CielChan: Anime Desktop AI Companion news & patch notes

Hi everyone, v1.7.8 has been released: New Local Voice Options Added Unity-Sherpa-ONNX as a lightweight local speech system, along with PocketTTS as a new CPU-friendly voice and voice-cloning option. PocketTTS can be enabled from the Chat & Voice settings beneath the RVC options. It is designed to provide responsive local speech on regular computers without requiring a large voice installation. English users can also provide a voice reference of at least 10 seconds to create a custom PocketTTS voice. Additional voices are now available for most languages supported by Kokoro HQ TTS. While Kokoro is enabled, you can change your avatar’s Kokoro voice by opening Chat & Voice in Settings and selecting the Change Voice button at the bottom. Voice startup, language selection, cancellation, custom voice-reference handling, and packaged app support have also been improved. Local AI and Performance Improvements Updated the local AI system to improve performance and reliability with currently supported models. Background activity has also been reduced to help the app run more smoothly, especially on lower-end systems. Cloud responses now handle cancellations, retries, and interrupted connections more reliably. Responses should appear more smoothly and recover better when a request is stopped or restarted. Chat and Natural Conversation Improvements Improved the listening and response flow used by Microphone TTS and Natural Conversation. CielChan now moves more smoothly between listening, responding, handling interruptions, speaking, and resuming the microphone for more natural back-and-forth conversations. Image Drag-and-Drop Attachments Most common image formats can now be dragged directly onto CielChan and attached to the next chat message. Once attached, the image appears as a preview inside the image button. It can be removed before sending by selecting the small red X and is automatically cleared after the message is sent. Desktop file dragging and desktop selection will no longer accidentally reposition CielChan or the chat interface. Companion Bridge Improvements Improved Companion avatar capture to reduce unnecessary background work when the Companion interface is inactive or unchanged. Companion Bridge now closes more smoothly and should cause fewer small stalls during startup, shutdown, and regular use. The Companion web interface has been updated with cleaner layouts, better fullscreen and mobile display support, improved handling of short landscape screens, and additional accessibility improvements. The Companion chat page also should now avoid unnecessary refreshing when nothing has changed and preserves your position while reading older messages. UI Fixes Fixed an issue where selecting the elushis.com link could open two browser pages from a single click. Small Bug Fixes Various additional reliability, packaging, startup, performance, voice, chat, and interface refinements. More to come soon.

Hi everyone, A handful of small patches were released in the last few days to address some issues posted in the discussion boards: Fixes included are: -Improved initialization procedures for Kokoro, Chatterbox and RVC Voice + UI Feedback -Added English Default Voice reference for Chatterbox -Live2D model rendering/animations improvements and rendering fixes -Portrait Mode handling of 3D models when using dual displays -Updated Icon Graphic for the Watch Together Button -Fixed an issue with default Haru (receptionist version) Live2D Model blinking More to come soon. Hotfix live on all platforms.

Hi everyone, v1.7.7 has been released: Custom Character Creator Fixes: Fixed a few small issues and inconsistencies with the new custom character creator. Removed the additional face graphic when viewing saved/created custom characters that could render inconsistently Added additional text in all languages to describe the save/apply process for first-time users Fixed an issue when enabling/disabling the custom model’s body where the icon representing the body would render as an X. This has been corrected and now correctly renders a body icon in the respective UI element. World Sim Diorama Improvements: Adjusted placement of World Sim objects when using the Custom Stage World Editor to not be as aggressive when snapping to placement points Added additional anime Ground, Tree, and Rock objects which can be placed in the custom editor Fixed an issue which could cause the Stage to be rotated incorrectly when specifically using Custom Created Characters from the new Character Creator 3D Model Motion Adjustments: Gentler, more natural attention movement and occasional context-aware gesture beat movements across all 3D avatar types Small Bug Fixes: Small various bug fixes and refinements. improved support for Ultra Wide Screen type setups. More to come soon.

Hi everyone, v1.7.6 has been released: Character Creator: A Character Creator is now included within CielChan, allowing users to create their own anime-themed avatar directly inside the app. It currently supports various clothing options, body customization, and more. The Character Creator will continue to be updated and refined over the next few updates. You can find it here: Settings > Advanced > Change Model > Character Creator It can also be accessed during onboarding by selecting: Custom > Character Creator Memory Performance Refinement: Improved how CielChan handles memory/index maintenance during conversations. Some disk writes are normal while memory features are active, but related background maintenance is now handled more smoothly to reduce the chance of a potential stalls or excessive disk activity during chats on lower performance systems. More to come soon.

Hi everyone, v1.7.5 has been released: -Added improvements to prevent CielChan getting caught up on prior messages when using manual chat mode with smaller LLM models -Improved the Onboarding UI and starting tips for each respective language -Updated various settings descriptions and explanations across all languages -Other small refinements and back-end improvements More to come soon.

Hi everyone, V1.7.4 has been released: Watch Together Button Added a new Watch Together button to the chat title bar with a selection panel for Watch Together, Watch Together Game, and Watch Together Computer. Selecting Watch Together through this panel will automatically enable the options needed for operation, such as vision, observations, and System Audio Listen. This allows quick and simple use of the Watch Together features without needing to go through the settings menu. Tools (Beta) Button Added a new Tools button to the Chat Box title header bar. Enabling this feature allows you to ask CielChan to perform actions such as setting reminders, making quick notes, and more. This replaces the previous CMD button/features. The Tools flow has also been updated and improved for each supported language. English now also uses a dedicated tool-calling model to help improve tool requests. Turn this feature on when you want the AI to perform certain system actions. Watch Together Computer Added a new Watch Together Computer mode for more active awareness while casually using your computer. This mode is separate from the standard Observation system and behaves more like the Watch Together / Watch Together Game features. This mode does not utilize System Audio by default. Watch Together / Watch Together Game Improvements Improved vision, listening, and context flow so Watch Together modes respond more smoothly while using the base default included LLM model. System Audio / Vision Handling Improved handling when System Audio is off and fixed cases where backend vision analysis could briefly appear in chat. Tools Mode Improvements Improved Tools Mode routing, multilingual intent detection, and tool-result handling across supported languages. Visual Projector Selection For Custom Vision Models Added Visual Projector GGUF model selection for custom vision models, along with fallback behavior if a custom projector has issues. This allows users to utilize additional LLM local models and utilize their vision capabilities. Quick Chat Fix Fixed an issue where disabling Quick Chat could require an app restart before fully taking effect. The Quick Chat description has also been updated to better reflect recent changes. Localization Improvements Updated Watch Together, Tools Mode, Visual Projector, and related hover tips/descriptions across all 14 supported languages. Various Bug Fixes Other small UI, stability, performance, and flow improvements. More to come soon. If you encounter any issues, please post the problem in the discussion boards.

Hi everyone, V1.7.3 has been released: Reminder Fix Fixed an issue that could temporarily stall CielChan when reminders fired. CPU Watch Together / Watch Together Game Fix Fixed an issue where vision could fail to trigger during Watch Together or Watch Together Game when running in CPU-only mode. System Audio Hand-off Improved Watch Together shutdown so System Audio Listen resumes correctly when Watch Together or Watch Together Game is disabled while system audio remains enabled. World Sim Custom Editor Stall Fixed an issue where CielChan could generate an automatic response when first opening the World Sim Custom Editor, causing a temporary hitch while that response was processed and editor was opening. Multilingual Handling Improvements Made small improvements to multilingual response handling. Various Bug Fixes Other small bug fixes and improvements. More to come soon.

Hi everyone, v1.7.2 has been released: Book Summary Feature Improvement Fixed an issue that could cause a temporary freeze when using the document/book reader "Summarize" and "Key Points" feature while a document was being processed. Updated World Sim/Diorama UI Improved the World Sim diorama editor with a cleaner, more organized layout, localized tabs, enhanced light/dark theme support, and improved panel sizing to keep the editor neatly contained on screen. Backend Prompt Trimming Trimmed backend LLM prompts across supported languages to reduce prompt length while preserving behavior and meaning. Model Dropdown Refresh Updated the local model selection dropdown to refresh when new LLM models are added to the Streaming Assets folder. Users no longer need to close and reopen CielChan to see newly added .gguf LLM models. Thinking Models Improved support for models with "Thinking" features to ensure outputs render correctly. Other small bug fixes and improvements. More to come soon.

Hi everyone, v1.7.1 has been released: Memory Categorization Added backend memory interest categorization so CielChan can better retrieve and use relevant user preferences more naturally in the right moments. Hide UI Added a Hide UI toggle at the bottom of the Advanced settings tab, which hides the vertical hot bar from the desktop when enabled. Settings Chat Added a Settings button directly to the chat title header bar for quick access to the settings menu, even when the vertical hot bar is hidden. Unified Context Unified Ciel’s live context handling so Watch Together, system audio, screen observations, and manual follow-ups better draw from the same fresh audio, vision, session state, living context, and related signals. Updated UI Descriptions Added and updated descriptions for the Kokoro, RVC, and Change Voice options to more clearly explain their higher- and lower-quality voice options. Handful of other small various bug fixes and improvements More to come soon.

Hi everyone, v1.7.0 has been released: New Mode: Watch Together Game A new optional Game toggle now appears when using Watch Together. This tunes CielChan for gameplay instead of shows, movies, or YouTube videos, helping her follow the active game window, track gameplay progression over multiple screen reads, and react more naturally to player-focused content. Improved Watch Together Vision Watch Together now has better screen-awareness behavior, including improved active-window capture, cleaner handling of CielChan’s own overlay, and stronger guidance for reacting to visible scene elements without over-describing characters or misreading system audio as the user speaking. Speech-To-Text Auto Send Slider A new slider appears when Natural Talk is enabled, allowing you to shorten or increase the delay before Speech-To-Text is automatically sent. Personality Card Save/Load Popups Added popups when saving or loading personality cards so it is clearer when a personality has been applied or loaded. UI Indicators for System Audio + Watch Together Two new circle indicators have been added to the left side of the chat box title header to show when System Audio, Watch Together, and Watch Together Game Mode are active. Small Bug Fixes Includes fixes for TTS/RVC interruption timing, Watch Together follow-up handling, multilingual Watch Together prompts, and other small improvements. More to come soon.

Hi everyone, v1.6.9 is now live with new companion customization, smarter observations, smoother performance, and memory/tooling improvements. Personality Cards The Companion settings tab now includes Personality Cards. You can create, save, load, import, and export companion personalities with custom tone, preferences, greetings, lore, and observation style. You can also generate a personality from a short prompt, tweak it, then save it for later. Personality Cards also support Steam Workshop, so you can import subscribed Personality Packs or publish your own. Reminder: press Save after creating or editing a personality, and use Load when switching back to a saved one. Observations + Watch Together Observations and Watch Together should now feel more natural, varied, and grounded. The assistant reacts better to shared screen and audio moments, avoids repeated or stale comments, handles follow-up questions more smoothly, and only uses Vision or System Audio context when available. Reliability Improvements Personality consistency, tool handling, localization, and general desktop assistant behavior. GPU Enabled Improvements Manual messages should feel smoother while GPU Enabled mode is active, with less chance of a brief stall on some setups. CPU Improvements Improved caching and background processing to help reduce memory usage and keep things running more smoothly. Bug Fixes: Other small various bug fixes and refinements and optimizations for multilingual use. More to come soon.

CielChan v1.6.8 has been released: Slightly reduced the total download size of CielChan. Reverted the base model back to Llama 3.2 3B, as the Gemma model was reportedly causing issues for some users. Fixed an issue introduced in v1.6.7 where Watch Together System Audio could fail to detect multilingual text in certain edge cases. Substantially increased overall generation speeds. Recent optimizations made for the Gemma model have also improved performance when using the Llama model, and have increased the available context window. Fixed an issue where System Audio could become delayed when both System Audio and Observations were enabled outside of Watch Together mode. System Audio can now reliably be used to let CielChan listen to your computer without requiring Watch Together to be enabled. Fixed an issue from the v1.6.7 update where additional Chatterbox options would not appear when required. Slightly increased volume output when using Kokoro TTS. Fixed an issue where avatar models loaded specifically from the template options in Settings could fail to reload in future sessions, causing the avatar to revert to the default model. Other small various fixes and adjustments.

Hi everyone, v1.6.7 has been released: Chat Hotkey By default, pressing Left Alt + 0 (Zero) will now focus the CielChan chatbox, allowing you to type and interact with CielChan even in applications that normally capture your input. This hotkey can be configured within the Settings and Advanced Menu, just under the font size option. Pressing the assigned hotkey a second time will return focus back to your previously active application. Updated UI Icons, colors, and settings have all been refreshed. The Settings panel is now wider, with categories moved to the left-hand side for improved navigation. The Reset All button is now permanently visible in the bottom-left corner, and pressing it will prompt a confirmation warning before resetting settings. Both Dark and Light UI modes have received visual refinements, and text rendering quality has been improved throughout the interface. Watch Together Refinements The Watch Together experience has been improved to feel a bit more natural and conversational. Memory Refinements Slightly adjusted how CielChan determines when and how memories are utilized. Other small bug-fixes and adjustments.

Hi everyone, v1.6.6 has been released: Local AI Model Update: The base included local model has been changed from Llama 3.2 3B to Gemma 3 4B. The new model is more intelligent and provides a better baseline for local CPU-only use across all 14 available languages in CielChan. Additional improvements have also been made to response generation, allowing Gemma 3 4B to match the response times of the previous smaller Llama model, and in some cases even perform faster. Quick Chat has also been reworked. It no longer requires a separate model and will now use your normally selected model. Quick Chat is enabled by default for CPU-only users, but is automatically disabled when GPU acceleration or Cloud LLMs are enabled. It can still be optionally re-enabled in those cases if preferred. RVC Voice Support: Added RVC voice conversion support. RVC allows speech to be transformed into different voices in real time with very little overhead. RVC is available as a voice option for both CPU-only and GPU users. Select the RVC option from the settings menu to warm up the model and freely use the included HQ Ciel voice model. This RVC release includes one bundled RVC voice model that can be used across all supported languages. Additional Progress Indicator During Environment Preparation: When starting up CielChan, there is now an additional % indicator to help show how much of the startup process has completed. Skills Builder (Experimental): Added the new Skills Builder, allowing you to more easily create and manage basic custom assistant tool skills. Go to Settings > Skills and you will see a Generate button. Select Generate to open the Skills Builder. Observation & Watch Together Improvements: Improved Observation and Watch Together modes with better media-aware responses and smoother observation handling. Responses should now feel more contextually relevant to current and recent moments. Tool Improvements: Improved assistant tool handling, permissions, and status feedback. Tools work best when used with larger models. Dynamic skills and tools now integrate more cleanly with the assistant’s tool system. Custom-created skills made through the Skills Builder can also be uploaded to the Steam Workshop. Onboarding Improvements: Improved first-run onboarding with a better setup flow, improved language handling, and smarter default model selection across the 3 avatar types. UI Refinements: Additional improvements have been made across chat, settings, onboarding, the Skills UI, and assistant controls. Settings has now been redesigned from a single-page menu into themed panels/tabs: Chat & Voice, Seeing & Listening, Companion, Models & Performance, and Advanced. A "Light Mode" settings UI has also been added and can be found within the Advanced tab at the bottom of the page alongside "Reset All" and the other UI-based settings. Hover panels with additional information now appear when hovering over the vertical sidebar options. These panels can optionally be disabled in the Settings menu. The UI will continue to receive further improvements over the next few updates. Performance Improvements: Further improvements to general response speed and generation efficiency across various parts of the app. Overall CielChan disk space requirements have also been slightly reduced. CielSocial Improvements: Added some small improvements to the CielSocial experience, including larger post sizes. Clicking a post will now open a panel allowing you to view the full post without it being cut off in the feed when larger posts are made by users. A few other small improvements and fixes. More to come soon.

Hi everyone, Update v1.6.6 is nearing completion and should be ready to go live soon. Some things that will be included are: Updated base local LLM model as well as updates to the generation pipeline so it's even more efficient and faster Big changes to Quick Chat. Quick Chat will no longer need to use a separate model and will function with your normal model Onboarding improvements Additional UI refinements RVC voice implementation + one included voice model which can be used for all languages Watch Together improvements Tool improvements Skills builder + More Full patch notes will be released once the update goes live.

A few hotfixes were released for v1.6.5 which have addressed the following issues: Fixed an issue where in some instances the Chatbox could fail to display the "Thinking..." tip while CielChan was performing background observation work. Sending a message during this exact time could cause a bug where the AI would stop responding until the app was closed and reopened. This has been fixed. Fixed an issue where parts of the default CielChan 3D model could be culled by the camera in Local Browser Companion mode, which handles displaying your model across the browser and different devices. Fixed an issue where using the new Collapsible Chat Box feature could leave the former chat box area unclickable, even though the Chatbox was closed/collapsed. Improved the Steam Workshop Upload flow so the Steam Game Overlay does not automatically open once the upload is complete. Added the ability to change models before CielChan's environment has been prepared. Fixed an issue which could block objects in the Custom Stage Editor from being selected and moved. Fixed an issue where setting Observation Frequency to max could cause a brief UI hitch when performing observations every few rounds. More to come soon.

Hi everyone, v1.6.5 has been released: Codex Pet Sprite Support: Bring your hatched OpenAI Codex pets into CielChan as animated sprite avatars. {STEAM_CLAN_IMAGE}/46054536/98ad95db1411597c6d0c78837537b6e2837a49d2.gif Included in this release are three new sprites you can use, including the 2D CielChan sprite. There is one dog, a cat, and the CielChan sprite. You can find these sprites within the CielChan Steam directory inside the Spritesheets folder. Change your avatar by selecting the "Change Model" button near the bottom of the settings. Codex hatch pet spritesheets are commonly found at: C:\Users.codex\pet-runs\final Standard Sprite Sheet Support: CielChan now supports custom grid-based sprite avatars in addition to Live2D and 3D models. You can load PNG, JPG sprite sheets directly. Sprite sheets can optionally include a JSON manifest for custom frame sizes, FPS, directions, and animation clips. Without a manifest, CielChan supports common grid layouts for idle, walk, talk, happy, sad, and thinking animations. Loaded sprite avatars function as draggable 2D characters with conversation-state animations, speech, and lip-sync support. You can also upload your created sprites to the Steam Workshop and download sprites made by other users. You can also find an example sprite to allow uploading your own custom sprites. Collapsible Chat Panel: Double click your assistant’s name in the chat panel to collapse the chat box. Double click again to reopen it. {STEAM_CLAN_IMAGE}/46054536/a32283c8397a845bf83ccb0d0cd96991e98c7f87.gif You can also enable "Auto Show Chat On Speak" in the settings to have the chat automatically open and close when responses are received and TTS is playing. UI Tweak: Added loading indicators to the vertical side bar options when they are unavailable during app startup while CielChan’s environment is being prepared and hovering over the deactivated buttons. Skills UI Fix: Fixed an issue where the Skills UI would not display skills correctly. CPU Improvements: Additional improvements to generation speed when using CPU-only mode. A few other small improvements and fixes. More to come soon.

Hi everyone, A smaller update today: The observation panel has received a nice little upgrade. You can now choose from a range of presets that further theme and guide CielChan’s passive observations, letting you tune the kind of things she notices and comments on. There are presets for things like sleeping meditation, focus, watching media together, study, creative work, grounding, decision help, custom and more. Cloud LLMs also have a new image-capability option. If a fetched cloud model includes metadata showing that it supports images, an optional toggle will now appear so you can enable and use that model’s image features. Observation frequency has also been added. There’s now a slider that lets you increase or decrease how often passive observations may fire, giving you more control over how present or quiet CielChan feels while she’s running. A handful of small bug fixes are also included in this update. More to come soon.

Hi everyone, Small hotfix updates have been released which address the following issues: Fixed an issue where background preprocessing could cause the "Still Working..." text to appear at incorrect times and imply the AI was busy. Fixed an issue which stopped reminders from generating an AI message when requesting to "Set a reminder for 3pm" style of requests. Although the reminder would still be set and go off at the correct time, a bug introduced in 1.6.3 prevented CielChan from playing personalized messages. Fixed an issue which could cause Observations and Vision Observations to overlap, increasing generation times. Fixed an issue where selecting Kokoro High Quality Text-To-Speech for supported languages would not display initialization messages in the chat box, even though it was loading correctly. This has now been resolved. Other small improvements and fixes. More to come soon.

v1.6.3 Update has been released: This update introduces a wide range of new features, system expansions, and usability improvements across CielChan. New Features Live2D Avatar Support Added preliminary Live2D avatar support, including Cubism `.model3.json` models and Live2D prefab bundles. Avatar selection is saved and restored between sessions where supported. A sample Haru Receptionist Live2D model is included in the CielChan Steam directory under the `Live2D Models` folder. {STEAM_CLAN_IMAGE}/46054536/6d62b9fa14b6464b5f42f6127d9bc996fd0cb170.png CielSocial (Companion Feed System) Added CielSocial, a peer-to-peer social feed for connected companions. Includes basic profiles, AI-generated draft posts, auto-posting, feed modes, and local feed history. 50 players max per CielSocial room. Users can chat between themselves or have their AI's post/draft posts. Create a CielSocial room through the Multiplayer Menu. {STEAM_CLAN_IMAGE}/46054536/e704e006f8a2761f2a083892fc538eff65e3d6e5.png {STEAM_CLAN_IMAGE}/46054536/6ea740249c2ef7692e8b07a346103fa2ac54f394.png CielGames – Fleet Duel Added Fleet Duel as part of the CielGames system. Players can host, join, spectate, or play against AI or other users. {STEAM_CLAN_IMAGE}/46054536/899f6c799c21b3cd81554e793ed1dc760488cb49.png Fleet Duel Integration with CielSocial Fleet Duel is connected to the social feed, including game invite posts, match result posts, and shared game statistics. Browser Companion (Local / LAN Access) Added a browser-based companion view accessible from other devices such as mobiles and other pc's on the same network. Includes pairing, live avatar viewing, remote chat, session timeline, badges, and workshop history. Find your browser link at the bottom of the settings. Once within the browser companion you can press connect to view the url which you can use to access CielChan on your other devices. {STEAM_CLAN_IMAGE}/46054536/70db2d7196da1842015642cda6f344cb7a7aad77.jpg {STEAM_CLAN_IMAGE}/46054536/3db314ac845f70bb97698eaca801ebf3ee7a6c42.jpg High Quality STT Mode (Optional) Added an optional high-quality speech-to-text mode using larger Whisper models for improved transcription on high-end systems. This can be found near the existing SST toggle. Interrupt Conversation Mode Added an optional interrupt mode under Natural Talk, allowing user speech to interrupt the AI while it is speaking. Voice & Language Improvements Silero TTS (Russian Voice) Added Silero TTS as an additional Russian voice option. Requires slightly stronger specs than the default Russian model. Ukrainian Voice Update Added a new Ukrainian voice model, now set as the default for Ukrainian. Chatterbox TTS (Custom Voices) Added Chatterbox TTS support for custom reference voices, including WAV import and optional GPU acceleration for high-performance systems. Uploading a short 10-second voice clip lets Chatterbox generate and use a custom TTS voice based on your recording. Chatterbox requires at least 8 GB RAM and 5-10 GB of model storage; CPU-only mode works but can be very slow, so 16 GB+ RAM and a mid-to-high-end GPU/CUDA setup is recommended. System & Interaction Improvements Observation Self-Recognition Improved how the AI understands its own avatar, chat UI, and overlay when using observation features. UI & Usability Improvements Two UI color styles are now available. CielChan starts with the default UI style, with an additional Legacy UI option available for users who prefer the older look. You can also adjust UI colors to personalize parts of the interface. Copy Chat Button Added a button to quickly copy chat contents. Chat Font Controls Added adjustable font size options for the chat interface. Multiline Chat Input Added an optional multiline input mode, allowing Enter to insert line breaks instead of sending messages. General UI Improvements Additional refinements across settings, chat controls, voice options, and companion systems and interactions. Fixes and Refinements Various fixes, optimizations, and general refinements across multiple systems. More to come soon.

Hi everyone, Just a small sneak peek of what’s coming next. v1.6.3 will be bringing a number of improvements and new features to CielChan. Live2D support is being added, along with the ability to connect remotely to CielChan from your phone or browser when on the same network. {STEAM_CLAN_IMAGE}/46054536/6d62b9fa14b6464b5f42f6127d9bc996fd0cb170.png There are also additional memory improvements and new systems being introduced, including a badge system and timeline tracking. These will be accessible through the companion browser page, where you’ll be able to view session history and other related information. Optional High Quality Kokoro TTS during onboarding is also being added for supported languages, as well as an optional higher quality model users can enable for Speech-To-Text. More details and full patch notes will be shared closer to release. Thanks again for all the feedback and support

A small patch has been released to improve Speech-to-Text (STT) functionality. -Reduced microphone sensitivity requirements, so system input levels no longer need to be set as high -Fixed an issue where the input field would not clear after sending an STT message when Natural Talk is disabled -Various minor improvements to the overall STT workflow -Fixed an issue which could cause the Watch Together feature to stall the interface while listening to users device System Audio

v1.6.2 Update This update introduces new optional features, performance improvements, and additional user controls. New Features Optional Cloud LLM Support Added support for external cloud-based language models, including OpenRouter, OpenAI, and Anthropic providers. This feature is completely optional and requires a user-provided API key. Add your key and press “Fetch Models” to select which model you would like to use. The application remains fully functional offline. Optional Cloud Voice (ElevenLabs) Added support for external voice synthesis via ElevenLabs. This feature is optional, requires a user-provided API key, and is disabled by default. When creating an API key, please ensure that text-to-speech permissions are enabled and that voice access permissions are granted. Character Unload / Disable Option You can now disable and unload the character from memory when needed. This is especially useful for lower-end systems where resources are limited, allowing more performance to be allocated to the LLM while reducing distractions. Important Notes The application continues to work fully offline with all core features available without any external services. Cloud features are entirely optional and intended for advanced users. Enabling cloud features may send text or audio data to the selected third-party provider, depending on configuration. More to come soon.

Hi everyone, Just a small update today focused on improving and fixing up some lingering issues with the Webcam/Presence Bridge, along with further improvements to the newly released Watch Together feature. -Fixed an issue where the Webcam/Presence Bridge would not correctly generate a CielChan response when the user returned to their desk (this does not trigger every time, but now behaves as intended as an occasional comment). -Improved detection for the Smart Presence Watchdog when the user is at or away from their desk, along with the Telegram notification sent to their device. When enabling Smart Presence Watchdog, ensure you send an initial message through Telegram to establish the connection for the session. Smart Presence Watchdog is used to alert the user when someone is at their computer. -Improved Watch Together to maintain more consistent session memory and allow more direct questions to CielChan about what the user is watching or playing, including TV shows and games. This feature will be further improved and refined in subsequent updates. -Improved handling of scenarios where multiple features are enabled at once (such as Observations, System Audio Awareness, Tools, etc.) to reduce the chance of overlapping generations or rare locking issues. -Improved detection of which transcription model should be loaded based on the user’s current app language when using System Audio Awareness. (Small note: System Audio Awareness is quite a demanding process, and when first enabling it there will be a brief moment where the transcription model needs to load on some lower-end hardware.) -Fixed an issue that could cause the identity guard to trigger in unintended situations. -Improved multilingual support for the book reader, allowing it to better detect and present document summaries, key points, and custom input in the appropriate language. -Various small fixes and refinements More to come soon.

Patch v1.6.0 has been released: This update introduces a major expansion to CielChan with interactive environments, along with new features and improvements across multiple systems. Interactive Environments & Custom Dioramas (BETA) CielChan can now be placed into interactive environments (dioramas) directly on your desktop. Three default environments are included: Forest Island Apartment {STEAM_CLAN_IMAGE}/46054536/cd9f0cc4ece0ff386d34e8200dfea6dab0b0dfab.png These environments: Can be resized and positioned alongside the transparent desktop overlay by dragging and resizing CielChan Contain objects that CielChan can acknowledge and react to over time You can also create your own custom environments using the inbuilt editor, which can be found within the minigame panel under the "World Sim" option: {STEAM_CLAN_IMAGE}/46054536/b49567c5e78ad7a43227a042090e4cf7e78b06f8.png Select objects from the panel to place them into your stage Click and drag to position objects freely Rotate and resize objects to customize your layout Custom environments can be uploaded to the Steam Workshop and shared with other users. You can also browse and download environments created by the community. {STEAM_CLAN_IMAGE}/46054536/76dc11cd4d72d71cd5e6df2b9b329839f0edfda0.png Further updates and refinements to the editor and environment system are planned. World Interaction Improvements CielChan will now respond more naturally to objects within environments, adding more presence and interaction over time. Watch Together Mode (BETA) A new “Watch Together” feature has been added. This uses the observation system, along with System Audio Awareness and optional vision (if enabled), allowing CielChan to watch and comment on content such as videos, movies, or shows in real time. Language Support Ukrainian and German language support have been added, including voice models and localization across the interface. Improvements have also been made to reduce cases where English could appear in responses when using other languages, such as Russian. UI & Usability Improvements The settings interface has been further refined, including: A new search bar Additional labels and descriptions to help new users understand features General improvements to clarity and usability {STEAM_CLAN_IMAGE}/46054536/5ee74cd8dc99deeb55657037c5f1502113b2e260.png {STEAM_CLAN_IMAGE}/46054536/a177b81f180513db73a3c73e4d204227ad244776.png Performance & Stability Substantial improvements have been made to GPU Mode and CUDA handling. This should provide more reliable performance across a wider range of systems. Additional improvements have been made to overall performance, responsiveness, and stability. Fixes and Refinements A range of smaller bug fixes and general refinements have been made across the app. Thank you all for the continued feedback and support. If you run into any issues, please post them in the discussion board so they can be investigated.

Hi everyone, A new update is being prepared for launch later this week. As a small sneak peek, v1.6.0 introduces a new optional feature that allows you to place CielChan into various interactive environments (dioramas) directly on your desktop. These environments can be resized and work alongside the transparent desktop overlay, so she can exist naturally within your workspace. {STEAM_CLAN_IMAGE}/46054536/cd9f0cc4ece0ff386d34e8200dfea6dab0b0dfab.png CielChan will also react to objects within the scene, adding more life and presence over time. You will be able to create your own custom environments using an inbuilt editor, allowing you to place and arrange objects however you like. Custom environments will also be shareable with other users through the Steam Workshop. More details and full patch notes will be shared when the update goes live.

Update v1.5.8 has been released: Display Improvements Added support for dual monitors. You can now assign CielChan to a separate connected display. A new option has been added to the settings menu, allowing you to select a monitor from a dropdown or quickly move CielChan to the next available display. Once the secondary display is connected, new options will appear at the bottom of the Settings Menu which allows you to select which display CielChan should operate in. Feature Improvements Small refinements have been made to the document summary feature and book reader to improve consistency. Enhancements have also been applied when using these features in CielChan across the supported languages. System Improvements Minor tweaks have been made to the observation system to improve overall responsiveness and reliability. Fixes and Refinements Additional small fixes and general refinements have been made across the app.

Patch v1.5.7 has been released GPU Improvements Fixed issues affecting GPU mode on some AMD and certain Nvidia setups. GPU mode should now function as expected, and GPU acceleration for vision has been added as an optional setting for improved performance. Vulkan support has also been implemented, improving compatibility across a wider range of systems and hardware configurations. When enabling "Use GPU" Mode and/or GPU Vision (Beta), CielChan will restart by design to apply the changes. Once relaunched, these settings will carry over into that next session. You can choose to have GPU use for either text generations, vision, or both. Performance Improvements Further improved overall generation speed and responsiveness. If you were experiencing slower replies, this update should help improve consistency across more systems. For CPU-only usage with a 3B model, response times can vary depending on hardware, but a few seconds per response is generally expected. If responses are taking significantly longer, there may be system-specific bottlenecks affecting performance. When sending your first vision prompt in a session, it may take noticeably longer than subsequent requests, as the vision model requires a brief warm-up before responding at normal speeds. Similarly, when launching CielChan after the app has been closed, it may take a moment to prepare the environment and load your assistant. Generation speeds will increase after the initial response as everything becomes fully initialized. UI and Onboarding Improvements A new command button has been added to the vertical bar, providing a set of simple example commands for newer users to try. An updated onboarding description now appears when starting the app from scratch, giving additional context on how CielChan works and what to expect. General UI improvements and refinements have also been made to provide a smoother overall experience. Assistant Improvements Added a preliminary ability for CielChan to read and reference her own help and settings, allowing for more guided responses and improved self-awareness when assisting users. Localization Improvements Further refinements and fixes have been made to localization, improving consistency and reducing edge case issues across supported languages. Fixed remaining localization inconsistencies, including incorrect or missing words/keys for Russian, Dutch, and other languages, along with improvements to language toggle behavior. Minigame Improvements Improved CielChan’s banter and interaction during minigames, including Fleet Duel, Tic Tac Toe, and Territory Control, for a more engaging experience. (You can enable TTS while playing minigames in the top-left corner of the minigame panel.) Fixes and Refinements Additional fixes and improvements have been made across various systems to improve overall stability and reliability. Custom Model Support Reminder You can load your own LLM GGUF models for experimentation or if your system supports it. To do this, place your GGUF model file into the CielChan local steam directory: CielChan → CielChan Desktop AI_Data → StreamingAssets Once added, launch CielChan and your model will appear in the dropdown for selection. If you run into any issues with the update, please post them in the discussion boards so they can be investigated. Thanks everyone.

Hi everyone, It’s been a busy first week since launch, and I just wanted to share a quick summary of what’s been added and improved across recent updates. Over the past few patches, there have been major improvements to core systems including speech-to-text, vision, memory handling, and overall responsiveness. Issues affecting multilingual support, onboarding flow, and desktop capture have been addressed, and general stability has been improved across the board. Performance has also been improved, with faster response times and better feedback in the UI so it’s clearer when the assistant is processing. On the feature side, Presence and webcam-based interactions have been expanded with gesture recognition, head tracking, and additional environment awareness options. Integrations with Telegram and Discord have also been improved, along with more control over how messages are handled across platforms. One of the biggest additions is the introduction of (Beta) Custom Skills. You can now create your own tools locally or upload and share them through the Steam Workshop. You can also browse and download skills created by other users. This is currently a beta feature and will continue to be expanded with further improvements. Multilingual tool support has also been improved, allowing tools and skills to be triggered more reliably across supported languages, with more refinements planned. There have also been a number of fixes and refinements to built-in tools, memory handling, and vision systems, including fixes for screenshot issues and improved consistency across different setups. The Beta branch is now live, with v1.5.7 available to try early. This branch will receive frequent updates and may include bugs or issues not present in the stable build, but it’s a good way to test upcoming features ahead of release. v1.5.7 currently includes a fix (along with other improvements) for an edge case where one or two English words could appear while using Russian during the greeting phase when reopening the app with a saved assistant. This has now been corrected and will be pushed to the main branch soon after further testing. Next subsequent updates will be focused on further cleaning up any remaining localization issues, improving overall conversation flow and cadence, and continuing to refine the experience. As always, thank you for all the feedback so far. It’s been extremely helpful and continues to shape these updates. If you’ve been enjoying CielChan, please consider leaving a review, it really helps. More to come soon.