1.5
2026-05-18
What's New in Version 1.5:
Per-Character Memory (Experimental)
Your characters now remember you. Facts about you and details they've shared about themselves persist across conversations. Say "please remember that..." to save something instantly, or let memories build naturally as you chat. Ask "do you remember what we talked about?" and they'll search your past conversations. View, add, edit, and delete memories from the character editor. Toggle on/off in Settings.
Wraith Originals — Custom Fine-Tuned Models
Introducing 4 personality models we trained from scratch: Ghostly (sarcastic roaster), Moth (devoted companion), Petal (tsundere), and Spectre (glitchy robot). These aren't system prompts — the personality is baked into the model weights. Find them under Wraith Originals in the Models tab. Highly experimental — let us know what you think!
Wraith Character Hub
Browse and download community-created characters at wraith.chat. Submit your own characters directly from the app with the "Share to Wraith Hub" button — your character data pre-fills automatically. All submissions are reviewed before going live.
Fixes and Polish
Fixed model switching crash that caused repeated loading errors. Fixed double-tap creating duplicate conversations. Fixed image generation being triggered when image model isn't downloaded. Improved model loading reliability when switching between characters. Memory toggle in Settings with Experimental label.
1.4.1
2026-05-10
What's New in v1.4.1
Improvements:
Lightning-Fast Image Generation
Completely rebuilt the image generation engine. Images now generate up to 4x faster than before. You can also control quality vs speed with the new Image Gen Steps slider in Settings.
Bulletproof Image Gen Cancel
Tap cancel and you're back to chatting instantly. No more frozen screens, no more force-quitting. Cancel works at any point — even before the first step starts.
Honest Loading UI
When the model needs to reload after image generation, the app now tells you exactly what's happening: "Cancelling image generation...", "Releasing GPU memory...", "Loading model..." — no more mysterious waits.
Smoother Streaming & Stop
Fixed multiple performance issues that made streaming feel sluggish and the stop button laggy on longer conversations. Text now flows smoothly at 60fps with frame-synced typewriter rendering.
Thinking Mode Polish
Thinking indicator only appears on models that actually support it (Qwen3, DeepSeek R1). Skip button lets you jump past long reasoning blocks without stopping the response.
Custom Model Improvements
HuggingFace model browser now accepts repo URLs, direct .gguf links, or repo names. Better error messages when a model fails to load — tells you if it's an architecture issue, memory issue, or corrupted download.
Bug Fixes:
Fixed model mismatch bug where switching characters could leave the wrong model loaded
Fixed custom character edits not taking effect until app restart
Fixed rare crash when cancelling image generation during CoreML setup phase
Fixed duplicate messages when double-tapping stop button quickly
1.3
2026-05-01
What's New in v1.3
- Tap any AI-generated image to view fullscreen, then save to Photos
- Bulk delete chats with new selection mode
- Custom system prompt editor for power users (Settings → Advanced)
- Better image generation — now draws what you actually ask for
- Live progress bar during image generation
Stability:
- Fixed app freezing on launch after a crash
- Fixed characters losing their personality after reopening chats
- Fixed crashes during image generation on lower-memory devices
- Generated images now persist across app updates
Thank you to our early supporters!