CompNav · Open-source projects
z-waif
A self-hosted companion project combining a local AI character, voice, VTube Studio, memory, and desktop or game modules.
What it is
z-waif aims to help users build an AI character that runs on their own computer. It combines a local language model, voice conversation, a VTube Studio avatar, and memory, with optional modules for Discord, Minecraft, vision, and alarms. Lorebooks or RAG can carry information forward across conversations. The project emphasizes modular assembly and customization rather than a ready-to-use hosted companion.
Core mechanisms and capabilities
Voice conversation can use components such as Whisper and RVC, with VTube Studio driving avatar expressions. Memory can draw on character data and custom RAG. Discord, vision, alarms, and Minecraft are optional modules that need their own integration setup.
Editorial take
The project has a clear local AI companion and virtual-character direction, with concrete scenarios and resource guidance in its README. It merits inclusion as a self-hosted character project. The README recommends a relatively capable GPU, and minimum needs vary with model choice. GitHub does not identify a standard SPDX license, so users should review the repository's terms for themselves.
Who it suits
Suited to people who want to assemble a local language model, speech pipeline, and virtual avatar and are willing to install dependencies and tune character and model settings.
Deployment and technical requirements
The project provides a WebUI, while its core workflow connects a local model with speech and avatar software, then adds modules such as Discord or games as needed. Users download the models and third-party tools and configure APIs, devices, and shortcuts; there is no single installer that covers every capability.
Runtime requirements
The README recommends Windows 10/11 and an NVIDIA GPU, with about 16 GB of VRAM as a reference; support for other GPUs has additional conditions. The project uses tools such as Oobabooga, RVC, Whisper, and VTube Studio, with exact requirements depending on enabled modules.
The project centers on local models, with the README mentioning options such as Oobabooga and LM Studio. Model, TTS, speech recognition, and avatar software are configured separately. The project recommends a higher-VRAM machine; actual load varies with model and enabled features.
Limitations and considerations
The README describes the project as early access and notes that some parts may be unstable or unintuitive. Although the README calls it open source, GitHub does not identify a standard SPDX license. Live2D models, voice models, and external tools may also have their own terms.
Official sources and further reading
Information checked: Oct 26, 2026.