V1.4.1 - Much needed bug fixes

[1.4.1] - 2026-06-08

Added

• Character Goal Tracking: Decision LLM now generates a short-form private goal (what the character wants from the conversation) on the same triggers as mood, injected into the chat prompt and editable via the Mood & Goal modal
• Hide Images from Carousel: Toggle individual images out of the chat image carousel with an eye button in the gallery
• Reset Swipes Debug Command: window.resetSwipes() console helper clears today's swipe count for testing

Changed

• Dark Mode Only: Removed the light mode toggle; the app is now always dark
• Removed Left-on-Read Follow-ups: Retired the unused left-on-read proactive feature
• Longer Login: JWT tokens now effectively never expire, so no more periodic re-login
• Description Editor: Auto-growing textarea that fits the content, with Save/Cancel buttons at the top of the Overview tab
• Freer Swipe Cards: Discover cards lift and rotate outside the frame while dragging instead of being clipped to a square

Fixed

• Character mood and goal not resetting on 30+ minute time gaps
• Chat sidebar not re-sorting or refreshing when new messages arrive
• Duplicate background mood system messages when the mood was unchanged
• Memory not being wiped on user-initiated unmatch or by the clear-all button
• Sidebar avatar not updating after a portrait change
• Chat image carousel not syncing when swiping image messages
• Proactive consecutive count getting stuck (empty conversations and trailing system/time-gap messages) and falsely flagging characters as capped
• Stale Library search results when typing during an in-flight load
• Deleting a TIME GAP marker leaving an orphaned message on reload

V1.4.0 - Activities & Attributes

[1.4.0] - 2026-03-13

Added

Activities System

• Icebreaker: Character asks a personality-driven question with 10 randomized styles, answer to start a timed chat
• Two Truths & A Lie: Character generates 3 statements (2 truths, 1 lie), guess the lie then chat about it
• This or That: 7 rounds of rapid-fire preference pairs, then timed chat discussing choices
• Would You Rather: 5 rounds of fun dilemma scenarios, then timed chat about decisions
• Activity Game Context Persistence: Game results (answers, choices, guesses) now stored on conversations and injected into every chat prompt, so the AI remembers what happened during the game
• Activity-Specific Chat Prompts: Each activity mode gets a unique chat context instead of generic "random chat"
• Activities Hub: Designed with cards for all activity types
• Activities Settings Page: Configure character selection options (include away/busy characters)
• Character First Message Chance: Configurable probability for who messages first in activities

Random Chat & Matching

• Random Chat Feature: Timed chat with random unmatched characters, option to match after timer
• Blind Date Mode: Hidden identity chat where only first initial is shown
• Proactive First Message: Characters can send the first message in activity chats

Character Attributes

• Configurable Attribute System: Replaced Big Five personality with fully customizable character attributes
• Attribute-Driven Behavior: Character attributes included in schedule generation and chat prompts

LLM Configuration

• Proactive LLM Config: Independent LLM settings for proactive message generation
• Random Model Pool: Configure a pool of random models for any LLM type
• Double Text Chance: Configurable probability for characters to send follow-up messages

Chat Improvements

• Image Gallery: View all images in a conversation with navigation
• Early Reactions: Characters can react to messages earlier in the conversation
• Typing Indicator Alignment: Improved positioning of typing indicators

Fixed

• Avatar rotation and gallery not updating when messages with images are deleted
• Asterisk words being incorrectly removed (e.g., "I really like that" losing the word)
• Stale model selector in LLM settings after switching providers
• Daily auto-match enable not working
• Smiley/emoji stripping in messages
• Image rotation not showing images from older paginated messages
• Message pagination being reset during conversation
• Autoscroll reliability improvements
• Double text when response already has multiple messages
• Several performance optimizations

V1.3.0 - Even more QoL

[1.3.0] - 2025-01-07

Added

Prompts System

• Prompt Presets: Save, load, and delete prompt configurations as named presets
• Current Preset Indicator: Shows which preset is active with unsaved changes warning
• Export/Import: Download prompts as JSON file or import from file
• Character Wizard Prompts: New configurable prompts for wizard (description, appearance, image tags)
• Schedule Day Prompt: Generate single day schedules with dedicated prompt

Chat Features

• Horizontal Images as Background: Display landscape images as chat background instead of sidebar
• Library Card from Chat: Access character's library card directly from chat menu
• Character State Modal: Click state pill in chat to manually edit character state

LLM Settings

• Reasoning Effort: Configure reasoning effort for OpenRouter models (low/medium/high)
• Retry on Invalid Response: Toggle automatic retry when AI returns invalid/empty response
• Request Timeout: Configure timeout duration for all LLM types
• Use Name Primer: Toggle character name priming at end of prompts

Behavior Settings

• Unlimited Proactive Messages: Set daily limit to 0 for unlimited proactive messages
• Dynamic Character States: States loaded from config file for easy customization

Discover Page

• Persistent Daily Randomization: Same random order maintained throughout the day
• Swipe Limit UI: Visual indicator for remaining daily swipes

Changed

• Refactored aiService.js into focused single-responsibility services
• Refactored backend routes (chat.js, users.js) into modular structure
• Refactored decision engine and proactive message services into smaller modules
• Refactored ChatHeader component with cleaner organization
• Character mood & state now included in decision context for smarter AI choices
• Improved proactive messaging with better decision context
• Improved character state handling and image tag generation
• Character wizard now uses configurable prompts instead of hardcoded ones

Fixed

• Library card modal not updating after generating content when accessed from chat
• Schedule parsing broken by chat-specific timestamp stripping
• Horizontal background controls positioning and visibility state
• Randomize orientation setting not being applied to generated images
• Thinking model output leaking into responses (reasoning tags now stripped)
• Sidebar not updating when character schedule is changed
• Library sorting for newly imported characters
• Library sorting by newest/oldest not working correctly
• Revert button not showing in Overview tab
• Auto-unmatch setting being ignored (added safeguards)
• Message counting for character state updates
• Image message editing now shows image during edit mode
• Advanced settings panel for NanoGPT provider
• Standalone timestamps in AI responses now stripped

V1.2.0 - Character States

V1.2.0 - Character States - 2025-12-12

Added

Character Mood & State System

• Character Mood: Dynamic emotional state displayed as purple pill in chat header (e.g., "feeling flirty and playful")
• Clickable Mood UI: Click mood pill to manually edit character's mood
• Character States: Special behavior states (drunk, showering, etc.) shown as orange pill
• State Instructions: Character states inject special behavior instructions into chat prompts
• Mood/state triggers on TIME GAP or every 25 messages via Decision LLM
• Background mood updates with 25-message cooldown

AI Providers & Settings

• NanoGPT Provider: Third AI provider option alongside OpenRouter and Featherless
• Extended Sampling Parameters: OpenRouter models now support top_k, repetition_penalty, min_p when model supports them
• Include Full Schedule: New behavior setting to include complete weekly schedule in chat prompts
• Proactive When Online: Configure probability of proactive messages when character is online (0-100%)

Storage & Multi-Device

• Backend-Only Storage: All character data now stored in backend SQLite (no more IndexedDB)
• Multi-Device Support: Login from any browser/device and all characters are available
• Per-User Config Files: Each user has isolated prompt and tag library configurations

Library & Settings

• Sort Dropdown: Sort characters by Newest, Oldest, or Random in Library
• SD Resolution Settings: Configure width/height for Stable Diffusion image generation
• Account Deletion: Delete your account and all associated data from Danger Zone in Profile

Changed

• Renamed "Content LLM" to "Chat LLM" in UI for clarity
• Proactive message generation now uses Metadata LLM (better for complex reasoning tasks)
• Character service refactored to use API calls instead of local storage
• Simplified ImageTab and CharacterProfile components (removed dual-save logic)

Fixed

• Character Rename: Now properly updates name in prompts and syncs to backend
• Empty Response Detection: Detect and retry when AI returns just "Character Name:" with no content
• RP Action Stripping: Strip asterisk roleplay actions (system prompt forbids this formatting)
• Schedule in Prompts: Full schedule now placed correctly after system prompt
• Delete-From Messages: Now removes correct message instead of all messages with same timestamp
• Typing Indicator: Added server-side pending request tracking to prevent stuck indicators
• Memory Degradation: Apply degradation once per compaction session, add safeguards against memory loss
• UI Lock on No Response: Frontend now unlocks when character decides not to engage (30% chance)
• User Created Date: Migration fixes users with invalid created_at timestamp

Removed

• IndexedDB storage (frontend/src/services/db.js, characterStorage.js)
• Character sync utilities (no longer needed with backend-only storage)

V1.1.0 - QoL features

[1.1.0] - 2025-11-30

Added

Library Improvements

• Search: Search characters by name and tags with debounced input
• Pagination: 24 items per page (grid), 50 items per page (compact)
• Compact View: Toggle between grid and list views
• Lazy Loading: Images only load when scrolled into view

Sidebar Improvements

• Search: Click "Matches" title to search matches by name
• Thumbnails: Auto-generated thumbnails for faster sidebar loading

Chat Features

• Message Swipes: Navigate between message variants with arrow buttons
• Image Regeneration: Regenerate image messages with new SD generation
• Conversation Export: Export chat history as JSON from header menu
• Persistent UI States: Avatar visibility and banner collapsed states saved
• TIME GAP Combining: Consecutive time gap markers merged into single marker

Character System

• V1 Card Support: Import JanitorAI and other V1 format character cards
• Character-Specific Image Prompts: Override main/negative prompts per character
• Post Instructions: Custom instructions for character social media posts

Memory System

• Memory Management UI: Add, edit, delete memories from chat header
• Remove All: Bulk delete all memories with confirmation
• Memory in Prompts: Memories now included in all AI prompts
• Memory Degradation: Configurable point-based memory decay system
• User Name in Memories: Display actual username instead of "User"

LLM Configuration

• Metadata LLM: Fourth LLM type for character generation (profiles, schedules, personality)
• Token Counter: View token usage in Overview tab

Settings

• Max Matches Limit: Set maximum number of active matches (0-50)
• Thought Frequency: Configure how often characters share internal thoughts

Changed

• Proactive message opener now randomly selects one variety instead of listing all
• Proactive check interval range extended to 5 minutes - 5 hours
• Message edit window now matches bubble size exactly
• Character profile layout changed to side-by-side

Fixed

• Multi-line message ordering (messages with same timestamp now sort by ID)
• Duplicate AI responses now rejected and trigger retry
• Incomplete think tags detected and rejected
• LLM settings validation for OpenRouter provider
• Max matches limit now only counts characters with active conversations
• Duplicate TIME GAP markers prevented
• EM DASH replacement now properly uppercases next letter
• Character-specific image prompt overrides now properly applied
• Proactive message timing enforces interval between consecutive messages

Removed

• Deprecated maxEmojisPerMessage setting
• Left-on-read settings (will reimplement later)

Cupid AI v1.0.0 - Release

Cupid AI v1.0.0 - Initial Release

Initial release of Cupid AI

Highlights

Triple LLM Architecture

• Content LLM: Generates character responses and conversations
• Decision LLM: Makes behavioral decisions (reactions, moods, unmatch, voice/image)
• Image Tag LLM: Generates Danbooru tags for Stable Diffusion

AI Characters

• Import character cards from PNG v2 format
• Character Wizard for AI-generated characters (4-step process)
• Dating profiles with interests, fun facts, and preferences
• Weekly schedules with online/away/busy/offline patterns
• Big Five (OCEAN) personality traits that influence behavior
• Tinder-style swipe interface (like/pass with undo)

Realistic Chat System

• Proactive Messaging: Characters initiate conversations after time gaps
• Schedule-Based Engagement: Characters respond differently based on their current status
• Memory System: AI extracts and stores up to 100 important facts per character (configurable 0-100)
• Conversation Compacting: Automatic AI summarization to manage context window efficiently
• Mood Effects: Dynamic chat backgrounds and message reactions based on conversation tone

Image Generation

• Stable Diffusion WebUI integration
• Context-aware image generation during conversations
• Smart AI decisions (can send, tease, or refuse requests)

Extensive Customization

• Behavior settings (proactive messaging, memory, compaction, swipe limits)
• LLM configuration (3 independent systems with separate providers)
• Stable Diffusion settings (sampling, CFG, resolution, ADetailer)
• AI prompt customization (system prompts, decision engine, memory extraction)
• Tag library management for image generation

Tech Stack

Frontend: React + Vite, Tailwind CSS, IndexedDB, Socket.IO
Backend: Node.js + Express, SQLite, JWT Authentication, Socket.IO
External Services: OpenRouter / Featherless (LLM providers), Stable Diffusion WebUI (optional)

Getting Started

Prerequisites

• Node.js 18+

One LLM provider:

• OpenRouter API key (get one here)
• Featherless API key (get one here)

Optional:

• Stable Diffusion WebUI (for image generation)

Documentation

• README.md: Complete feature documentation
• CHANGELOG.md: Version history
• backend/.env.example: Environment configuration guide

License

MIT License - see LICENSE file for details