablets.
• Improved sidebar auto-collapse behavior and navigation animations.
• Improved AuthGuard session handling for expired or refreshed login states.
• Improved file processing timeouts and batch parsing responses for more stable extraction workflows.
• Improved native model tool and parameter configuration UI for clearer tool selection and option control.
• Improved tool-run handling with clearer errors when a run does not produce a final answer.
• Improved Prompt Preset and Skill injection logic for more stable conversation context composition.
• Improved chat placeholder text and mention selection interactions.

🐛 Fixed

• Fixed registration email validation to reject control characters and normalize SMTP recipients.
• Fixed cross-platform ID parsing issues caused by integer-size differences.
• Fixed load-more sentinel logic that could retrigger while more items were already loading.
• Fixed chat font variable configuration that could cause inconsistent font rendering.
• Fixed incomplete error handling when tool runs returned no final answer.
• Fixed frontend route type path issues that could affect type generation or builds.

✨ Feature

• Added password reset from the login page with email verification code and admin-controlled availability.
• Added OpenWebUI user import for admins, including preview, email deduplication, balance import, and credit conversion.
• Added user avatar management with uploaded image avatars, generated avatars, and public avatar access.
• Added period billing overage handling: plan credit is used first, then extra usage can be charged from account balance.
• Added chat display customization for content width, chat font, and font weight.
• Added Redis TLS configuration support for encrypted Redis deployments.
• Added HTTP protocol support for OpenTelemetry OTLP exporters.
• Added a billing attribution time field to usage ledger records for more accurate billing history and reporting.

⚡ Optimization

• Improved file avatar lookup performance with a dedicated index for file-backed avatar URLs.
• Improved period billing ledger attribution for more stable cross-period usage and historical billing records.
• Improved OAuth/OIDC login safety by detecting provider email conflicts and guiding users to sign in before binding accounts.
• Improved GitHub and Discord provider email handling with verified-email awareness.
• Improved conversation metadata refresh behavior to reduce unnecessary polling and avoid title generation when no titleable content exists.
• Improved system prompt layer priority behavior with clearer ranking semantics.
• Improved chat mention menu placement and composer tool menu animation.
• Improved Markdown rendering so currency values like $1.23 are not misread as math.
• Improved subscription and balance visibility in period billing mode, including clearer top-up access.
• Improved file deletion behavior by waiting for server confirmation and showing clearer in-use conflict feedback.
• Improved billing input layout in the admin user editor for easier account and period billing management.

🐛 Fixed

• Fixed model editor behavior so model capabilities can be fully cleared.
• Fixed accidental LaTeX rendering for dollar-denominated currency text.
• Fixed confusing OAuth/OIDC behavior when a provider email already belongs to another account.
• Fixed period billing usage flow so users are not blocked immediately after included credit is exhausted when balance can cover overage.
• Fixed no-content conversation metadata cases with clearer localized error handling.
• Fixed file list state changing too early when deletion fails or a file is still in use.