Skip to content

Releases and Updates

Release notes for each Trio Mate version, and the desktop download.

Quick navigation

2.5.19

  • New: MCP services can be authorized and signed in from Settings.
  • Improve: when a finished background task does not return its result to the current conversation, the app shows the reason and offers retry or dismiss.
  • Improve: chat typography and button colors are unified, along with general stability improvements.

2.5.18

  • New: preferences and key details shared in chat can now be remembered instantly, taking effect in future conversations with clear save feedback.
  • Improve: generated artifacts now support quick right-click actions and dropdown switching for faster file viewing and access.
  • Improve: opening Settings now preserves your active conversation and composer state without resetting the chat.
  • Improve: smooth sidebar collapse and expand transitions, with aligned window top chrome and toggle buttons.

2.5.17

  • Fix: Personal Memory now reliably remembers the important information you ask it to keep.
  • Improve: conversations and tool calls now run reliably for a smoother experience.

2.5.16

  • Major upgrade: Personal Memory now remembers key facts and preferences long-term across your devices while protecting your privacy.
  • New: Personal Memory syncs across devices, with sync status and manual sync available in Account Settings.
  • New: view, edit, or remove specific memories directly in chat with a quick confirmation.
  • New: answers now include numbered citations and source links for easy verification.
  • New: browse workspace files and generated artifacts in the chat sidebar, with quick preview and system app launch.
  • Improve: pause or clear Personal Memory anytime in Account Settings without affecting chat history.
  • Improve: new image viewer with thumbnail previews, supporting zoom, pan, and continuous browsing.
  • Improve: automatic retry on connection hiccups without repeating completed actions; fixes occasional chat layout and interaction glitches.
  • Improve: consolidated account and organization settings pages.

2.5.15

  • New: sensitive data protection in the Security Center automatically detects and shields phone numbers, emails, IDs, and bank cards before sending, with smart masking enabled by default.
  • Improve: settings panel now closes automatically after signing in, keeping your view clear.

2.5.14

  • Improve: channel connections and message delivery are reliable, and confirming the QR code completes the connection right away.
  • Improve: channel settings focus on the essentials, and sign-in and sign-out behave the same way across channels.

2.5.13

  • Improve: rebuilt web search to work reliably even on restricted networks, instead of waiting a long time and failing.
  • Improve: search answers now include clickable source links, so you can verify where the information came from.

2.5.12

  • Improve: local file cards can reveal the file in its folder, save a copy, or open it with a system app, so its location is always clear.
  • Improve: image attachments appear and send more smoothly, keeping long editing sessions responsive.
  • Fix: images returned by tools now show up in messages.
  • Fix: the send button is enabled as soon as you paste plain text, so you can send right away.
  • Fix: file preview failures now show the real reason instead of a vague message.
  • Improve: overall stability.

2.5.11

  • Improve: new installs connect to the official service address by default; existing accounts and history remain available after upgrade.
  • Improve: prefer the model's native vision for images, and fall back to on-device recognition only when needed.
  • Improve: channel setup supports QR onboarding, with a separate Lark entry and clearer save results.
  • Fix: keep sign-in continuous after switching organizations; isolate scheduled tasks by account so they do not leak across accounts.
  • Improve: stop automatically showing the organization welcome dialog.

2.5.10

  • Fix: restore messaging for accounts unable to send messages after upgrading.
  • Fix: prevent completed onboarding and usage reminders from appearing again after an upgrade or service address change.
  • Improve: preserve conflicting files during history migration and show the correct old-data options after switching accounts.
  • Fix: display the Agents list in the selected interface language.

2.5.9

  • Fix: after switching to the new API domain, automatically merge existing conversations and history so they remain visible after upgrade.
  • Improve: show enterprise knowledge citations more completely, with clearer entries for sources already read.
  • Improve: make login and intro screen buttons and colors more consistent.
  • Improve: serve account email and profile details from the identity service so they stay continuous after sign-in.

2.5.8

  • New: ask for clarification when key requirements are unclear, with preset choices or a custom answer.
  • Improve: group clarification questions at the end of replies and collapse answered details to save conversation space.
  • Improve: unify drag resizing for the sidebar, composer, and preview, with size reset and preview collapse.
  • Improve: compact visualization cards and separate action buttons from content.
  • Fix: center expanded visualization previews at their content size with padding; oversized content scrolls within the window boundaries.

2.5.7

  • Improve: adapt content previews in the right-hand visualization area to the window, with full-window viewing and less clipped content.
  • Improve: refine conversation scroll following, message navigation highlights, and user message visibility.
  • Fix: preserve folder references when pasting and improve working folder selection and message sending.
  • Fix: resolve title flicker when switching conversations, title bar alignment, and unresponsive overlay buttons.
  • Fix: improve signature verification compatibility for automatic updates on older versions of Windows 10.

2.5.6

  • New: verify scheduled channel deliveries with a test message to the destination before saving the task.
  • Improve: refresh chat, title bar, and sidebar layouts, including group folding, drag sorting, and task status displays.
  • Improve: resize the input area by dragging, enjoy consistent send animations, and see immediate feedback when sending images.
  • Improve: refine settings loading and usage heatmaps, reducing duplicate page loads and conversation list refresh issues.
  • Improve: find tools more reliably and get clearer explanations and recovery guidance for execution failures and access restrictions.
  • Upgrade: bundle an independent Python environment on macOS to reduce document processing dependencies on local setup.
  • Upgrade: improve dependency compatibility, template filling, and structure validation across Word, Excel, PowerPoint, and PDF skills.
  • Fix: exclude time spent waiting for replies or approvals from task execution limits to reduce unexpected timeouts.
  • Fix: keep unanswered clarification cards visible after a reply ends and close their pending state when a task stops.
  • Fix: refine working folder permissions and dialog display, improving Windows folder access and skill execution compatibility.
  • Fix: improve tool-calling reliability and reduce task execution failures.

2.5.5

  • Improve: refine the overall interface layout and interaction experience.
  • Fix: calculate task elapsed time correctly across multi-turn conversations.
  • Fix: resolve builtin Skill identity validation errors.

2.5.4

  • Fix: improve tool-calling reliability in ongoing conversations and reduce unexpected task interruptions.

2.5.3

  • Fix: improve WhatsApp connection, reconnection, and outbound messaging reliability, reducing failures caused by service protocol changes or temporary network instability.
  • Improve: show the failure stage, error code, service node, retry schedule, and handshake protocol version in WhatsApp Advanced Diagnostics.
  • Fix: prevent scheduled and background task histories from duplicating user instructions or omitting replies and tool results.
  • Fix: load scheduled tasks from the correct storage location during startup so existing tasks remain available and updatable.
  • Fix: localize channel sign-out dialogs and WhatsApp QR code labels to match the app language.

2.5.2

  • Improve: enhance the performance and stability of built-in Skills.

2.5.1

  • Improve: warn and guide the Agent to change strategy on repeated tool calls before pausing, avoiding premature task interruption.
  • Improve: keep the chat input prompt simple instead of switching text with every task state.
  • Improve: retain the full pre-compaction conversation so the Agent can retrieve earlier details when needed instead of relying only on the summary.
  • Improve: use two-pass compaction for long conversations, preparing older context in the background and combining recent messages at compaction time, with automatic fallback when preprocessing is unavailable.
  • Upgrade: update the built-in runtime for more reliable legacy-session recovery, continued chats, and cache reuse.
  • Fix: prevent final text or thinking frames from appearing in a sudden jump and clear the generating state when streaming ends.
  • Fix: preserve partial replies, tool results, and conversation records after stopping tasks, compacting long chats, or handling errors.
  • Fix: persist long-conversation compaction results so reopening or continuing a chat does not reload the full history.
  • Fix: compact legacy conversations already beyond the context threshold instead of allowing unusually inflated input token counts.
  • Fix: prevent late Stop or conflicting terminal states from overwriting completed results, and publish background-task completion only after data is saved.
  • Fix: prevent stale requests from overwriting the Usage heatmap and correctly record usage for Responses conversations.
  • Fix: keep ordinary chats available when MCP tool preflight fails and show the correct error category.
  • Fix: show that files and commands are not sandbox-protected when Full Access is enabled.
  • Fix: allow completed, failed, or cancelled background-task cards to be dismissed together with their related conversation data.
  • Fix: restore missing native context menus in chat inputs and other editable fields, localized to the app language.
  • Fix: keep long-conversation navigator previews at a stable width instead of shifting with their content.

2.5.0

  • Add: interrupt a running task with follow-up instructions or a new direction, without starting over.
  • Add: a message navigator for long conversations to jump to any message you sent along the timeline.
  • Add: import Agents and Skills from the Apps home with AI review before import; view details, manage, or uninstall local content.
  • Add: direct network access in non-sandbox mode so tasks can use the local network environment.
  • Improve: reuse stable context in ongoing conversations, leveraging model caching to reduce repeated network calls and model usage.
  • Improve: combine usage trends, feature and Agent distribution, and an activity heatmap on the Usage page.
  • Improve: reorganize Settings and page navigation, including support for a collapsed sidebar, narrow windows, and scheduled-task pages.
  • Upgrade: update MCP integration for the latest protocol.
  • Fix: resolve interface state issues after sending messages, switching tasks, and completing Plans.
  • Fix: resolve some interaction issues with a collapsed sidebar, narrow windows, page navigation, and scheduled-task pages.

2.4.8

  • Improve: keep context stable across consecutive turns for better cache reuse.
  • Fix: avoid writing duplicate dynamic runtime information into conversation history.

2.4.7

  • Add: the Usage page now shows recent records, expandable details, and JSON export, with input, cache, output, and total tokens separated.
  • Add: drag folders into the chat composer to send their files while preserving the directory structure.
  • Fix: automatically continue when model output reaches its limit, instead of stopping the task and asking you to click Continue.
  • Fix: recover corrupted configuration files during startup with a backup and repair flow, with clear status when recovery cannot complete.
  • Fix: remove the empty-state flash when sending the first message or receiving the first reply in a new chat.
  • Fix: explain blocked external resources in HTML previews and render self-contained charts correctly.
  • Improve: preserve explicitly selected Agent workspaces across sends, switches, and restarts.
  • Improve: close Plan checklists when tasks finish normally and keep the completed state after restart.

2.4.6

  • Fix: sandbox policy initialization failures in Windows environments using only local administrator accounts.
  • Improve: task execution stability in specific environments.

2.4.5

  • Fix: addressed known issues.
  • Improve: overall stability.

2.4.4

  • Add: display inline and block math formulas in conversations for calculations and technical content.
  • Improve: open local directory paths directly from chat results for file and workspace tasks.
  • Improve: built-in document, PDF, presentation, and spreadsheet skills now include file checks for more stable creation and editing.
  • Improve: removed the Smart Routing configuration that is no longer needed from Settings.

2.4.3

  • Improve: chat activity and task progress at a glance.
  • Improve: the selected Skill from Apps routing is now preserved when starting a new chat, reducing repeated selection.
  • Improve: pasting rich-text content now preserves usable text more reliably, reducing information loss.
  • Improve: increased reliability after Agent tasks finish, with a better diagnosis and recovery experience when something goes wrong.

2.4.2

  • Major update: image text recognition is now fully built into the app. No Python setup or additional model download is required, and OCR works offline from the first use.
  • Improve: the OCR engine is reused reliably throughout the app session. Switching personal workspaces, organizations, or conversations no longer rebuilds the recognition environment, making attachment processing faster and more dependable.
  • Fix: expired attachment access and unstable message binding when editing and resending messages with historical attachments.

2.4.1

  • Improve: PowerShell startup compatibility in sandbox mode.
  • Fix: stability issues affecting messaging channels, historical attachments, workspace history, and scheduled tasks.
  • Improve: the macOS stable installation experience and security for diagnostics and local data.

2.4.0

  • Add: the Apps catalog brings together installed agents, skills, connectors, and available content. Filter by type, category, or role and start using them directly.
  • Add: draft input is saved per conversation, so you can continue editing unsent content after switching conversations or workspaces.
  • Add: the usage page shows tenant Token usage and validity, with a reminder as you approach your allowance.
  • Add: the Web channel can connect to an online desktop device. Create or continue tasks in your browser and follow their progress and results in real time.
  • Add: choose agents, skills, and desktop workspaces from the Web channel, and attach files to a task.
  • Add: the Web channel shows device availability and conversation history, so you can return to the browser and continue following a task.
  • Add: the Web channel shows operations that need confirmation. Sensitive actions can require reauthentication before a decision is submitted.
  • Fix: incorrect ordering or grouping of temporary and newly created conversations in the sidebar.
  • Fix: missing current-time context for scheduled tasks.
  • Fix: an incorrect restore button after maximizing the window.
  • Fix: missing fonts when creating Chinese content with document, spreadsheet, and presentation features.
  • Fix: the originally selected agent was not restored when continuing a Web conversation, and completed remote tasks could leave residual interface state.
  • Improve: recovery and feedback for short network interruptions during a remote Web task, preventing conversations from remaining in a waiting state.
  • Improve: built-in file tools now read long files, apply multiple edits, and write content more reliably. A cancelled or failed task does not leave a partial modification.
  • Improve: web content retrieval now extracts long pages and finds relevant passages more reliably.
  • Improve: external file handling under the workspace-contained sandbox policy, so a task cannot access files outside the current workspace.
  • Improve: Security Center policy interactions are clearer: sandbox policy changes take effect per conversation, while approval and network policies take effect from the next task turn.
  • Improve: confirmation prompts now present risk more clearly before you continue.

2.3.1

  • Fix: incorrect knowledge permissions when granting access to employees and departments.
  • Fix: duplicate messages sent through alternate entry points while a reply was generating.
  • Fix: web_fetch incorrectly blocked by the sandbox policy for public web URLs.
  • Fix: incorrect scenario assistant avatars.
  • Fix: the material picker remaining visible after a chat starts.
  • Fix: stale scenario labels remaining after switching organizations.
  • Fix: AI title and permission-advice requests incorrectly blocked by content moderation.

2.3.0

  • Add: the Apps home now unifies agents, skills, and apps, with category filters and a view of installed items.
  • Add: five scenario assistants for finance, HR, legal, marketing, and sales, with capabilities and skills organized by scenario.
  • Add: scenario assistants can search authorized organizational materials and show readiness, keeping answers grounded in relevant content.
  • Add: MCP apps are now available with connection checks and status details, making tool sources and availability clearer.
  • Add: agents can save scoped memories, avoiding information overlap between agents.
  • Add: local agents and skills can be imported in tenant workspaces while the organization store stays centrally managed.
  • Add: a built-in PDF parsing skill is available for local PDF files.
  • Add: usage data collection can be disabled in Settings, with your choice remembered.
  • Add: the local data storage location can be customized per device environment.
  • Improve: agent workspaces and materials are isolated by conversation to prevent cross-task mixing.
  • Improve: conversation search matches history faster and presents results more clearly.
  • Improve: prompt history preserves the user's visible language and skill summaries for a more complete review.
  • Improve: scheduled task and task-plan states now use a consistent presentation, with no lingering loading state after completion.
  • Improve: assistant activity is shown as concise summaries that are easier to scan.
  • Improve: HTML/React preview scrolling is consolidated, eliminating extra scrollbars for long content.
  • Improve: the login page, app icons, and window visuals now use a consistent Trio Mate brand style.
  • Fix: Windows and macOS installers now block unsupported operating system versions and explain why.
  • Fix: Windows installation now handles required runtimes, reducing post-install startup failures.
  • Fix: inconsistent display and binding states across scenario assistants, organizational materials, personal workspaces, and tenant permissions.
  • Fix: external links, Markdown content, and agent-material access now enforce clearer security boundaries to reduce unnecessary data exposure.
  • Fix: moderation blocks, gateway quota errors, and network interruptions now show clearer messages and finish gracefully.

2.2.0

  • Add: the task plan panel now shows task steps, execution progress, and completion state above the conversation.
  • Add: desktop now receives task plan changes, including expanded steps, completion, and failure state.
  • Add: enterprise knowledge search is available to Agents inside desktop conversations.
  • Add: channel message sending is available from conversations to external channels.
  • Add: Agent detail cards show avatars, and the Skill/Agent Store supports category filters and paginated loading.
  • Add: Skill content follows the current language, keeping store previews aligned with local content.
  • Add: the Security Audit Center displays high-risk tools, sandbox activity, and permission records.
  • Add: HTML/React artifacts open in an isolated preview environment to reduce local data exposure risk.
  • Improve: task execution, schedules, sub-tasks, and prompt rendering now use a unified desktop engine path.
  • Improve: background task and schedule storage migrations preserve existing history after upgrade.
  • Improve: tool-call status now uses lightweight indicators, with cleanup after recovery and final replies.
  • Fix: files pasted from Finder or the system clipboard are handled as attachments while preserving multi-file order.
  • Fix: HTML previews support horizontal and vertical scrolling, preventing long content and wide tables from being clipped or covered.
  • Fix: Windows install paths, VC++ runtime handling, publishing scripts, shortcut labels, and install cache handling.
  • Fix: the Agent page sidebar, personal tenant labels, store titles, empty avatar states, and knowledge hints now display consistently.
  • Fix: stale state in historical tool calls, hover loading, duplicate notifications, and binding checks.

2.1.2

  • Improve: the file drop zone is expanded from the chat input to the entire conversation page, making attachment seamless.
  • Fix: fixed an issue where copying a full web page as an image could cut off long pages.
  • Fix: fixed repeated confirmation prompts for certain common actions, reducing task interruptions.
  • Fix: error messages are now clear and actionable, with improved recovery from network fluctuations or unexpected interruptions.
  • Improve: upgraded the desktop engine for stable OCR recognition, message editing, and overall operation.

2.1.1

  • Add: Office document skills now prefer local Office capabilities, preserving layout fidelity when converting Word, Excel, and PowerPoint files to PDF.
  • Fix: tool-call status display now uses lighter prompts, keeping conversation content more focused.
  • Improve: Windows sandbox execution stability.

2.1.0

  • Add: sidebar workspaces and conversation lists now support drag-and-drop reordering with a redesigned interaction.
  • Add: pasted text in the chat input box can now be edited directly before sending.
  • Add: local file paths in chat are recognized as openable links, making it easier to access generated and project files.
  • Add: added a channel send message tool for sending messages to external channels directly from conversations.
  • Add: WeChat settings now include proactive message toggles for managing push behavior.
  • Add: the skill store now supports taxonomy filtering and role-title display for faster skill discovery.
  • Add: added a skill update check entry in Settings to view and update installed skills.
  • Add: added sandbox execution policy settings and approval controls for customizable command execution boundaries.
  • Add: generated files are now displayed as file blocks with visible file sizes.
  • Fix: fixed Mermaid diagram images not being copyable to the clipboard.
  • Fix: fixed temporary conversations not being correctly nested under their workspaces.
  • Fix: fixed recovery display for empty replies and max-turns fallback scenarios.
  • Fix: fixed missing binding notifications popping up repeatedly.
  • Fix: fixed sidebar not staying at the latest message position after collapsing.
  • Fix: fixed generated file paths being exposed in conversations.
  • Fix: fixed local file links not opening correctly on Windows.
  • Fix: fixed the conversation list not refreshing after invite redemption, with permissions not taking effect immediately.
  • Fix: fixed organization skills not appearing in Settings.
  • Improve: improved tool call result display with clearer summaries.
  • Improve: improved permission dialog layout for more intuitive approval workflows.
  • Improve: improved conversation search performance and matching accuracy.
  • Improve: unified multi-channel message delivery experience.

2.0.4

  • Fix: tray and menu bar no longer show old icons.
  • Improve: workspaces can now start chats and rename directly.

2.0.3

  • Fix: answering assistant follow-up questions now continues to a visible result, keeping selected information in the current task.
  • Fix: input history no longer mixes in skill prompt content, keeping history navigation limited to what the user actually typed.
  • Improve: assistant follow-ups are confirmed one question at a time, keeping context order clear across multiple choices.

2.0.2

  • Add: local file paths in chat are recognized as openable links, avoiding manual path copying for generated and referenced files.
  • Add: sidebar conversations are organized by workspace, with current-workspace and temporary-chat entries grouped together.
  • Add: organization profiles can be edited in the desktop app, allowing team details to be updated directly.
  • Fix: Mermaid diagrams now render in an isolated sandbox, reducing page interference and repeated artifact replay.
  • Fix: built-in skill runtimes stay isolated per skill, keeping execution state stable after skill updates or installation.
  • Fix: the input box focuses after the new-chat shortcut so the next task can be typed immediately.
  • Fix: locked agents no longer show an unavailable remove action, avoiding confusion about whether the current choice can be cleared.
  • Improve: enhanced the built-in PDF, Word, Excel, and PowerPoint skills so the desktop app covers more common office-document tasks.
  • Improve: the sidebar supports width resizing, remembered collapse state, and pinned grouping for clearer navigation through long chat lists.
  • Improve: non-blocking activity-log errors are toned down so execution feedback stays focused.

2.0.1

  • Fix: aligned desktop app icons with native macOS and Windows shapes so launcher, taskbar, and window icons no longer show a square plate.

2.0.0

  • Add: upgraded the desktop home screen, sidebar, sign-in page, and input box with a reorganized action layout.
  • Add: added desktop organization management so team names, members, and current account details can be reviewed in one place.
  • Add: added account email binding and management, covering sign-in, rebinding, and duplicate-email feedback.
  • Add: added secure browser-to-desktop launch after sign-in, reducing accidental account handoffs.
  • Add: grouped file attachments, skill selection, quick menus, and skill completion in the new-chat input box so tasks can start from the input area.
  • Add: added Spanish UI and release notes to cover multilingual usage scenarios.
  • Add: refreshed Trio Mate brand icons, startup visuals, and avatar styling for a unified desktop look.
  • Fix: isolated local conversations by account and organization so identity switches no longer mix history from another account.
  • Fix: migrated legacy account conversation folders so history archives stay in a stable location on Windows.
  • Fix: reduced interruptions when long conversations generate files, preserving long-task results.
  • Fix: completed interaction records for older conversations so review keeps the full context.
  • Fix: refreshed agent recommendation questions from the current selection so they stay aligned after switching agents or context.
  • Fix: aligned account, organization, and email-binding state so sign-in and rebinding feedback stay consistent.
  • Improve: split settings into account, AI capabilities, smart routing, and interaction sections so option changes follow a direct path.
  • Improve: removed chat background and logo settings so appearance controls focus on the current desktop visual system.
  • Improve: unified agent icons, user avatars, and member avatars so identity cues stay consistent across lists.
  • Improve: clarified the input background hierarchy in dark mode so the input area no longer blends into the page at night.
  • Improve: emphasized the username in the sidebar account block while keeping organization details visible, making the current identity easy to confirm.

0.2.33

  • Fix: prevented background-service crashes when handling long web content or command output.
  • Fix: truncation no longer splits Chinese and other multi-byte characters, reducing garbled text and failed runs.

0.2.32

  • Add: integrated local skill entries into the skills list, streamlining importing, managing, and starting new chats from the store.
  • Fix: resolved issues where some models only displayed thinking processes or text was overwritten at the end of long replies, ensuring complete response display.
  • Fix: resolved occasional blank pages, lost selection states, and card overflows after selecting or installing agents and skills, improving reliability when starting new chats.
  • Fix: optimized permission inheritance and Python dependency installation for subagents, ensuring smoother and uninterrupted automated task execution.
  • Improve: reduced sensitive data exposure in system logs, enhancing sign-in and diagnostic record security.
  • Improve: refined categories and empty-state feedback in the skill and agent stores to optimize browsing and installation workflows.

0.2.31

  • Add: skills can now be browsed and published in the store, and new chats can start directly from agents or skills.
  • Fix: new chats stay as drafts until the first message, so they no longer appear in the conversation list too early.
  • Fix: removed duplicate navigation entries and the welcome-screen back button, making agent selection clearer.
  • Fix: cleaned up older Agent wording so the interface consistently refers to agents and skills.
  • Improve: agent recommendation questions on the home screen can now be generated and cached, making new-chat prompts better matched to the selected agent.

0.2.30

  • Add: added a prevent-sleep toggle so the computer can stay awake while long tasks are running.
  • Fix: reduced interrupted replies and streaming parse failures for some models, making long responses more stable.
  • Fix: restored desktop tool passing and parsing during bridged execution, making file and conversation tasks more reliable.
  • Fix: session renames on Windows now retry after transient failures, reducing out-of-sync conversation titles.

0.2.29

  • Fix: resolved false failure reports when deleting conversations on Windows, ensuring the conversation list stays in sync.

0.2.28

  • Fix: resolved long permission dialog content being clipped or unable to scroll.
  • Fix: added confirmation before deleting conversations and fixed lists not restoring promptly when deletion fails.
  • Fix: visual generation (SVG/HTML/charts) now shows progress earlier for clearer generation feedback.
  • Fix: improved response speed for automatic titles and permission advice.
  • Improve: usage statistics are now handled by the server, simplifying local desktop records.

0.2.27

  • Add: upgraded the desktop task execution path so tool calls and conversation handling stay consistent.
  • Fix: time-related requests no longer lose desktop tool access, making current-conversation and file tasks more reliable.
  • Fix: channel images and attachments are preserved in history, and related data is cleaned up when a conversation is deleted.
  • Fix: scheduled-task conversations now match the chat message style, making images, attachments, and execution steps easy to review.
  • Improve: long conversations open on recent history first, reducing lag when switching sessions.

0.2.26

  • Fix: message rendering flicker on some devices with non-standard resolutions.

0.2.25

  • Add: chat images can now be opened in a larger preview, making details easier to inspect.
  • Fix: older conversations now show more complete assistant actions and results, making them clearer to review.
  • Fix: reduced blank replies, out-of-order thinking text, and copy buttons on empty messages, keeping conversation history cleaner.
  • Improve: images, replies, and step cards now fit the window better, making mixed text, images, and long content easier to read.
  • Improve: image chats and history loading are more stable, with clearer waiting and restore feedback.

0.2.24

  • Fix: reduced first-message flicker and card jump-back in new chats, making the transition into the conversation smoother.
  • Fix: automatically handle image messages for models without image support, keeping replies more reliable.
  • Improve: unified the waiting state before replies, making streaming output connect more smoothly.
  • Improve: added a safety cooldown for dependency installs, reducing supply-chain risk.

0.2.23

  • Add: chats now support pasting clipboard files and images directly as attachments, making context sharing faster.
  • Add: conversations can now be pinned, keeping important chats at the top for easier switching and return visits.
  • Add: Agent chats now bind to a workspace, so the working directory is clear before a task starts.
  • Add: Settings now offers Auto, Fast, and Best smart-routing preferences, making model routing better match each use case.
  • Fix: historical images, artifact placeholders, and attachment references are less likely to duplicate or disappear when conversations are replayed.
  • Fix: Agent creation, selection, and built-in Agent recognition are more reliable, so new and built-in agents appear in the right places.
  • Fix: several window, sidebar, and scroll-area layout boundaries now behave more consistently, reducing clipped content and mismatched scrollbar styles.
  • Fix: date and time recognition now degrades more safely when parsing fails, reducing interrupted chat requests.
  • Improve: Settings and Usage are more centralized, making account, appearance, language, logs, and usage information easier to review.
  • Improve: WeChat channel settings are more streamlined, keeping common options easier to configure and maintain.
  • Improve: chat runtime, background tasks, and external-channel sessions restore history more consistently, keeping context steadier when switching sessions.
  • Improve: WeChat, Feishu, and related channel flows are more stable, with more reliable message records, retries, and status feedback.

0.2.22

  • Improve: faster and more stable software updates for a smoother upgrade experience.
  • Improve: optimized header interactions on the Agents page for Windows users.

0.2.21

  • Fix: chat scrolling now follows new replies more naturally, with fewer jumps or missed auto-scrolls while reading.
  • Fix: desktop dialogs, dropdowns, and other floating layers are less likely to be blocked by the title-bar drag region, making menu clicks and dialog closing more reliable.
  • Improve: Agent Store cards are easier to browse, with clearer names, icons, versions, platform badges, and install states.
  • Improve: the Agents page now uses more consistent tabs and action menus, making installed agents, Store agents, and Agent publishing easier to manage.
  • Improve: the publish-Agent dialog handles long content scrolling more reliably, so descriptions, capabilities, and changelog fields are less constrained by window height.
  • Improve: terminal tool runs now keep the full command string, making permission prompts and policy checks better match what the user actually typed.

0.2.20

  • Fix: when external channels, scheduled tasks, or background tasks handle content automatically, they no longer produce preview results that only work in the main desktop chat, making replies faster and more stable.
  • Fix: Feishu, WeCom, WeChat, and WhatsApp keep automatic handling permissions after starting or resetting a session, reducing repeated confirmations.

0.2.19

  • Fix: external channels now merge consecutive messages, images, and files into the same conversation turn more reliably, reducing misplaced, missing, or duplicated replies.
  • Fix: external channels now correctly use the default YOLO setting, so automated message handling needs less manual handoff.
  • Improve: progress updates while external channels handle messages are cleaner and shown less repetitively.

0.2.18

  • Add: WeCom channels now support QR-code onboarding and receiving media attachments such as images.
  • Add: Feishu and WhatsApp channels now support receiving attachments, with clear in-chat feedback when downloads fail.
  • Add: channel sessions now support /new, /reset, and /help commands with localized replies.
  • Improve: channel execution progress now streams into Trio Mate conversations in real time, and WeChat replies show typing status while streaming.
  • Fix: WeChat messages with attachments now enter the same conversation turn in the correct order, reducing mismatches between text and attachments.
  • Fix: Feishu group-chat configuration upgrades now preserve existing group access permissions instead of overwriting them with defaults.
  • Fix: oversized Markdown tables in Feishu cards are split automatically to reduce message send failures.

0.2.17

  • Fix: Intel Macs now choose a compatible ONNX Runtime version when installing local OCR, avoiding failures caused by missing platform wheels.
  • Fix: WeChat QR sign-in now shows the verification-code input when needed, with clearer scan, expiry, and verification prompts.

0.2.16

  • Add: WeChat channels now support receiving image attachments, which automatically use local offline OCR.
  • Add: WeChat channels now support /new and /reset commands, so context can be reset directly from WeChat.
  • Improve: Feishu, WeCom, WeChat, and WhatsApp channels now group messages and attachments that arrive close together, reducing incomplete replies caused by split callbacks.
  • Improve: channel sessions and scheduled tasks now default to YOLO mode, making automated work smoother.
  • Fix: channel sessions now show tool calls and reasoning instead of only saving the final reply.
  • Fix: improve offline OCR service stability on Windows.

0.2.15

  • Add: image attachments are recognized with local OCR first, so text inside images can be included in chat context.
  • Add: Local OCR management is now available in Settings, including status, install, repair, and uninstall actions.
  • Improve: delete confirmations for scheduled tasks, skills, and memory are clearer to reduce accidental removal.
  • Fix: conversations opened from Feishu, WeCom, WeChat, and other channels stay read-only to prevent accidental edits.

0.2.14

  • Add: PDF, Word, Excel, PowerPoint, and other common office documents can now be added directly to chats.
  • Improve: software update downloads are more stable, with download status shown sooner after you click.
  • Improve: pointer feedback is more consistent when hovering over clickable buttons and menus.

0.2.13

  • Fix: browser sign-in could fail to open Trio Mate in certain environments.
  • Improve: quitting the app now asks for confirmation, reducing accidental interruptions.
  • Improve: desktop stability across different system environments.

0.2.12

  • Add: Trio Mate can now start background tasks, so long-running work can keep progressing without occupying the current chat.
  • Add: the sign-in page now supports Google, Microsoft, and other third-party accounts through OAuth 2.0 / OpenID Connect, making enterprise account setup more flexible.

0.2.11

  • Add: authors can declare supported platforms when publishing agents to the Store, and the Store now shows macOS / Windows platform badges while browsing.
  • Improve: scrollbars are more consistent across chat, settings, usage, and scheduled-task views, making long content easier to browse.
  • Fix: simultaneous tool permission requests are handled in order, reducing dialogs that overlap or block the flow.
  • Fix: expired tool permission dialogs now clear automatically instead of lingering and affecting later actions.
  • Fix: the system prompt field in Prompt settings no longer pushes the options below out of view.
  • Fix: the publish-Agent dialog now stays within the default window size, keeping footer actions visible.

0.2.10

  • Fix: LLM streaming recovery is steadier for long-context or slow-response cases, and replies are more likely to continue correctly after tool execution.
  • Fix: Python tool dependency installation and code execution now use the same managed environment, reducing cases where an installed package still cannot be imported.
  • Improve: same-session authorization memory is more accurate for similar command work, reducing repeated confirmations.
  • Improve: Agent and Skill workspace detection is clearer, so temporary session folders no longer interfere with project capability discovery.
  • Improve: macOS menu-bar and Windows tray / app icons are clearer, reducing black blocks, undersized glyphs, and blurry display.

0.2.9

  • Add: files can now be dropped directly into the chat composer, with clear feedback for unsupported files.
  • Add: prompt settings now include a global YOLO default mode, and the sidebar has a direct settings shortcut.
  • Improve: settings now use a left-navigation layout, and Back returns to the page you came from.
  • Improve: usage now uses separate Overview and Breakdown views, and Back returns to the page you came from.
  • Improve: tool-call and Render UI loading states are steadier and no longer disappear too early during long-running tool work.
  • Fix: transient network or LLM streaming drops now retry automatically, reducing unexpected failures in long tasks.
  • Fix: large tool results are saved by reference instead of being pushed directly into context, and interrupted tool results can continue through the flow.

0.2.8

  • Fix: when a tool permission is explicitly denied, the current task now stops immediately instead of continuing down the wrong path.
  • Fix: session-scoped permission upgrades are now remembered correctly, reducing repeated confirmations and making the approval flow more stable.
  • Improve: streamline startup by removing the extra splash window and tightening the transition timing.
  • Improve: large text attachments are now passed as file references, making long-content handling more stable and more context-efficient.

0.2.7

  • Fix: isolate the desktop web-login session from the browser so sign-in state no longer leaks across them.
  • Fix: refresh the Windows main and tray icon assets so the old blurred/misaligned icons no longer show up.
  • Fix: tighten compatibility and the auto-move behavior when macOS launches from outside /Applications, reducing accidental moves.
  • Improve: the handoff sign-in page now renders the account info without the editable-input look.

0.2.6

  • Add: pasting long text into the chat input now collapses into a chip while the full text is saved into conversation history, making sending and browsing lighter.
  • Add: the Agent Store now supports uploading new versions for already-published agents, smoothing Agent iteration.
  • Improve: refreshed desktop app icon for a more cohesive and polished look.
  • Improve: SKILL doc reading is now structured, and chat network errors retry with backoff for steadier long tasks.
  • Improve: tightened auth session lifetime and the web refresh flow for more reliable sign-in state.
  • Improve: cron tool visibility within cron sessions is tightened to avoid recursive triggers.
  • Improve: clearer `run_process` tool contract guidance for steadier model calls.

0.2.5

  • Fix: the model settings page now stays in sync with upstream provider catalog changes, reducing stale or missing model entries.
  • Fix: adds a recovery migration for users who previously missed the `Trio Mate` -> `Trio Mate` data move, making old chat history restoration more reliable after upgrade.

0.2.4

  • Add: the login dialog now offers "Sign in via browser", letting you finish sign-in in your browser and return to the desktop app automatically, skipping another email code round-trip.
  • Add: refreshed layout and copy for Store filters for a smoother browsing experience.
  • Fix: smoother startup flow, and migration now finishes before the main window is shown for more consistent startup feedback.
  • Fix: corrected the nested button layout inside model cards for steadier interactions.
  • Improve: unified HTTP request logging across the main process for more consistent diagnostics.

0.2.3

  • Add: the chat Agent picker now supports search, so you can filter agents faster by name, description, or entry file.
  • Fix: tool execution now recovers more reliably in complex tasks, reducing cases where the chat shows a tool call but does not keep working.
  • Improve: long tasks now organize search and tool output earlier, and complex workflows get a higher default step budget for a smoother path to completion.

0.2.2

  • Fix: in YOLO mode, write and edit actions in chats without a workspace now follow the auto-execution policy reliably.
  • Fix: once a tool call is explicitly denied, repeated retry loops now stop promptly to keep conversations responsive.
  • Improve: safe JavaScript process handling in the command execution path is now unified for more stable runtime behavior.

0.2.1

  • Fix: YOLO mode is now more reliable in channel conversations, and auto-execution follows the active chat correctly.
  • Improve: YOLO mode now handles routine confirmations automatically, while dangerous actions remain blocked.

0.2.0

  • Add: policy engine with permission confirmation dialog — file writes and command execution now prompt for approval, with an option to ask AI for advice.
  • Add: YOLO mode (auto-approve) to skip routine confirmations; dangerous commands are still intercepted.
  • Add: file backup and rollback — files modified by AI can be restored with a single "Restore to here" action.
  • Add: prompt history navigation — press ↑/↓ in the input box to browse and reuse previous inputs.
  • Add: streaming output for command execution, showing real-time progress.
  • Add: WeChat and WeCom (Enterprise WeChat) channel integrations.
  • Add: Agent Creator for building and publishing agents from conversations, with a new file tree preview for outputs.
  • Add: support for downloading multimedia attachments; improved stability for media previews.
  • Improve: Windows-specific optimizations including long path support and fixes for Python encoding issues.
  • Improve: simplified sidebar conversation list UI and added auto-focus for the login flow.
  • Improve: macOS prompt to move the app to /Applications on first launch.
  • Improve: significantly faster Python execution with smarter runtime caching and environment reuse.
  • Improve: unified branding to Trio Mate, including automatic user data directory migration.
  • Fix: more reliable @mention recognition and context continuity in Feishu group chats.
  • Fix: Python encoding and script compatibility issues on Windows.
  • Fix: incorrect Enter key behavior in the login modal.

0.1.10

  • Add: appearance theme settings for switching the chat background, logo, and interface color palette.
  • Fix: Feishu, WhatsApp, and scheduled tasks are now correctly included in usage statistics.
  • Fix: resolved cases where the AI had finished but the desktop app did not show the final reply.
  • Fix: when a task reaches its execution limit or returns an empty final reply, the app now gives a clearer closing message instead of appearing unresponsive.
  • Fix: resolved an occasional blank-screen issue on some devices, making startup recovery more reliable.
  • Improve: increased the default execution budget for complex tasks and added repeated tool-call protection to reduce loop-related slowdowns.

0.1.9

  • Add: Feishu replies now support interactive cards with rich text (Markdown).
  • Fix: follow-up replies in Feishu group chats and topics are now more accurate.
  • Improve: context carryover in ongoing Feishu conversations is now more stable.
  • Improve: long conversations now use context more intelligently, making replies more coherent.

0.1.8

  • Improve: Feishu and WhatsApp channel settings now present connection status and setup guidance more clearly.
  • Fix: follow-up messages in Feishu now preserve context more reliably, reducing the chance of losing earlier conversation context.
  • Fix: channel conversation titles now update automatically after new replies, making different threads easier to distinguish.
  • Fix: channel conversations now clearly remind you to reply in the original channel, reducing accidental sends from desktop.
  • Fix: running conversations are now protected from accidental deletion, helping avoid interrupted work.

0.1.7

  • Add: edit conversation titles directly to keep chats organized more easily.
  • Improve: smoother sidebar conversation actions and clearer display for long titles.
  • Improve: improved window dragging experience on Windows.
  • Fix: agent and workspace bindings are now more stable in conversations.

0.1.6

  • Improve: stopping long-running actions manually now responds faster and finishes more smoothly.
  • Fix: message sending and session saving are now more reliable during rare reconnection issues, reducing occasional timeouts.
  • Improve: selected skills are now displayed more clearly and are easier to recognize.

0.1.5

  • Add: each thinking section now opens by default and collapses with a short delay after completion for clearer feedback.
  • Fix: cancelling long-running actions now stops more quickly and reliably.
  • Fix: improved background execution stability to reduce repeated retries during unexpected failures.

0.1.4

  • Add: Agent Store for browsing, searching, and installing AI agents with one click.
  • Add: Agents page now uses a tab layout (Store / Installed), defaulting to Store.
  • Add: store-sourced agents show a source badge in the Installed list.

0.1.3

  • Add: brand new Usage Statistics page with multi-dimensional token consumption tracking.
  • Add: redesigned Feedback page (formerly Logs) with support for attaching conversation context.
  • Improve: ensure selected Agents have higher persona precedence in chat sessions.

0.1.2

  • Add: copy button shown on message hover.
  • Add: edit and resend previously sent messages.
  • Add: one-click diagnostic log upload from the logs page.
  • Improve: overall interface layout optimizations.
  • Fix: display issues on certain pages.

0.1.1

  • Add: `Cmd+N` / `Ctrl+N` shortcut to quickly start a new chat.
  • Add: back-to-chat button added to settings, agents, and scheduled tasks pages.
  • Improve: redesigned chat bubbles for a more comfortable reading experience.
  • Improve: thinking process and tool calls now display as compact single lines with hover preview.
  • Improve: visual refinements to sidebar and chat area scrollbars.
  • Improve: scrolling up during streaming no longer gets pulled back by new messages.
  • Improve: on Windows, the installer now automatically closes the running app before updating.
  • Improve: various UI refinements and stability improvements.

0.0.21

  • Add: search conversation titles and message content quickly with `Cmd+K`.
  • Add: artifact preview now supports copying source and exporting files.
  • Fix: resolved bash tool failures when skill script paths contain spaces.

0.0.20

  • Improve: enhanced compatibility of web_fetch tool with certain enterprise websites.
  • Fix: fixed content not refreshing when switching scheduled tasks.

0.0.19

  • Fix: resolved a startup crash caused by a main-process bundle conflict in certain newly released builds.
  • Fix: corrected the auto-update feed issue that caused repeated downloads when the same version metadata was reused.

0.0.18

  • Add: support for individual and bulk deletion of cron job history with confirmation dialogs.
  • Add: enhanced skill isolation in Agent mode, ensuring specialized agents only load their dedicated capabilities.
  • Add: significantly improved script compatibility on Windows for more stable automated task execution.
  • Improve: redesigned the Scheduled Tasks (Cron) interface with unified headers and enhanced responsive panels.
  • Improve: refined history logs with auto-selection of the latest record and cleaner message previews.
  • Improve: standardized Agent configuration for a more consistent user experience.

0.0.17

  • Add: support for Sub-Agent mode, enabling specialized agents for complex tasks.
  • Improve: refactored "About" and "Version Log" pages with standardized `yyyy-MM-dd` date formatting and improved version matching logic.
  • Improve: updated the chat home logo banner and removed opacity filters for a sharper visual experience.
  • Improve: various stability optimizations and performance enhancements for a smoother user experience.

0.0.16

  • Add: full support for Portuguese (pt) language pack, including translated historical release notes.
  • Improve: refactored release metadata and removed redundant Markdown folder in favor of centralized YAML management.
  • Fix: optimized main process network request mode to resolve the "fetch failed" error in certain environments.

0.0.15

  • Add: support for rendering image content blocks in AI responses, making multi-modal interactions more intuitive.
  • Improve: redesigned the release notes layout by aligning version metadata and removing redundant titles for a more compact UI.
  • Improve: added a 10-minute memory cache for release notes to speed up page transitions.
  • Fix: optimized the default selection and configuration check for the built-in TySmart provider for smoother connection.
  • Fix: resolved a race condition in image saving to improve the stability of attachment handling.
  • Refactor: decoupled the changelog URL from the renderer and moved its configuration to the main process for better maintainability.

0.0.14

  • Add: built-in T&Y Smart x HKBN model provider with no extra configuration required.
  • Add: automatic conversation titles to help organize chats.
  • Add: manual conversation renaming (double-click the conversation title).
  • Improve: single-instance mode; relaunch focuses the existing window.
  • Fix: Windows tray icon resources not bundled correctly.
  • Fix: clearing Base URL now works when the API key is masked.
  • Fix: Windows update feed now includes release notes for this version.

0.0.13

  • Add: feedback entry in the sidebar with direct access to feedback channels and submission guidance.
  • Add: usage tracking page for viewing usage data inside the app.
  • Improve: upgrade to React 19 to strengthen the desktop app foundation.
  • Fix: replace the ASCII art banner with SVG to resolve rendering issues on Windows.
  • Fix: use `.ico` assets for the Windows tray icon to avoid incorrect icon rendering.

0.0.12

  • Add: multi-color theme support with 5 color palettes (Teal, Pink, Blue, Green, Orange) to choose from.
  • Add: custom titlebar with frameless window support and window control buttons.
  • Add: Appearance settings page for theme color switching.
  • Add: cron job edit dialog for modifying existing scheduled tasks.
  • Improve: cron job creates independent conversation per execution to avoid sharing the same conversation context across multiple runs.
  • Improve: updated app icons to official Trio Mate branding.

0.0.11

  • Add: you can now stop the current chat session while a streamed response is in progress, with clearer in-progress and stopped feedback in the UI.
  • Improve: session errors are isolated to the affected chat, reducing the chance that one failed session disrupts other active conversations.
  • Add: expanded skill compatibility so workspace and agent projects can discover skills and legacy commands from `.claude/skills`, `.claude/commands`, and `.agents/skills`.
  • Add: model lists are now persistently managed in Settings, with support for manually adding, editing, deleting, and preserving fetched model capability metadata.

0.0.10

  • Fix: resolved shell PATH initialization in the packaged desktop app, reducing cases where CLI tools cannot be found at runtime.
  • Add: when reopening WhatsApp settings during QR login, the latest QR code now remains visible instead of forcing a full refresh cycle.
  • Refactor: reworked WhatsApp reconnect handling to distinguish disconnect reasons and apply backoff retries only when reconnection makes sense.

0.0.9

  • Added a model picker to the chat composer so you can switch models per conversation without changing the global default.
  • Reworked the composer toolbar to keep Agent, Workspace, attachments, and send actions in one place for a smoother chat flow.
  • Slash-triggered Skills are now explicitly selected before sending, with removable skill chips that make combining multiple Skills more direct.
  • Refined the update notes scroll area so changelog content is more stable after checking for updates.

0.0.8

  • Added the full Skills system, including slash-triggered Skills in chat and centralized management in Settings.
  • Improved long-conversation overflow handling so extended chats stay more stable and usable.
  • Hardened WhatsApp session routing to reduce session splits, misrouting, and unexpected replies.
  • Fixed the auto-update install flow so downloaded updates can restart and install from the app button or system notification.
  • The About page now shows update notes directly after checking for updates instead of only showing the version number.
  • Trimmed unwanted leading and trailing whitespace from AI replies for cleaner message output.

0.0.7

  • Added WhatsApp self-chat auto-reply support so messages sent to yourself can also trigger Trio Mate.
  • Simplified the macOS update artifact strategy around the ZIP installer path to reduce format-related update issues.
  • Added `data-testid` coverage for desktop interactive elements to support more stable E2E regression checks.

0.0.6

  • Improve: simplified the macOS release flow to ZIP-only for a more stable auto-update path.
  • Improve: reduced macOS artifact branching so CI maintenance and issue diagnosis are more direct.

0.0.5

  • Fix: restored the bundled updater configuration to avoid packaged update-check failures.
  • Improve: stabilized the packaged desktop update flow to support subsequent auto-update releases.

0.0.4

  • Added the full cron task workflow so AI scheduled tasks can be created, managed, executed, and reviewed in the app.
  • Added visual cron execution status and moved the related entry into the Activity Bar.
  • Added suggestion cards to the new-conversation empty state for a faster start.
  • Fixed out-of-order `blocks` rendering in AI responses.
  • Fixed accidental message sending when pressing Enter during IME composition.
  • Fixed Mermaid rendering failures with a graceful fallback and an AI repair entry point.

0.0.3

  • Improve: pointed macOS auto-update metadata to the ZIP installer so the update path matches the shipped artifact format.
  • Improve: adapted Electron download, signing, and Node environment handling for the new CI runner to improve release stability.

0.0.2

  • Added desktop auto-update support backed by the R2 update feed.
  • Added the About-page update entry with manual checking, download progress, and restart-to-install actions.
  • Added a native system notification when an update is ready.

0.0.1

  • Shipped the first desktop release with the core chat workspace and a three-column main layout.
  • Added multi-provider configuration with dynamic model discovery and OpenRouter support.
  • Added WhatsApp, Feishu, and 5G channel integrations with auto-connect and visible status on startup.
  • Added Skills and Agent package management, workspace selection, persistent memory, and related settings pages.
  • Added desktop workflow capabilities including Generative UI, file attachments and previews, log viewing, and knowledge-base settings.

That's the full release history.