AIRI: The Open-Source Self-Hosted AI Companion to Build Your Own Cyber Neuro-sama

Project Overview

After AI VTuber Neuro-sama exploded in popularity, more and more developers started exploring how to build their own controllable AI companions. The open-source project moeru-ai/airi is a standout example of this wave — it has already earned over 42,000 stars and 4,200+ forks on GitHub, with as many as 110 new stars added in a single day, reflecting impressive momentum.

AIRI's positioning is clear and direct: a self-hosted, you-owned AI companion system. In the project's own words, it is "a vessel for the soul of a waifu and cyber life," with the goal of bringing virtual characters truly into reality and achieving the same depth of interaction as Neuro-sama.

What Is Neuro-sama

For readers unfamiliar with the background: Neuro-sama is an AI-powered virtual streamer capable of real-time conversation, playing games while interacting with viewers, and has built a large and loyal fanbase on Twitch through its humanlike performance and distinct personality. It proved that AI characters can do far more than chat — they can participate in complex, real-time interactive scenarios.

The technology behind Neuro-sama represents an important milestone for AI VTubers. Its developer, Vedal987, deeply integrated large language models, real-time text-to-speech (TTS), automatic speech recognition (ASR), and streaming platform APIs to achieve a millisecond-level response loop. This architecture requires the system to simultaneously handle chat input, game state awareness, and output generation — placing extreme demands on latency control. VTuber (Virtual YouTuber) culture emerged in Japan around 2016, using Live2D or 3D models initially driven by real humans. Neuro-sama pushed this format into a new era of truly AI-driven autonomous characters, marking a deep fusion of AI and anime virtual character culture. AIRI's goal is to open-source this capability, so anyone can own an AI character of their own.

Core Features

AIRI is not an ordinary chatbot — it is a complete system with multimodal, real-time interaction capabilities.

Real-Time Voice Conversation

The project natively supports real-time voice chat, allowing users to converse with the AI character as naturally as with a real person, completely eliminating the jarring "type → wait for reply" experience. Low-latency voice interaction is key to creating a genuine sense of companionship and is the core technology for closing the psychological distance between humans and AI characters.

Game Interaction Capabilities

AIRI has the ability to actually play games — the project officially supports Minecraft and Factorio. Having an AI truly "play" a game rather than merely "talk about" it is far more technically complex than ordinary conversation. In Minecraft, for example, the AI needs to understand the environment through visual input or structured game-state APIs, decompose high-level goals (such as "build a house") into specific action sequences, and handle environmental feedback during execution — a domain that sits at the intersection of Embodied AI and reinforcement learning. Factorio demands even higher planning capabilities, as its automated pipeline design is essentially a constraint-satisfaction and optimization problem. Research projects like OpenAI's "Universe," DeepMind's "Agent57," and "GROOT" designed specifically for Minecraft have all explored similar directions in recent years, but combining this with real-time role-playing and audience interaction remains a significant engineering challenge. Having an AI character "actually play" is what distinguishes AIRI from ordinary conversational agents, and brings it much closer to Neuro-sama's form of playing games while live-streaming and interacting in real time.

Full Platform Support

On the deployment side, AIRI supports Web, macOS, and Windows. Cross-platform design significantly lowers the barrier to entry — whether you want to quickly try it in a browser or run the full system on a local desktop, there's an option for you.

Tech Stack and Open-Source Ecosystem

AIRI uses TypeScript as its primary development language, making it a natural fit for the web ecosystem and facilitating development of front-end interfaces, real-time communication, and cross-platform desktop applications (such as Tauri-based solutions).

Choosing TypeScript as the primary language reflects AIRI's web-first design philosophy. TypeScript introduces a static type system on top of JavaScript, greatly improving code maintainability for large-scale collaborative projects — particularly important for an open-source project with 4,200 forks and many contributors. Tauri is a rapidly growing cross-platform desktop application framework. Compared to Electron, it uses Rust as the backend and the system's native WebView as the rendering layer, resulting in final binary sizes less than one-tenth of an equivalent Electron app and significantly lower memory usage. For an AI companion application that needs to run in the background long-term and continuously communicate with local AI models, Tauri's lightweight nature has practical significance. The combination of TypeScript frontend and Tauri/Rust backend is becoming a new paradigm for building next-generation desktop applications that prioritize performance and user experience.

As a fully open-source self-hosted system, AIRI's significance stands out especially in today's context. Self-hosting in the AI application context means running inference, data storage, and character state entirely on hardware the user controls — not relying on cloud API calls. From a technical implementation standpoint, this typically involves local LLM deployment (e.g., via Ollama or llama.cpp runtimes), local TTS engines (e.g., VITS, GPT-SoVITS), and vector databases (for character memory management). The core trade-off of self-hosting is sacrificing the convenience of cloud compute in exchange for data sovereignty, reduced network latency, and freedom from API rate limits. With consumer-grade GPU performance rising rapidly, running models in the 7B–14B parameter range on personal computers has become practical, which is also the hardware prerequisite that makes projects like AIRI viable. While mainstream AI companion products mostly rely on the cloud with user data controlled by vendors, AIRI lets you run the entire system on your own device and truly own your data and characters. This protects privacy and gives developers ample room for customization — you can freely shape the character's personality, integrate different LLMs, or extend entirely new interaction scenarios.

Significance and Reflections

From Tool to "Companion"

The direction AIRI represents reflects a broader trend: AI applications are extending from pure productivity tools toward emotional companionship and entertainment scenarios. When large models are capable enough in conversation, voice synthesis is natural enough, and multimodal interaction is fluid enough, building an AI character with personality, interactivity, and the ability to participate in daily activities becomes genuinely within reach. The 42,000+ stars clearly demonstrate that demand for this kind of experience is substantial among developers and users alike.

The Core Value of Self-Hosting

In an era of proliferating commercial AI companion products, AIRI's commitment to the "you own it" philosophy is especially valuable. Users don't have to worry about characters being taken down by the platform, data being collected, or subscription fees quietly increasing — everything stays in their own hands. This decentralized, user-sovereignty-first approach is likely to become an important trend in personal AI applications.

Challenges Still to Overcome

Of course, truly reaching Neuro-sama's level is not something that happens overnight. Long-term personality consistency and memory management are among the most difficult technical problems in AI companion systems. Current mainstream large language models are limited by context window length and cannot natively "remember" conversation details from weeks ago. The industry typically uses RAG (Retrieval-Augmented Generation) architecture to address this: historical conversations, user preferences, and key events are stored as vectors in a local database (e.g., Chroma, Qdrant), and relevant memories are dynamically retrieved and injected into the prompt during each conversation to simulate long-term memory. Personality consistency is maintained through carefully designed system prompts and character cards — the RP community has established de facto standards such as V2 character cards. How to keep a character's core personality stable while it continues to "grow" is a central challenge that AIRI and similar projects continually iterate on. Additionally, self-hosting itself carries a certain deployment and compute barrier for ordinary users. But as an active open-source project, AIRI provides an open experimental ground, allowing the community to collectively push the boundaries of AI companion technology.

Summary

moeru-ai/airi is an ambitious open-source AI companion project built on the core principles of self-hosting and full user ownership, combining real-time voice conversation with game interaction capabilities, and supporting Web, macOS, and Windows. For developers interested in AI characters, virtual companions, or Neuro-sama-style VTubers, AIRI is undoubtedly an open-source benchmark worth deep attention — it not only showcases the imaginative potential of personal AI applications, but also provides solid technical practice toward the vision of "bringing AI characters truly into our lives."