v2.0.6

What a predictable creature!

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 synced custom theme Accent Pulse opt-in so CSS themes can request the built-in pulse with --marinara-theme-accent-pulse: enabled.
• Added a Stable/Staging update channel selector with staging warnings and channel-aware apply checks (#2912).
• Added searchable Home FAQ controls, saved Professor Mari chat history management, Game Mode manual background generation, and lorebook vector deletion controls (#2913, #2909, #2902, #2900).
• Added Up/Down controls for alternate greetings in the Character Editor so card authors can reorder greetings without copy/paste work (#2917).
• Added native Gemini API embedding support for Google and Vertex Gemini connections so lorebook vectorization and memory recall can use Gemini embedding models (#2889).
• Added a per-chat AI translation prompt override in Chat Settings, with a restore-default action, so chats can customize the translation system prompt without losing the built-in default (#2883).
• Added llama.cpp sidecar embedding endpoint controls for pooling type and physical batch size so Gemma and other embedding GGUF models can use OpenAI-compatible lorebook/memory embeddings when they require non-default pooling (#2863).

Changed

• Chat Branches no longer shows a separate "Active" pill; the checkmark and active row highlight identify the selected branch, while rename/delete actions remain available for the active branch.
• Memory recall chunking now behaves as read-behind storage when a chat message limit is set, keeping the active prompt tail out of durable memory chunks (#2862).

Fixed

• Fixed compact UI layout polish around the Browser source menu, Settings tab labels, Game Assets import actions, Advanced update/admin buttons, and Lorebook overview control sizing/tooltips.
• Fixed Import Profile and Advanced Danger Zone settings buttons so they use the shared neutral Marinara chrome button styling, with Danger Zone actions stacked one per row.
• Fixed CodeRabbit review findings around recovery-boundary safety, Professor Mari chat switching, FAQ search accessibility, lorebook export/default compatibility, chat metadata patching, import mode validation, update channel checkout safety, and avatar crop normalization.
• Fixed Chats sidebar Conversation rows so they match Roleplay/Game row density while blank/new Conversation chat fallback icons use the cyan mode color instead of the custom chat accent.
• Fixed the Professor Mari home experience so desktop opens the chat inline in place of the home menu while mobile keeps its prior full-screen focus flow, the FAQ opens by default only on desktop, the FAQ/Professor launch card stays taller and evenly split with centered welcome copy on larger screens, the chat composer starts as a single-line input, achievements align to the home card width, missing-connection guidance points at the chain selector, closing the desktop chat no longer flashes homepage text, tutorial copy uses Chat Chrome text colors, Professor chat history controls live inside the chat window, and an in-progress Professor Mari chat can follow the user as an accent-bordered dismissible floating companion after they leave the home screen or open mobile detail sheets, with a DJ-sized circular mobile button and without loading the floating chat machinery while it is hidden on the home shell.
• Fixed broad app slowness paths by avoiding full chat-list refetches when opening Chat Settings, removing eager settings preloads, showing immediate settings/branch loading feedback, skipping Game snapshot copy work for non-Game branches, and stripping bulky internal prompt/debug payloads from branched/exported/imported messages (#2914, #2913).
• Fixed Game Mode setup list editing, guided generation macro resolution, game portrait/background prompt handling, NovelAI image sizing, image-prompt style compaction, and roleplay empty-send behavior (#2915, #2906, #2905, #2903, #2902, #2894, #2893, #2892).
• Fixed chat export/import fidelity by preserving mode/persona metadata, resolving macros in exported transcripts, preserving group-chat speaker snapshots after member removal, stripping internal export payloads, and exporting compatible lorebook entries as arrays (#2913, #2910, #2904, #2901, #2897, #2895).
• Fixed Professor Mari workspace/home behavior by saving previous chats on restart, exposing prior chats for rename/delete/reopen, preserving the selected persona, preventing repeated command-failure loops, asking clarifying questions before vague persona creation, and using a Termux-compatible mari shim path on Android (#2911, #2909, #2899, #2891).
• Fixed persistent black-screen recovery, Android/touch popover dismissal around Author Notes and Chat Settings pickers, JannyAI detail imports, blank preset variable options, and default vectorization state for new lorebooks (#2908, #2907, #2900, #2898, #2896).
• Fixed impersonation generations so preset-driven prompts skip regular preset instructions while preserving marker-provided context, preventing conflicting "respond as the assistant" system text from contaminating /impersonate prompts (#2886).
• Fixed Professor Mari home-chat restart so chat messages are deleted only after the workspace reset succeeds, preventing failed restarts from causing delayed chat history loss (#2887).
• Fixed Professor Mari workspace privileged-route access so trusted LAN/Tailscale clients can use the workspace when loopback-only mode is disabled, while database command execution remains loopback-only (#2884).
• Fixed privileged-route parameter errors so missing or invalid admin access is not rewritten as a generation-parameter warning (#2884).
• Fixed chat exports so saved thinking/reasoning content is included in text exports and mirrored in JSONL exports/imports (#2881).
• Fixed sprite prompt compilation so concise user descriptions survive prompt review/compaction, and reviewed prompts no longer receive a second layout/negative suffix (#2871).
• Fixed memory-recall branch contamination by pruning native chunks whose timestamp span no longer matches the current chat message log (#2862).
• Fixed v2.0.6 release metadata across packages, the homepage-visible app version, Windows installer sources, PWA manifest, README release pointer, and Android APK metadata.

Platform Notes

• Android versionName is 2.0.6 with versionCode 25.
• Windows, macOS/Linux, Termux, Docker, APK, and PWA users can update through the usual v2 updater paths once release assets are published.