AIvy

AIvy is a next-generation desktop mascot whose character dialogue is generated by AI on the fly.When you talk to AIvy, or when the character spontaneously speaks up about something happening on your PC, the words aren't pre-written scripts — the AI thinks them up and replies in real time. The generated dialogue is then read aloud by a text-to-speech engine, synchronized to the character's lip movements.[What the AI generates]- The character's dialogue (text)[What the AI does NOT generate]The 3D character itself (VRM model), facial expressions, animations, UI, and sound effects are all pre-authored assets. No images, video, or music are generated by AI during use.[About the AI models]Dialogue generation (default, local inference): AIvy is a local LLM runner built on llama.cpp (the same kind of architecture used by LM Studio, Jan, GPT4All). On first launch you pick from a dropdown the open-source language model best suited to your language — Sarashina 2.2 3B (MIT) for Japanese, Qwen 2.5 3B (Apache 2.0) for the others — and download it once from HuggingFace. From then on inference happens entirely on your PC, and your conversations are never transmitted externally. The in-app catalog only offers models under permissive licenses (Apache 2.0 / MIT).Dialogue generation (optional, cloud LLM): For users without local-LLM-capable hardware, AIvy additionally offers optional cloud LLM integration with OpenAI, Anthropic Claude, and Google Gemini. Network traffic occurs only when the user has explicitly registered their own API key with the chosen provider in the in-app settings. API keys are stored with OS-level encryption (Windows DPAPI) on the user's device and are never sent to any AIvy server (we don't run one). Usage costs are billed directly by the chosen provider. When this feature is not used, no external traffic occurs and inference runs entirely locally.[Streaming / video creation]You may freely stream, record, and share content created with AIvy. Just follow the license terms of the VRM model, each TTS engine, and the language model you actually use — commercial use, monetization, and credit requirements vary per VRM/engine/model.[Things to know]AI responses vary every time, depending on the model you choose, the personality you configure, and what you say. For the rare case of an unexpected response, you can switch personality presets, change models, or pause speech at any time.