CompNav · Open-source projects
Artemis
A self-hosted AI girlfriend project combining local characters, Live2D, a desktop pet, voice, and chat bots.
What it is
Artemis is a self-hosted project centered on local AI girlfriend characters and multi-character interaction. It combines character chat and memory, Live2D avatars, a desktop pet, speech synthesis, and entry points such as QQ, Telegram, and WebChat. Users can switch characters with separate memories and import SillyTavern character cards. It is not a single client; it is a system assembled from services including OpenClaw, llama.cpp, GPT-SoVITS, and ComfyUI.
Core mechanisms and capabilities
Characters can be switched while retaining separate chat and memory data. Live2D presents emotional expressions, while the desktop pet supports voice and proactive reminders. Character cards can be imported from SillyTavern, and WebChat, QQ, and Telegram provide different interaction entry points.
Editorial take
It merits inclusion because it brings character companionship, voice, desktop presence, and multi-character management into a runnable project with recent community attention. Its scope is broader than a typical chat frontend and requires managing several local services. Hardware figures in the README describe the author's own setup, not guaranteed requirements for every device. GitHub does not identify a standard SPDX license, and users can review the repository terms themselves.
Who it suits
Suited to users who want local character chat extended into voice, desktop avatars, and multiple messaging channels and can manage models, TTS, image generation, and platform connections.
Deployment and technical requirements
Artemis runs as a set of cooperating local services, with WebChat and optional QQ or Telegram access. Its README and usage guide explain launch paths, service roles, and memory or character settings. Users choose and run the required components; this is not a single standalone installer.
Runtime requirements
The documented main path uses Windows or Linux, requires setting paths and language, and involves services such as OpenClaw, llama.cpp, and GPT-SoVITS. QQ and Telegram entry points require their own bot configuration. The README says default scripts target NVIDIA GPUs; other hardware paths need separate verification.
The main path uses local models and local voice or image services, including llama.cpp, GPT-SoVITS, and ComfyUI. Setup depends on hardware, model choice, and service startup order. The README describes an 8 GB VRAM example using hot-swapping, not a universal minimum for running every feature simultaneously.
Limitations and considerations
The system has many dependencies, and model inference, voice, image generation, and Live2D compete for resources. The author's 8 GB VRAM hot-swapping approach is a specific implementation strategy, so users still need to adapt it to their hardware and models. Character, voice, and Live2D assets may have third-party terms, and the repository does not show a standard SPDX license.
Official sources and further reading
Information checked: Oct 26, 2026.