Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Creative & Roleplay

Open-LLM-VTuber Review: Self-Hosted Voice Chat with a Live2D Avatar

·7 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

Open-LLM-VTuber (version 1.2.1) is a free, open-source voice chat companion with a Live2D avatar that you install and run yourself on Windows, macOS or Linux. According to the project it can run fully offline when every component is local, but it needs a terminal-based setup and the repository has had no functional change since 11 February 2026.

Key Takeaways

  • Maintenance: the newest release is from 26 August 2025, and the README says the maintainers are concentrating on a full v2.0 rewrite and ask for no v1 feature requests.
  • Cost: no paid tier; the README links a donation page.
  • License: MIT for the code; the bundled Live2D sample models fall under a separate Live2D license.
  • Setup: the backend is installed from a terminal; the Electron desktop client is only a front end.
  • Components: speech recognition, language model and voice are separate choices, so one setup can mix local and cloud parts.

📍 In One Sentence

Open-LLM-VTuber is a self-hosted program that connects a speech recognizer, a language model and a text-to-speech voice to an animated Live2D character, so you can hold a spoken conversation with it.

💬 In Plain Terms

You start a small server on your own computer, open its page in Chrome or the desktop client, and talk to a cartoon character; which parts run on your machine and which call a cloud service is up to you.

⚠️Warning: No functional update since 11 February 2026. The latest release is 26 August 2025, and the only newer commit (15 May 2026) edited the README. The project may be unmaintained: check the repository before relying on it.

📌Note: This review is based on the project website, the GitHub repository and its release pages, checked on 24 October 2026. PromptQuorum has not installed or run the software, so audio quality, latency and avatar behavior are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

Open-LLM-VTuber is a free, open-source project that lets you talk to a language model by voice while a Live2D character answers on screen. You run it yourself on Windows, macOS or Linux and pick the speech recognition, language model and voice components. This review is based on the project website, the GitHub repository and its release pages, checked on 24 October 2026; PromptQuorum has not tested the software hands-on.

Open-LLM-VTuber Review: Self-Hosted Voice Chat with a Live2D Avatar

What Is Open-LLM-VTuber?

Open-LLM-VTuber is a voice-interactive AI companion: you speak, a language model answers, and a Live2D character reacts with expressions and a synthesized voice. The repository describes it as talking to any LLM with hands-free voice interaction and voice interruption, running locally across platforms. Its MIT-licensed code is copyrighted to Yi-Ting Chiu in the LICENSE file, and the project is run as a community on GitHub.

The name comes from the goal of recreating the closed-source AI VTuber neuro-sama with open components that run offline on systems other than Windows. Today it is used as a general companion or character chat, with a web interface and an Electron desktop client that has a window mode and a desktop pet mode.

Get It

Open-LLM-VTuber is installed as a backend server on Windows, macOS or Linux; the desktop client and the browser page both connect to it.

Option
How to get it
Backend (all systems)Open-LLM-VTuber-v1.2.1-en.zip from the GitHub Releases, or git clone with the recursive flag
Desktop client, Windows.exe installer from the client releases; needs the backend running
Desktop client, macOS.dmg from the client releases; needs the backend running
Websiteopen-llm-vtuber.github.io
Privacy policyNo privacy policy page was found on the website or in the repository
Source codegithub.com/Open-LLM-VTuber/Open-LLM-VTuber, MIT license

Version reviewed: 1.2.1, published 26 August 2025 per the GitHub Releases page; the desktop client 1.2.1 appeared on 21 August 2025. A terminal and developer tools (Git, FFmpeg, uv, Python 3.10 to 3.12) are required for the backend, so this is not a double-click install. The quick-start docs warn that the unsigned desktop client may trigger operating-system security warnings. The app is free.

Requirements

The project sets no fixed minimum; what you need depends on which components you run locally and which you call over an API.

Minimum

What the sources say:
If every heavy component (speech recognition, language model, voice) uses an API, the quick-start says even a Raspberry Pi can host it

Local use

What the sources say:
Recommended: a Mac with Apple silicon, an Nvidia GPU, a recent AMD GPU, or a CPU strong enough to replace one

Tools

What the sources say:
Git, FFmpeg (required) and uv; Python 3.10 or newer but below 3.13; CUDA Toolkit and cuDNN for Nvidia GPU use

Browser

What the sources say:
Chrome only is recommended; Edge and Safari have known issues

Remote access

What the sources say:
Using the server from another device needs HTTPS, because browsers only allow the microphone in a secure context

These are the project's own statements; PromptQuorum did not measure memory use or speed on any hardware.

Features Confirmed by the Sources

Every item below comes from the website, the quick-start guide or the README; none has been independently tested.

  • Voice conversation. Real-time speech input, voice interruption without headphones, optional AI-initiated speech and a display of the AI's inner thoughts that is not spoken aloud.
  • Live2D avatar. Cubism 3 to 5 models, expression mapping controlled from the backend, touch interaction and your own imported models.
  • Vision and tools. Camera, screen and screenshot input, MCP tool calls, and, per the website, browser control.
  • Pluggable models. Language models via Ollama, any OpenAI-compatible API, Gemini, Claude, Mistral, DeepSeek, GGUF, LM Studio or vLLM; recognition via sherpa-onnx, FunASR or Whisper variants; voices via sherpa-onnx, Coqui-TTS, Bark, Fish Audio, Azure and others.
  • Streaming and groups. A built-in Bilibili live-chat client, group chat with several AI characters and saved chat history.

Local or Cloud: Data Flow

The project says everything can run offline with local models and that conversations then stay on your device; it publishes no privacy policy, and the pages we read do not mention telemetry. In a fully local setup (for example Ollama for the model and a local recognizer and voice) audio and text are processed on your computer. The software itself is open source, so the claim can be checked in the code.

Each component you switch to a cloud option sends data to that provider. The supported cloud choices include OpenAI-compatible APIs, Gemini, Claude, Groq Whisper and Azure, and the quick-start example uses Edge TTS for the voice, which needs an internet connection. Models and dependencies are downloaded during setup, so the first install is online.

📌Note: These are the maintainers' declarations, not audit results. PromptQuorum has not read the code or inspected network traffic, and a setup is only as private as its least private component.

Who Should Use It

  • Tinkerers and developers who want to wire their own speech, language and voice components to an avatar.
  • Streamers and character-chat fans who want a talking Live2D character they control, including a desktop pet.
  • Privacy-minded users who are willing to run every component locally and verify it themselves.

What We Could Not Verify

  • Quality and speed. PromptQuorum did not measure latency, recognition accuracy, voice quality or the hardware needed for a smooth conversation.
  • Maintenance outlook. The README says a full v2.0 rewrite is in early planning and asks for no v1 feature requests; whether v1 will receive further fixes is not stated.
  • Long-term memory. The website describes Letta-based memory, the README says it is temporarily removed; which statement holds for version 1.2.1 was not checked.
  • Commercial use of the avatars. The README says the Live2D sample models may need extra permission from Live2D Inc. for medium or large companies; other models carry their own licenses.
  • Desktop client signing. The docs say the client is unsigned and may trigger security warnings; whether the installers are otherwise safe was not checked.

Competitors and Alternatives

SillyTavern

Platforms:
macOS, Windows, Linux, web
Price / license:
Free / AGPL 3.0
Key difference:
Text-focused roleplay frontend built around lorebooks and persona management
Articles about SillyTavern (9)

Also mentioned in:

+7 more not shown

Backyard AI

Platforms:
macOS, Windows
Price / license:
Freemium / Closed source
Key difference:
Desktop character-chat client with a built-in Character Hub and one installer
Articles about Backyard AI (4)

Also mentioned in:

These are same-category apps from the PromptQuorum directory; confirm each app's current price, license and platforms on its own page.

Compare in directory

Frequently Asked Questions

Can I run it without a GPU?

Yes, if you send the heavy parts to APIs: the quick-start says each component can use an API, and a lighter local model or a CPU-friendly recognizer is possible. Local use is recommended on Apple silicon, an Nvidia GPU or a recent AMD GPU.

Can I use my own Live2D model?

Yes: put the model folder in live2d-models, register it in model_dict.json and select it in the character configuration. Cubism 3 to 5 models work, Cubism 2 does not, and you should check the model's own license.

Can I talk to it from my phone?

The project says you can run the server on one computer and open it from another device, but the browser microphone only works over localhost or HTTPS, so remote use needs an HTTPS setup.

Can it join a live stream?

It has a built-in Bilibili live-chat client that receives viewer messages; other platforms need an interface you implement yourself.

Which browser works best?

The quick-start recommends Chrome only and names known problems with Edge and Safari.

Verdict

Open-LLM-VTuber is a free, open-source way to build a voice-driven Live2D companion from components you choose, with a documented route to running everything offline. Against that, setup is a developer-style install, the example configuration still uses an online voice, the repository shows no functional update since 11 February 2026, and none of its quality claims could be checked here. It suits tinkerers who can use a terminal and are happy to treat it as a project that may no longer be maintained; anyone who needs a supported, polished app should test it carefully first.

Sources

← Back to Power Local LLM

MCP