- v1.3.034 - 9 September 2026
v1.3.034 — 2026-09-09 Improvements The chat window now shows memory-processing progress and message status. It explains when a message is waiting for memory processing and displays progress, queue status, model processing and completion or failure. Improved chat-history loading and maintenance. Large histories load more efficiently, with more reliable data recovery after upgrades, restarts and unexpected interruptions. Improved localized character-card error messages. Card-change and pending-deletion messages now follow the interface language across all 31 supported languages, with wording corrections in Romanian and Thai. Fixes Fixed chat becoming unavailable after some upgrades from older versions. History merge conflicts and storage initialization failures are now handled without discarding existing records, allowing new messages to be saved and read normally. Fixed individual unread…
- v1.3.033 - 5 September 2026
v1.3.033 — 2026-09-05 Fixes Fixed local AI failing to load on some otherwise compatible devices. Loading no longer stops unnecessarily when system information checks fail or outdated startup records remain. Fixed local AI loading being blocked when VRAM usage could not be estimated. The app can now attempt to load the model when an estimate is unavailable, while retaining the existing checks for insufficient VRAM. Fixed local AI reload remaining blocked after a stop timed out. Once the previous model process has exited, loading can be retried without restarting the app. Fixed misleading local AI failure messages. Model and component errors are no longer incorrectly presented as GPU or VRAM failures, and reopening Settings preserves the latest failure details. Improvements Improved logging when local AI fails to load. The original failure reasons are retained more reliably to help diagno…
- v1.3.032 — 2026-09-04 - 4 September 2026
v1.3.032 — 2026-09-04 Fixes Fixed the app sometimes remaining stuck on the startup screen after a pet model failed to load. When a model fails or the renderer exits unexpectedly, the app now recovers automatically and advances to an available fallback pet instead of repeatedly loading the same broken model. Fixed pet models that stopped responding during loading causing an indefinite wait. A stalled load now ends cleanly so recovery can continue. Fixed model switching sometimes failing while the pet was recovering. Recovery and user-requested switches now run in the correct order, preventing a successfully loaded model from leaving the main window hidden. Fixed Settings being unable to select another model after every automatic fallback failed. A valid model selected in Settings can now restore the pet, while a model that does not load is not saved.
- v1.3.031 - 3 September 2026
v1.3.031 — 2026-09-03 Improvements Improved local AI background process lifecycle management. Model loading, switching, stopping and app shutdown now transition more reliably. Improved logging when local AI fails to load. Failure reasons are now recorded more completely and clearly, making model-loading problems easier to diagnose. Fixes Fixed an issue that prevented the built-in local AI model from loading on some devices. Previously, loading could fail repeatedly and incorrectly report that the GPU was unavailable even when the device met the requirements. Fixed local AI background processes sometimes remaining active. Related processes now terminate correctly after a loading failure, model switch or app shutdown instead of continuing to use VRAM and system resources. Fixed incorrect application icons. The main executable, installer and created shortcuts now display the official appli…
- v1.3.030 - 2 September 2026
v1.3.030 — 2026-09-02 New Character cards can now be imported and exported as structured JSON. Standard V1, V2 and V3 PNG or JSON cards are accepted. Export automatically creates a V3 PNG when the card has a usable portrait, or JSON when it does not; cards with multiple linked world books let you choose which one to embed. World books can now be imported and exported as widely compatible JSON. Compatible world-book data embedded in PNG files can also be imported. Improvements Editable character cards now use structured V3 JSON as their canonical data. Existing Markdown cards receive a matching JSON copy while the original file remains untouched for rollback to an older app version. This removes Markdown-heading ambiguity without discarding existing data. Character-card and world-book round trips preserve data the app does not edit. Standard fields, extension data, asset declarations, em…
- v1.3.029 - 29 August 2026
v1.3.029 — 2026-08-29 Fixes Live2D window size and placement are now based on the model's actual rendered mesh. Models whose artwork extends beyond their declared canvas (some Workshop and custom models) were clipped at the edges or sat off-center; they now render fully and correctly positioned. Custom Live2D models no longer jump position on first load. Once the measured size is known, the window adjusts in place, keeping the model's bottom-center anchor fixed. Fixed rendering errors when the detail preview and the control-panel preview coexist. The two WebGL canvases now hand off Cubism shader ownership correctly — no more garbled or blank previews. The control panel no longer saves a stale thumbnail. A leftover frame from the previous model could be stored as the new model's thumbnail while the panel was open; this no longer happens.
- v1.3.028 - 27 August 2026
v1.3.028 — 2026-08-28 Fixes Memory Manager character list fixes. The chat-history tab now lists only characters with messages, and the memories tab lists every character with memories — characters whose chat history expired no longer lead to an empty chat view. Unsubscribed Workshop characters no longer appear in the list, avatars now load for all listed characters, and a duplicated emoji in the panel title is removed.
- v1.3.027 - 27 August 2026
v1.3.027 — 2026-08-27 Improvements Chat history is no longer deleted after 30 days. Messages are now kept long-term in monthly-sharded storage, capped at the most recent 20,000 messages per character. The new layout also makes Steam Cloud sync upload only the current month's file, so syncs are faster and playing on multiple PCs conflicts less. Fixes Characters with memories but no chat history now appear in the Memory Manager list. Previously their memories existed but had no way to be viewed. Interrupted Workshop uploads no longer leave staging folders behind. Leftover folders older than 24 hours are removed at startup — previously each could keep a full content-package copy on disk indefinitely. Crash dump files are now capped. The app keeps the 5 most recent crash dump files.
- v1.3.026 - 27 August 2026
v1.3.026 — 2026-08-26 New Choose which GPU runs the local AI. Multi-GPU PCs (dual NVIDIA cards, or a dedicated GPU alongside integrated graphics) can now pick the adapter in Settings → AI. The choice also applies to GPU voice synthesis, and recommendations, VRAM budgeting and the resource monitor all follow the selected card. "Auto (recommended)" keeps the previous behavior. Fixes Right-click menu icons and labels are no longer cramped together. A small gap now separates the emoji icon from the menu text.
- v1.3.025 — 2026-08-23 - 24 August 2026
v1.3.025 — 2026-08-23 Fixes The pet now recovers by itself when the GPU drops the model's rendering context. A context loss (for example while a heavy game was running) previously left a white square with no way back; the app now reloads the pet with a bounded retry budget — no more permanently dead pet. Model switching can no longer jam forever. Switch requests queued during a recovery now time out cleanly instead of leaving the Settings page unresponsive to any further switching. VRM models no longer get clipped at the bottom during lying/sitting animations. The camera now compensates for each animation's actual envelope, so poses like lying down stay fully in frame. Right-click menu text and icons are now properly aligned. Menu items get a fixed-width icon slot, so labels line up in a straight column. The Send button no longer overflows the mini chat panel. Long translations (Russian…
- v1.3.024 — 2026-08-22 - 23 August 2026
v1.3.024 — 2026-08-22 Improvements Faster VRM model switching. Shader compilation now happens during background preloading instead of blocking the reveal, and the renderer is created directly with your saved quality preset — no more long stall when switching VRM models. Workshop model cards now show a single category badge. VRM and MMD cards no longer display a duplicate badge in the top-right corner — each card carries exactly one label. Fixes Switching pet categories no longer flashes models in a wrong position. The old model stays put and the new model is revealed only after its window position, size and camera are all settled — no more misplaced frames in either direction. Models keep their screen position when switching categories. The pet's position is now anchored to the model's feet rather than the window corner, so switching categories with different window shapes no longer mak…
- v1.3.023 — 2026-08-22 - 22 August 2026
v1.3.023 — 2026-08-22 Fixes Dragging the pet no longer makes attached UI sink on high-scaling displays. On 125%/150% scaled screens, the drag loop's position updates could slowly grow the window frame, pushing the voice indicator and other attached UI further below the model after every drag. The drag now pins the window's content size, so nothing accumulates. Models with missing files can now be deleted. Removing a model whose files are no longer on this computer (for example, an entry synced from another PC) now removes the library entry as expected instead of failing — while still refusing to delete files that exist but can't be verified. Shimeji and Static tabs recover from unavailable saved models. Like the other pet types, switching to these tabs now falls back to the built-in default instead of failing when the previously used model is missing. Pet categories remember your chosen…
- v1.3.022 — 2026-08-20 - 22 August 2026
v1.3.022 — 2026-08-20 Fixes Model switching no longer dead-ends on machines synced from another PC. When a model or mode points to a path that does not exist on this computer (for example synced settings from a different machine), the switch now falls back to that mode's built-in model instead of failing outright and leaving only one usable pet. Attached UI no longer drifts away from the pet after drags. Roam camera compensation now only applies to displacement produced by the roam system itself, never to your own drags, so the voice indicator and other attached UI stay put relative to the model.
- v1.3.021 — 2026-08-20 - 20 August 2026
v1.3.021 — 2026-08-20 Fixes On-screen input UI no longer runs off-screen when the pet is moved. The voice indicator, chat bubble, danmaku, and input panel now stay attached to the pet and clamped inside the visible area while dragging, roaming, and dancing; Shimeji dragging now uses the same reliable window-drag path as the other pets. Gemini works again in streaming mode (the default). Errors returned inside the Gemini stream lost their status code, so the no-search fallback retry never ran for streaming users — free-tier keys saw a permanent "Quota exceeded" error. The status is now preserved and the retry works in both streaming and non-streaming modes.
- v1.3.020 — 2026-08-19 - 19 August 2026
v1.3.020 — 2026-08-19 New New 25 FPS frame rate cap option. The global frame rate cap in Settings now offers a 25 FPS step alongside Match Refresh Rate, 60 FPS and 30 FPS, and it takes effect across every rendering mode (VRM, Live2D, Spine, MMD and Shimeji). Improvements The frame rate cap is now managed in one place. The Settings tab is the single source of truth for the frame rate cap — the VRM pet control panel no longer carries its own separate control, so the two settings can no longer disagree.
- v1.3.019 — 2026-08-19 - 19 August 2026
v1.3.019 — 2026-08-19 New New "Paragraph Synthesis" TTS option (off by default). When enabled, replies are synthesized paragraph by paragraph (split at line breaks) instead of sentence by sentence — each paragraph starts playing as soon as it is complete, giving more natural prosody on both built-in and cloud TTS engines. Over-long paragraphs automatically fall back to sentence-level synthesis. Improvements Clearer voice-output settings. The "Dialogue Only" read-aloud mode is renamed to "Quoted Speech Only" in all 31 languages — characters whose replies contain no quotation marks no longer fall silent without explanation. The descriptions for Read Aloud, Bubble Display, Skip Action Descriptions, and Paragraph Synthesis are now always visible below each option instead of hidden in hover tooltips. Fixes Local AI chat no longer fails with a cryptic "HTTP 400" error when the conversation ou…
- v1.3.018 — 2026-08-18 - 18 August 2026
v1.3.018 — 2026-08-18 Improvements Spine version guidance now matches the actual runtime range. Notifications correctly identify Spine 1.6 through 4.3 as supported and ask newer models to be re-exported as 4.3 or earlier. Fixes Workshop companions and their conversation data are now preserved across upgrades and subscription reconciliation. The app no longer treats a subscribed character as a disposable cache, restores character cards left in an interrupted 1.3.017 deletion state, rebuilds stable Workshop character identities from legacy metadata, and creates a byte-exact recovery evidence snapshot for affected installations before opening their live data. Pet selections synchronized from another PC can no longer prevent the app from starting. Device-local model paths are safely rebound when the same asset exists on this computer; when it does not, the app opens with a temporary built-i…
- v1.3.017 — 2026-08-17 - 17 August 2026
v1.3.017 — 2026-08-17 New Push-to-talk voice input is here, in three places. A new voice button next to the chat input starts recording with a live waveform and timer — confirm to transcribe and send, or cancel anytime. A global hotkey (F2 by default, customizable) works in either hold-to-talk or press-to-toggle mode. And the voice indicator on the pet itself is now clickable — once to start, again to send. New "Music" tab — a full local music library and player. Browse by All Songs / Artists / Albums / Liked / Recently Played / Most Played / Recently Added, with instant search and sorting; title, artist, album, cover art, genre and duration are read automatically. Create normal or smart playlists, drag tracks, albums and artists into them, manage and save the play queue, view synced .lrc lyrics and a live spectrum, and keep folders synchronized automatically. Loudness normalization, cr…
- v1.3.016 — 2026-07-16 - 16 July 2026
## v1.3.016 — 2026-07-16 - fix: **Selecting a cloud voice by ID now applies to spoken replies as well as to the voice test.** Previously a voice entered by ID (for example a Fish Audio reference ID) could be used by the "Test voice" button while replies were still spoken with the previously selected voice, and the two could stay out of step across restarts. The active cloud voice, the selected provider and the voice test now always resolve to the same saved selection. - improvement: **The voice-input indicator now clears immediately when speech input is stopped or the input mode is changed** while a phrase is still being processed, instead of briefly remaining on "Processing".
- v1.3.015 — 2026-07-14 - 14 July 2026
## v1.3.015 — 2026-07-14 - fix: **Some of the built-in local voices could fail to speak, leaving a sentence silent.** The affected voices now synthesize audio correctly. - fix: **Long messages sent to the built-in AI model are now processed in full.** Previously only the earlier part of a longer message reached the model, so its reply could miss details or instructions written further down. - fix: **The Microsoft Edge online voice now plays correctly, and importing video, audio and certain image files works again.**
