Warashi v0.1.4

Warashi v0.1.4

Hotfix for no voice on Windows.

• Fixes "[WinError 2] ... error loading or converting generated audio file to wav" / no sound on Windows. The default voice produces an mp3, which the app decodes with a bundled ffmpeg. But the decoder also ran an optional "probe" step that needs a separate ffprobe tool we weren't bundling — on a clean Windows machine that step failed and killed the whole voice output. The probe is now optional: if it can't run, the app just decodes the audio directly. The default voice now works out of the box on Windows with no manual install.

If you're on Windows and got "[WinError 2]" or had a reply with text but no sound, grab v0.1.4. Includes everything from v0.1.2 / v0.1.3 (bundled ffmpeg, graceful handling when the optional translation service isn't running, one-click local-model download, editable default character, attach images/documents).

Warashi v0.1.3

Warashi v0.1.3

Hotfix on top of v0.1.2:

• Fixes "Error processing response: local variable 'req' referenced before assignment" / no voice. The audio-translation path now degrades gracefully — if the optional DeepLX translation service isn't running, it falls back to the original text instead of breaking the whole reply.
• Cross-language voice translation is now OFF by default. It's an opt-in advanced feature (speak in a different language than the on-screen reply). The default setup no longer tries to reach a DeepLX server you don't run, so there's no more "connection refused" on a fresh install.

If you downloaded v0.1.2 and saw the "req" error or had no sound, grab v0.1.3. Includes everything from v0.1.2 (bundled ffmpeg so the default voice works out of the box, one-click local-model download, editable default character, attach images/documents, etc.).

Warashi v0.1.2

Warashi v0.1.2

Changes since v0.1.1:

Voice now works out of the box — even on Windows.
A static ffmpeg is bundled (via imageio-ffmpeg), so the default edge-tts voice no longer needs a manual ffmpeg install. This fixes the silent-voice "[WinError 2] … ffmpeg not found" error on Windows (macOS/Linux too if ffmpeg wasn't installed).

Easiest free brain.
The first-run wizard can now one-click download the recommended local model (qwen2.5:3b) — no API key, no terminal. Local Ollama is the recommended free default; Ollama Cloud and free hosted APIs are documented as options.

New in Settings:

• "About the player" — a global prompt telling every character who you are and how to address you.
• Web search / tools (experimental) toggle — note: only works with an LLM that supports tool-calling.

Other improvements:

• The default / preset character is now editable in the app.
• The attach button is always available — attach images, and now also text documents (.txt / .md / .html).
• Server-set default background option.
• Security hardening: path-injection fixes; errors no longer leak stack traces to the client.
• Default host is 127.0.0.1 (avoids localhost/IPv6 connection failures on some systems).
• Docs: corrected LLM setup, README available in 5 languages.

Download Warashi-v0.1.2.zip below, unzip, and follow the README.

This project is built on top of Open-LLM-VTuber (MIT). Default characters are Live2D Inc. sample models used under the Live2D Free Material License. No copyrighted anime/game characters are bundled.

Warashi v0.1.1

Warashi v0.1.1.

Highlights since v0.1.0:

• 5-language UI (English / 繁體中文 / 日本語 / 한국어 / 简体中文) plus a single Reading language that drives subtitle + chat history + memory; the spoken voice follows each character's own voice pack.
• Mobile/tablet: floating mic + raise-hand and zoom/move controls, iOS safe-area, tap-to-look, a phone sidebar opener.
• Chat: attach images; icon-only corner status badges.
• Fixes: settings hot-apply on character re-select, correct character display name, translator config writes, fresh-install robustness.

How to run: download Warashi-v0.1.1.zip, unzip, double-click start-companion.command (macOS) / start-companion.bat (Windows). The first launch downloads a ~1GB speech-to-text model — one-time, a few minutes, leave the window open. Bring your own LLM (a cloud API key or a local Ollama model).

Warashi v0.1.0

First public release of Warashi — a free, open-source desktop AI companion with a Live2D avatar.

Features: long-term memory, proactive conversation, natural barge-in voice chat, character manager (with skin thumbnails), first-run setup wizard, and a fully bilingual (English / 繁體中文) UI. Bring your own LLM — a cloud API key (OpenAI / Claude / Gemini) or a local Ollama model.

Download & install

1. Download Warashi-v0.1.0.zip below and unzip it.
2. Double-click the launcher: start-companion.command (macOS) or start-companion.bat (Windows).
3. Your browser opens to a setup wizard — add your LLM, then chat.

See the README for full setup, the macOS Gatekeeper / Windows SmartScreen note, and how to add your own characters.

Built on Open-LLM-VTuber.