Arc, Visual References & Chat Controls Two new ways to shape a scene — and everything you need is now one tap away. PersonaChat Arc — a model that follows the scene • Arc is a new choice in the model list. Everyday replies stay quick and light. • When a scene reaches a difficult character moment or a real turning point, Arc can hand that message to a model better suited to it, then settle back down again. • There is nothing to set up and nothing to tune. Arc reads the moment for you. • Choose Arc when you begin a story and it stays with that story. A conversation already under way can't be moved onto it, so start a new one to try it. Arc works in one-to-one chats on the built-in models. Visual References — show them something • Attach a photo from your camera or your library, and the character answers what is actually in it — a room, an outfit, a place, a look on someone's face. • GIFs work too. The first frame is used. • The picture is read and described so the character can respond to it inside the story. • Leave a couple of ordinary exchanges between pictures, so a scene stays a scene rather than a slideshow. Chat Controls — everything in one place • A new controls panel inside the chat holds your AI model, your Lorebook, your Living memories, and the last summary. • Last summary is on the start screen as well, so you can read the rolling summary of earlier messages before picking up where you left off. • No more leaving the conversation to change one thing and then hunting for your place again. Models • Qwen 3.5 Flash has been replaced by Qwen 3.7 Flash. • Mistral NeMo has been retired. • Sampler defaults are retuned across the built-in models, and Top K now starts at 0 — set it to 0 to consider every token, or 1 to keep only the most likely one. It's hidden for the models that don't use it. Also • Ads now come round less often — every 5 messages in one-to-one chats instead of 4, and every 8 in group chats instead of 7. Premium removes them altogether. • A new app icon. It no longer changes with your theme — one icon, all the time. • Lore drafts now behave consistently when you're using your own OpenRouter key. Thanks for the reviews and feedback!

Memory & Lore Long stories used to lose the small things. Now they don't. Living Memory — remembered for you • Characters now hold on to a short list of precise details from much earlier in the story: promises, secrets, relationships, turning points, and threads left hanging. • Each one is brought back only when it's actually relevant to what you just said, so nothing crowds out the conversation. • These sit alongside the rolling summary instead of replacing it. The summary carries the broad plot; memories carry the exact details a summary eventually smooths over. • Automatic memories are read-only. When you want to take control of one, tap Save to Lorebook. Lorebook — you control these • Write down any fact you want a character to know: the name of the puppy you adopted, how two kingdoms border each other, a family tie that only came up in chapter nine. • Add an entry by hand, or build one from any message in the transcript, including older messages you can no longer edit. • Choose where it applies: This Story, or All Chats with that character. • Entries stay until you change them, and are recalled only when relevant. Nothing gets pasted into every message. Finding it • Open Memory & Lore from a character's start or continue screen. You can add Lore before you even send the first message. • Delete records one at a time, or clear every automatic memory without touching your Lorebook. • Living Memory can be paused any time in Settings. Your Lore stays available either way. • Everything here lives on your device, and applies to one-to-one chats. One thing worth knowing: only chats you begin from now on can build automatic memories. A conversation already in progress can't be upgraded to them, and Start Again clears that chat's progress. Your Lorebook works in any chat. Help & Guides • A new Help & Guides section in Settings, with a full walkthrough of memories, Lore, and summaries, plus guides on getting started and on writing speech, action, and thought naturally. • You can also open it straight from the Memory & Lore screen. Bring Your Own Key • If you use your own OpenRouter key, Memory & Lore now runs on it too, and you can choose which models handle that work in Settings. • OpenRouter setup has moved out of the intro and into Settings, so it's there when you want it and out of the way when you don't. • Everyone using their own key keeps unlimited messages. New setups see occasional ads, which Premium removes. If you set yours up before this update, it stays ad-free as a thank-you. Fixes • The close button on the plans screen works right away, with no countdown to wait out. • Curly quotation marks from German and other keyboard layouts are handled properly now, so styled speech no longer breaks. Thanks for the reviews and feedback!

Your Stories Just Got a Whole New Look Chat Wallpapers • You can now set any image behind a chat, with simple sliders for how see-through and how blurred it looks. • Set one default wallpaper for every chat in Settings, or give a single character its own scene in the Workshop that overrides the default. (Group chats use the default; the Iron Wake group has its own.) • Every built-in character now comes with a wallpaper that fits them, so chats open inside a matching scene instead of a flat background. • A dark, cinematic reading look turns on automatically whenever a wallpaper is set, so your text stays easy to read over any background in both light and dark mode. • Opening a chat with a wallpaper plays a short reveal: the scene starts sharp and zoomed in, then eases back as the messages fade in. Help Me Reply • Stuck for what to say? Tap the new sparkle button in the message box and we'll draft your next line for you, written in your persona's voice. • It appears when it's your turn, it's yours to edit, and nothing is sent until you choose. Works in one-on-one and group chats. A Bigger, Refreshed Cast • Five brand-new characters: Teresa (a corporate founder), Dorian (a vampire who rules the old quarter), Mireille (a privateer captain), Dax (a cyberpunk fixer), and Malachar (an ancient demon bound to you by a pact). • Eleven rarely-used characters were retired from new installs. If you've already chatted with one, they stay in your library. • Many of the characters we kept were rewritten so their opening scene hands you a clear role and a real choice instead of a passive intro (Lily is now a goth musician, Hermann's tavern is a pick-a-quest hub, and more). • Every character has fresh artwork in one consistent PersonaChat style, so the whole cast finally looks like it belongs together. • The Iron Wake crew got all-new portraits and a tighter, more polished story. Restore Built-In Characters • New in the Workshop menu: "Restore built-in characters" updates the built-ins to their newest writing and artwork, brings back any you removed, adds the new arrivals, and gives them their wallpapers. All your conversations are kept, and your own created or imported characters are never touched. Smarter Long-Chat Memory • Long stories now hold on to much more of what happened earlier, for noticeably better recall as the chapters pile up. • The running recap is now one smooth summary, so characters stay true to what actually happened. • Summarizing runs quietly in the background instead of interrupting you with a review screen. Prefer to check it first? Turn on a review step in Settings to read or edit each summary. • We can now improve how this works from our side, so future upgrades reach you without an app update. • Fixed a crash that could happen the first time a long group chat created its summary. More Models, Longer Replies • Three new AI models for richer, more natural conversations: GLM 4.7 Flash, MiMo V2.5, and Mistral NeMo. • Replies can run longer now: one-on-one chats allow longer responses, and group chats gained a new "Long" option. Group Chat Polish • Roomier message bubbles, plus an always-visible "⋯" button on every message for quick Edit, Rephrase, and Delete (no more long-pressing to find them). Smoother Start, Clearer Privacy • The intro and setup screens now scroll properly at large text sizes, so the Continue button is always within reach. • Clearer wording that your created and imported characters stay private on your device, and that bringing your own API key is completely optional. Thanks for the reviews and feedback!

New AI Models • Two new built-in models — DeepSeek V4 Flash and Qwen 3.5 Flash — fast, open-source, and included with Premium • DeepSeek V4 Flash is the new default, tuned for vivid, character-driven replies • Qwen 3.5 Flash adds strong multilingual support for chatting in many languages • Each model ships with its own tuned defaults for better quality out of the box Better Conversations • Replies keep clean paragraph breaks, so longer responses are easier to read • Characters stay in their own voice and no longer write your side of the chat • A longer "Long" response-length option for more detailed replies OpenRouter (Bring Your Own Key) • Set a custom response length — now up to 500 tokens for longer replies • All your model settings now apply when using OpenRouter, for finer control over responses Polish • Smoother onboarding, and your existing settings carry over automatically • Various refinements and fixes Thanks for the reviews and feedback — keep it coming!

Character Cards • Less overwhelming to create — optional metadata fields are now tucked together and collapsed by default • New Personality field on 1:1 character cards, separate from Scenario • Expandable editors for long AI prompt and scenario fields • Persona names are trimmed of extra whitespace automatically Group Chats • Shared system prompt — editable per model from AI Settings • Daily-limit and ad-unavailable alerts now show properly instead of failing silently OpenRouter (Bring Your Own Key) • No ads while your OpenRouter key is active • Strict mode with separate model settings for summarization, group turn picking, and card metadata • API keys are validated before saving Thanks for the reviews and feedback — keep it coming!

NEW AI MODELS & FASTER BACKEND A refreshed lineup of five carefully-tuned models — Wayfarer, Anubis v1.2, Fallen, Euryale, and EVA Qwen — each with a plain-language description so you can pick by what you want to do. Wayfarer is the new default for high-stakes adventures; Anubis v1.2 brings a stronger character voice; Fallen handles deeper themes; Euryale stays coherent over long conversations. LIVE STREAMING REPLIES Messages now appear word-by-word as they're written, so you see progress instead of waiting on a blank screen. A stop button lets you cancel a runaway reply and keep whatever's already there. REDESIGNED CHAT INPUT The composer grows up to 10 lines so longer storytelling messages are easier to write. Tap anywhere on the chat to dismiss the keyboard. GROUP CHAT IMPROVEMENTS - Replies now stream word-by-word, with the warmup countdown and stop button working per character — same experience as solo chat. - Character cards now show full names — with a unique tag when names still collide — so look-alike characters can be told apart in the picker, the chat, and the typing indicator. - Response length is configurable up to Medium, so characters can advance the plot instead of only reacting. - Solo-chat summaries no longer carry over into group chats; each conversation stays in its own world. AI SETTINGS - Penalty and sampling sliders are unlocked for reasoning models like Fallen, with an inline note explaining the trade-off. - "Reset to Default" now honors your onboarding response-length preference, and newly-added OpenRouter models inherit it too. - Picking a custom OpenRouter model from the picker now activates it automatically — no separate toggle needed. Removing your active OpenRouter model demotes you cleanly to another one rather than silently failing on the next send. ONBOARDING - Sharper copy for the Lighter and Darker atmosphere choices. - Live countdown during model warmup so the first reply on a cold start doesn't feel like a hang. Plus stability fixes throughout.

URGENT: Fixed a critical bug where chat responses were lost when users briefly switched apps, causing frustration and data loss for active conversations. - Improved connection reliability: the app now automatically retries when the network drops or the server is temporarily busy, so you're less likely to see errors during a chat - Better error messages: instead of a generic error, you'll now see what actually went wrong (network issue, server busy, etc.) - Fixed a layout bug on iPad where the edit message screen was squeezed to one side

What's New: - Bring Your Own Key: connect your OpenRouter API key to chat with any AI model - Monthly subscription plan now available alongside yearly and weekly - Smoother chat scrolling with rebuilt conversation view - Fresh Liquid Glass design for buttons on iOS 26 - New layered app icons that look great everywhere - Group chat now validates your setup before starting - Support for reasoning AI models with smart response trimming - Faster OpenRouter responses with automatic provider routing - Improved AI response length control with Short, Medium, and Long tiers - OpenRouter models now appear in Group Chat AI Settings - Optimized Fallen reasoning model with locked sampling parameters - Fixed group chat character names and profile pictures not displaying - Various bug fixes and performance improvements

Group Chat Adventure Mode Experience a new way to roleplay with multiple AI characters at once! Lead a team, navigate group dynamics, and watch characters interact with each other in real-time. - Chat with multiple AI characters simultaneously - Characters respond naturally based on context and who's being addressed - Smart conversation flow - characters chime in when it makes sense - Built-in story summarization keeps long adventures running smoothly - Try it now with "Iron Wake Crew" - a sci-fi survival scenario included free

Fixed crashes and performance issues in long conversations by optimizing memory usage

- Fixed crashes related to character management and editing responses (when cards were renamed) - Improved clarity of the conversation summary feature to prevent confusion about which messages are being condensed

• Fixed a critical bug where long conversations could crash the app due to incorrect token counting after summarization • Repaired message editing functionality for conversations with summaries, ensuring the correct messages are modified or deleted

- Improved Onboarding, allowing users to choose different AI response lengths and lighter/darker themed AI interactions. - Made creating character cards easier. The sections "Story Description" and "Story Starter Summary" are now pre-filled by AI based on the written scenario and Story Starter that you provide. - Added an amazing new AI model called TheDrummer-Valkyrie-49B-v1. - AI Settings now has Save and Cancel buttons at the top to make it easier. - Changed the default Summarizer prompt, and you can now change the maximum length of the summarizer. - Improved API security and reliability with HMAC encryption - Fixed statistics gathering for AI Model usage. - Removed statistics gathering from Custom cards. It now only gathers usage of PersonaChat cards.

• Fixed chat deletion error: Resolved an issue that could occur when deleting or rephrasing messages in very long conversations. You can now manage your chat history smoothly regardless of conversation length. • Fixed character images not appearing in navigation bar when chatting with custom characters created in Workshop. • Fixed location detection issues: Resolved an issue where some users were incorrectly receiving limited daily messages due to location detection failures. The app will now properly detect your region for appropriate message limits.

- In the Onboarding screen: Fixed Continue button visibility on Dyslexia settings screen when using larger text sizes. - Added privacy screen to onboarding highlighting that chats are never stored or used for training. - Updated character editor navigation with Cancel and Save buttons in the toolbar for improved usability. - Fixed app icon not changing with theme selection on iPad devices. - Fixed an issue where character edits made in the Workshop would occasionally disappear after opening and closing a chat with that character. - Fixed character name changes not updating across tabs  - Fixed persona placeholders being permanently replaced instead of preserved for reuse with different personas. - Automatic restoration of corrupted character templates where persona placeholders were permanently replaced - please delete and restart any affected conversations. - You can now regenerate conversation summaries with the new Re-Roll button (up to 5 attempts) to get the perfect summary. - Summaries are now more concise and factual, focusing on key events and plot points rather than emotional reflections. - Customize how summaries are generated with the new "Summarizer (Llama 3.3)" model in AI Settings, where you can edit the summary prompt to your preferences.

- Simplified character management by removing template/instance separation - now edit any character anytime while preserving chat history - Characters with updated images have been completely rewritten from scratch - Enhanced dialogue and personalities for Adara, Annette, Zane, and Jack - Favourites tab is now correctly showing the Favourite cards upon opening the app. - Fixed keyboard covering text in summary review screen - Fixed swipe-to-delete gesture conflict in Workshop that caused edit view to open instead of deleting character - Fixed an issue where the cursor would jump to the end of text when editing messages on iPad. - Fixed tab bar becoming transparent when launching directly into Workshop tab - Fixed chat view jumping to white screen when receiving large AI responses by improving scroll behaviour and message rendering stability. - New Theme selection in Settings - New App Icons, which will match the selected Theme - New Workshop design - Minor bug fixes

- Improved image caching performance - Eliminated duplicate image download requests that could cause excessive network activity - Fixed app crash when editing messages in chat - Resolved index out of bounds error that occurred when editing messages, particularly in longer conversations - Fixed scroll position jumping after editing messages - Chat view now maintains proper scroll position when returning from the edit screen, preventing disorienting jumps to unexpected locations - Fixed Edit Message screen not using dyslexia font - The message editor now properly respects the dyslexia font setting when enabled - Reduced unnecessary app initialization calls - Added throttling to prevent repeated initialization between messages

- Fixed a crashing bug when there is a network error during the ReWrite function. The app will now restore the state of the chat to the pre-rewrite state. - Fixed the blue speech bubble problem with the Fallen model. - Fixed the empty photo bug in Personas. - Fixed the TextEditor bug on iPad when editing a message, where the cursor jumps to the end while typing. - Fixed the usage of incompatible quote-markups in Character "Theo". - Added three new story characters: * Greta Baumann, a sharp-tongued farmer’s daughter in medieval Germany. * Sol Mire, a haunted survivor of a failed deep-space mission. * Ivy Larkspur, a sarcastic teen sleuth in a fog-shrouded coastal town.

- Fixed an issue that could cause the app to unexpectedly close during conversations. - Fixed an issue that could cause the app to close unexpectedly when loading certain characters. - Fixed quote sign issue with Fallen model. - Raised frequencyPenalty in AI Settings to 0.2 to avoid repetitive chats across all AI models. - Added three new story characters: Lucien Vale, a brooding vampire with a haunted past. Arden Solari, a memory architect in a near-future romance. Theo Brightwill, a shy teen baker in a magical village. - Updated system libraries. - Minor fix for Fallen model.

- Added new AI Models: Anubis, Hanami, Fallen. - Replaced Magnum model with a newer Special Edition version. - Reworked the AI Settings and System prompts for all AI models. - Each AI model now has its own AI settings and system prompt, so you can experiment better and switch between them more easily. - Your current AI settings are now migrated to Euryale 2.3. You can reset them in Workshop to revert to default settings if you wish. - Fixed a minor syntax bug with AI model processing. - Improved stop tokens of AI Models. - Removed example dialogues from existing characters, so new games remain unpredictable and aren't affected by existing dialogues. - Added improved analytics to see which AI models are used less, so we can retire them in the future. - Added 6 new character cards: Elias, Cassian, Sasha, Milo, Clara, Jeremy. - Settings screen improved. - Improved onboarding screens.

* Summarization: In the previous release, we made a mistake and created a diary from the player's perspective, instead of the character's perspective. We have now fixed that. * Updated core dependencies to the latest stable versions.

- Improved the Summarization algorithm - Improved the Rating mechanism - Updated core dependencies to the latest stable versions

- Added an AI Settings shortcut to the Model Selection screen, allowing easier access and modification during ongoing chats. - Removed Dialogue Examples from Character creation to simplify the process and reduce user confusion. - Retained Euryale 2.2 model with updated description to provide continued access to this language model option. - Updated core dependencies to latest stable versions

- You can now switch between four different AI models during chat: Euryale 2.2, Euryale 2.3, Kunou 1.0, and Magnum 4.0. - The default AI model has been updated to Euryale 2.3. If you encounter any issues, you can switch back to the previous version or try a different model. - Added a new character card: Ronno. - Implemented a minor fix to the Adara character card. - Updated Zane's character card photo. - Various other minor improvements and fixes.

- Fixed text editor layout in chat history summarization view to properly fill the screen on iPad devices. - Added ability to edit conversation summaries by tapping the menu button (⋯) on any AI message after an initial summary point has taken place.