itch.io

Try using the desktop app and not the web-app or vice versa

Hi, ty for your feedback. You need to manually save your modifications -> there is a save button at the bottom of each panel (you need to scroll to the bottom to see it).

You can run any model you want locally using LM Studio or Ollama — I'll leave it at that.

Ty, I'll look into it - if you have a decent GPU I'd really recommend using OmniVoice for local voice cloning it's really good.

Tysm for the detailed feedback !

On OmniVoice and AMD: good news, AMD is actually already supported, but only on RDNA3 (I'm on a 7800XT). It looks like what I set up for RDNA3 doesn't carry over to RDNA4, which is why the server won't start on your 9070XT. I'm fairly confident I can get it working on RDNA4 pretty quickly — it should just take a couple of back-and-forths and some testing. If you're up for helping out a bit, feel free to join the Discord server and we'll sort it out together.

On the AI response length: you can already adjust this! Open the Settings panel → AI tab → click "Show advanced options," and you'll find a Response length slider right at the bottom.

On the lore section: you can set this up today by creating a custom persona. Head to the Persona tab and edit the default "The witty historian" — copy it into a notepad, add your own lore elements, then create a new persona with that content and assign it to a new Character.

Thanks again for taking the time to write all this out — it genuinely helps me improve the app!

Watch this at 2:07

That's a big piece of work so at least 1 month

Thanks! Tool calling and MCP support are the next big piece of work I'm focusing on — this will include web search, browsing websites, controlling Windows, Spotify, and more. So yes, this is definitely planned for a future update.

Scaling options are on the way

Adding Chinese as an input language is already on the roadmap, so that one's coming. For the TTS side though, I'll have to pass on Volcengine. Each new TTS provider takes a considerable amount of work to integrate, and I don't add new providers just to support a single language. The good news is that Chinese already works really well with OmniVoice, a local TTS option built into the app — it supports voice cloning and the quality is genuinely great.

Actually there are two problems. First, Chinese isn't implemented as an input language in the app — Whisper somehow picks up what you say in Chinese and translates it into English on its own. And if I were to add Chinese as an input language, it would do the reverse: translate what you say in English into Chinese. Supporting two languages at the same time isn't possible at the moment (and probably won't be in the future).

Yep i recommend gemma 4 right away on the get started tutorial video every user see when opening the app for the first time, it's hands down the best (free) model for the app. I'll add the gemma 4 reasoning tags in a future update.

Oh I see — so I have a strip for <thinking> tags in the LM Studio connector, but it only works when both the opening and closing thinking tags are in the complete response, even though it's plugged in streaming mode. You can try increasing the response length to the max (600) in the AI tab so the full response includes both opening and closing thinking tags. Also, I'm a solo dev working on this — thanks again for the support!

Ty the feedback! I highly recommend the Gemma 4 models from 'unsloth', especially gemma-4-26b-a4b-it and gemma-4-e4b-it (depending on your VRAM) — they work great and reasoning is disabled.

Yes, you can talk to the AI — it has both voice and text modes. If you run the LLM locally (via LM Studio or Ollama), nothing leaves your computer; everything, including screenshots, is stored locally. If you use online APIs like Google Gemini, then your messages and screenshots are sent to their servers. Feel free to join the Discord and ask people who've been using it for months.

Atm it's a fixed timer, but adding a randomized range is definitely something I can do

Hello! Yes, it has an auto-speak feature. In the future, I plan to make it more reactive so it'll auto-speak when you receive an email, when a tweet gets posted, that kind of stuff.

Hello and thank you for your support! To answer your questions:

• The 250-character limit in text mode is static, but I can definitely increase it.
• Qwen3-TTS is a high-end TTS provider — the minimum requirement to use it is an RTX 4080+. That said, I just released a new update yesterday (v0.7.13) featuring OmniVoice, a new local voice cloning TTS provider with very good quality and performance. Join the Discord and check the #updates channel for more details!
• Simply use your itch.io email address — both work, but most people use the email address linked to their itch.io account.

Ty !

make sure your .vrm filename has no spaces or special characters

use your itch.io account email address to login into the app

Thank you! I actually spent quite a bit of time trying to implement Qwen3-TTS (which is genuinely impressive), but it requires a high-end NVIDIA GPU to achieve acceptable latency for real-time usage — on CPU, it takes around 30 seconds to generate just 3 seconds of audio. I'll hold off on adding it until a DirectML streaming solution becomes available.

yes you can use kokoro, pocket tts, orpheus and voicevox

Yes you can run local vision models through LM Studio and Ollama

There already is spanish support with Kokoro, voices have an english accent but use the spanish pronunciation

Join discord to get help

Usually this happens on low-end hardware or laptops
I made a "lite" version for this specific case, join discord to get the link

That never happened to anyone before, are you sure it's not a third party problem

Load a model inside LMStudio and make sure the LMStudio server is running

Thank you for your support ! No worries let your current free license expire, and next time you see the license popup simply use your itch.io email address to activate your lifetime license

Hi, yes the lifetime license includes all future updates.
For now, updates are delivered manually, but you’ll always have access to the latest version as it releases.

I did not automate this specific case (yet), so you may need to join discord and DM me a screenshot of the purchase. I already upgraded your free license to lifetime ty so much for your support :)

Yeah it's a bit confusing but to put things simply :
- Early access here on itch.io is lifetime license
- Patreon has free trial (14 days) then it's a monthly plan (3$)

Pour l'instant la vf est bof, le model multilingue de reconnaissance vocale (whisper) est très moyen en français et la seule voix fr dispo sur Kokoro "siwis" est horrible, vraiment. MAIS j'y travaille les prochaines updates vont améliorer ces deux points :)

Hello, it somehow is but i would not recommend, there is actually no really good french voice unless you use premium TTS providers like Azure or ElevenLabs. It's definitely on my roadmap tho.