v2.1.1

Important

Android APK notice: The APK is a Termux bootstrap + WebView shell, not a native Android server build. It opens an already-running local Marinara server, and on first launch it can download Termux from F-Droid, hand it to Android's installer, and start Marinara through Termux after Android permission prompts. Follow the Android wrapper guide if Android blocks the bootstrap handoff.

Added

• Added a Grok CLI (Subscription) connection provider that routes chat requests through a local grok CLI login for SuperGrok / X Premium+ users, with in-app setup instructions and no API key or base URL fields.
• New, fancy animations and images for dice rolls.

Changed

• Bumped release metadata to v2.1.1 across packages, the PWA manifest, the README release pointer, Windows installer sources, Android APK metadata, and the home-page-visible app version.
• Reorganized Settings so Addons combines Themes and Extensions, Generations houses image/video generation and prompt overrides, Imports uses the plural label, generation prompt editors start collapsed, and Danger Zone uses accent-color selections with no default checked categories.
• Refreshed the in-app Credits modal from the GitHub contributors list and added credits sync/check helpers to the release workflow.
• Split the server generation route into focused prompt, provider, context, and command-runtime modules so Conversation commands, Professor Mari actions, turn games, provider setup, and post-processing paths are easier to review and maintain.
• Removed unused API Key and Base URL fields from local subscription/auth providers such as Claude Subscription, ChatGPT/Codex auth, and Grok CLI.

Fixed

• Fixed Marinara profile/full backup exports so generated scene-video MP4s and reusable Conversation Call character video clips are included with their storage metadata and accepted on profile ZIP import (#3315).
• Fixed Card Evolution Auditor review proposals being falsely marked stale from cached character data by refreshing the character card before validation. He added an explicit stale override path that appends edited replacement text after confirmation instead of blocking the proposal outright. (#3314).
• Fixed Memory Recall embedding creation for giant memories by chunking oversized recall text before sending it to embedding providers, preventing single giant entries from exceeding provider limits (#3317).
• Fixed Windows/git updater and launcher flows so they use the pinned pnpm version, avoid stale aggregate clean/build commands, and rebuild shared/server/client in update-safe package steps (#3318, #3323, #3324).
• Fixed Bot Browser avatar importing for sources that return generic or missing image content types, including Datacat, Character Tavern, Janitor/Janny, Pygmalion, and Wyvern avatar CDNs (#3319).
• Fixed Game Mode NPC portrait generation so GM-generated NPC descriptions are distilled into canonical portrait guidance for the initial and later NPC asset prompts (#3321).
• Fixed the Roleplay setup wizard's Use Settings Presets shortcut so it still seeds the selected character's first message when creating an empty chat (#3322).
• Added phonetic name fields to Character and Persona metadata, resolves phonetic-name macros, and uses those values when sending Conversation Call voice lines to TTS so names can be pronounced correctly (#3325).
• Fixed Conversation Call prompt macro resolution so live audio prompts, persona/character context, lorebook entries, daily history, and call transcript content resolve macros such as {{user}} before provider dispatch (#3326).
• Fixed custom preset choice confirmation so edited multi-select preset variables can intentionally be saved empty instead of disabling Confirm Choices (#3327).
• Fixed Conversation message avatars that were cropped in editors but appeared oversized in DMs/groups by restoring the relative avatar crop container (#3328).
• Fixed Conversation mode prompt assembly so selected preset wrap formats (XML, Markdown, or None) are respected for instructions, daily summary/date blocks, current context, memories, and lorebook injections instead of always wrapping supplemental sections in XML.
• Restored the Echo Chamber chat as a collapsible panel instead of making close/collapse hide the chamber permanently for the chat (#3329).
• Fixed manual group Conversation character triggers so the prompted character receives recent visible group transcript context and does not answer from profile/lorebook content alone (#3330).
• Fixed Roleplay /guided respond for <character> in individual group chats so Manual mode asks for an explicit target instead of no-oping, and Smart mode honors the requested character instead of the queued responder badge.
• Fixed first-turn lorebook activation so opening chat messages remain keyword-scannable during the first user generation even when the recent-message scan window would otherwise drop earlier group greetings (#3332).
• Fixed Text to Speech character voice assignment so users can add character voices using provider default voices, including ElevenLabs defaults, even before custom/account voices are loaded.
• Added a per-chat Conversation Calls setting to disable generated bracketed voice cues for TTS providers that do not accept [whispering], [laughing], [sighs], and similar tags.
• Fixed Grok CLI (Subscription) connections so they no longer seed stale grok-build-* model aliases, can use the local CLI default model with a blank model field, and fetch selectable model IDs from grok models.
• Fixed Grok CLI (Subscription) roleplay requests by preferring the safe headless Composer model when discovered, starting Grok CLI connections with a safer 32k context window, and surfacing a clearer context-limit hint when the CLI reports max turns reached.
• Added Lorebook entry-status sorting, fixed continued assistant messages so appended text starts after a blank line, and removed the obsolete tracked pr-evidence/ artifacts while ignoring future evidence folders (#3336).
• Fixed Conversation-created scene chats so their Prompt Preset selector remains visible while still defaulting to None.
• Made the Connections panel's unfiled/root drop area more forgiving so desktop users, including Windows users, can reliably drag connections out of folders without hitting a tiny target.
• Fixed native lorebook import so nested folders keep their parent/child hierarchy instead of being flattened to the top level (#3347).
• Fixed the character Lorebook tab so an embedded lorebook can be removed from the card: added a Remove from card action (including for cards with no linked copy), clarified that the row delete only unlinks the standalone while the embedded copy stays, and renamed Edit Linked Lorebook to Edit Embedded Lorebook (#3359).