Pasta-Devs/Marinara-Engine v1.5.5

on GitHub

Added

• New agent: Card Evolution Auditor that actively updates your characters as they grow.
• Polska gurom!!! In Game mode.
• GM can now add party members during the game and create character cards for them.
• Turn, Scene Analysis, and Assets Image Generation retry button in Game mode.
• Improved Game mode's structure and prompts.
• Custom widgets, notes/books, session summaries, and inventory in Game mode are now all editable.
• You can now upload custom NPC portraits in Game mode when clicking on the portraits.
• The Characters tab now opens a full-page library with large card browsing, creator-note previews, and a selected-card overview before editing.
• Chat galleries and character galleries now support selecting and uploading multiple images in one action.
• Chat branches can now be switched from a selector at the top of the chat bar instead of only through Manage Chat Files.
• Conversation schedules now let you customize per-character idle and DND response delays, plus inactivity follow-up timing.
• Character titles to mirror the ones Personas have.
• Various macros, see all under /macros.
• Game mode combat improvements (statuses, abilities).
• Bulk delete.
• Search filters for chats in the Chats tab.
• TTS support.
• FAQ on the home page.

Fixed

• Fresh installs and client builds no longer fail with ERR_PNPM_RECURSIVE_RUN_FIRST_FAIL because the shared package now builds from root entrypoints instead of the client package's nested predev and prebuild hooks.
• The lite container release workflow now inspects the correct -lite image tag instead of the nonexistent *-lite-lite tag, so tagged lite image publishing completes successfully again.
• Fixed sidecar startup state and enabled logs for Ollama to see what's going on.
• You can now use tab when writing lorebook entries.
• Some image generation endpoints.
• Clicking roleplay image attachments now opens them in Marinara's in-app lightbox instead of a new browser tab.
• Auto-play in game mode now pauses when you're reading a note, a book, or doing a QTE event.
• Opening a conversation no longer resets the autonomous-message inactivity timers just because the message history finished loading.
• OpenAI-compatible connections no longer send reasoning payloads to models that do not support them.
• Selfies and sprite generation no longer force a character avatar as a hidden reference image by default.
• Explicitly adding or editing an agent no longer persists it as globally disabled.
• Memory recall now stays inside a dedicated prompt budget before injection, preventing recalled history from crowding out agent and thinking context.
• Exporting a modified character to PNG no longer reuses stale embedded card metadata from the avatar image.
• Sprites get displayed automatically when you add Expression Engine to your chat, and their setup was moved to the Agents section of Chat Settings.
• More ComfyUI fixes.
• Group chats' inconsistent injections: now, upon regenerations, the model knows who should respond.
• Game mode scene-wrap now only sends the current party's character names instead of the entire imported character library, preventing large libraries from tripping the 100-name limit.
• Professor Mari now has access to all the fields in character cards/personas/lorebooks/etc. and can correctly split info into them.
• The Windows installer now downloads Git from a valid prerequisite URL again instead of failing the autodownload step with a missing PowerShell -Uri argument.
• Mobile UI fixes for Game mode.
• Increased the output size to 16384 tokens on the new Game setup generation to prevent malformed JSON errors.
• Decreased padding for text in boxes in the Glued Side Panel avatars option.
• Edit Sheet in Game mode black screen bug.
• CYOA choices can now be edited.
• UI fixes.
• Lorebook entries now don't stay active after they've been activated once, and the lorebooks respect the token limits of how many active entries there may be at once.
• Custom widgets now may change between sessions.
• No more looping music/ambiance in Game mode.
• If a provider accepts a smaller context size than the overall model allows, we now automatically reduce the output size to match the allowed size.