Overview
AIRI is an open-source AI companion that gives a chat model a voice and a character body, in a browser or on the desktop.
The project is published by the moeru-ai GitHub organization as Project AIRI, and the MIT license names Neko Ayaka as copyright holder (2024-present). The README describes it as a re-creation of the virtual streamer Neuro-sama, and the repository description calls it a "Self hosted, you-owned Grok Companion". Stable release 0.15.4 came out on 12 November 2026, a pre-release 0.15.5 followed on 14 November, and the latest commit is dated 18 November 2026, so development is active and no stale-project warning applies.
Per the README, the project uses web technologies such as WebGPU, WebAudio and WebSocket. The browser version is meant to show what browsers can do, while the desktop version can, per the README, use NVIDIA CUDA and Apple Metal natively. The docs describe the desktop version as suited to more advanced use such as VTuber streaming, computer use and access to local LLMs.
The README lists what the project can do: play Minecraft, chat in Discord and Telegram, listen through browser or Discord audio, speak through several speech providers and animate Live2D and VRM models. Factorio support is marked work in progress, and roadmap items such as a memory system are still unchecked.
These descriptions come from the maker's README and docs. PromptQuorum has not run the app and has not assessed voice quality, latency, character behavior or resource use.
How to Get It
You can open AIRI in a browser or install the desktop app from GitHub Releases, winget, Scoop or Homebrew Cask.
The README shows download buttons for Windows, macOS and Linux. Stable release 0.15.4 lists a Windows installer (.exe), macOS .dmg files for Apple silicon and Intel, and Linux AppImage, .deb, .rpm, .flatpak and .pacman packages. The release also contains an Android APK, but neither the README nor the docs describe Android as a supported platform, so it is not listed here. The docs offer a pre-release and experimental nightly builds as separate channels.
Platform | How to get it |
|---|---|
| Windows | Installer (.exe) or winget install MoeruAI.AIRI |
| macOS | Disk image (.dmg) or brew install --cask airi |
| Linux | AppImage, .deb, .rpm, .flatpak or .pacman |
| Browser | Open the web version, no install |
| Website | Official site and documentation |
The sources do not say whether the installers are code-signed, so expect possible operating-system warnings and check them before you proceed.
Requirements
AIRI states no hardware requirements, so what you need depends on the providers and models you choose.
With cloud providers most of the work happens remotely. Running a local model through a tool such as Ollama needs hardware sized for that model, which AIRI does not control.
Item | Stated figure |
|---|---|
| Operating system | Windows, macOS, Linux (release files) |
| Browser | Web version available |
| RAM / VRAM | No requirement stated |
| GPU | Optional; README mentions CUDA and Metal |
| Disk | No figure stated |
| Microphone | Needed for voice input (docs) |
Screen and window capture is available only in the desktop version, per the web manual.
Features
AIRI's core is a chat model with a voice and a character body, plus optional integrations. All items below are maker claims from the README and docs.
- Chat with a model from a provider you pick, including OpenAI-compatible APIs, OpenRouter, Ollama, vLLM, Anthropic Claude, Google Gemini and DeepSeek, among others in the README list.
- Voice input from the browser or Discord, with client-side speech recognition and talking detection.
- Speech output through several providers, including ElevenLabs, Microsoft/Azure Speech, OpenAI-compatible text-to-speech, Alibaba Cloud Model Studio and local Kokoro TTS.
- Live2D and VRM character models with automatic blinking, look-at and idle eye movement.
- Playing Minecraft; Factorio is described as work in progress with a proof of concept.
- Chat bots for Discord and Telegram.
- Screen and window capture for vision in the desktop version (docs).
- In-browser memory storage with DuckDB WASM or pglite; the roadmap still lists a memory system and pure in-browser local inference as unfinished.
- A plugin system that the README describes as work in progress.
Privacy and Data Flow
Where your data goes depends on the providers you connect, and no privacy policy was found.
AIRI does not run one fixed model. The docs say you add a chat provider such as OpenRouter, an OpenAI-compatible service, Ollama or Google Gemini with an API key or a local service address, and separate providers for speech, transcription and vision. If those are cloud services, your text and audio can go to them under their own terms; with Ollama on your machine, the model runs there.
The docs also offer signing in to an official AIRI provider as an alternative. The v0.15.4 release notes describe a subscription with a Flux balance for it. They also say the desktop app, web app, mobile app and sign-in pages can send error reports under the same data-collection switch as product analytics. What those reports contain, and how the official provider handles chats, is not stated in the sources PromptQuorum read.
The docs list vision providers and say screen and window capture is available in the desktop version. If you combine them with a cloud provider, screen content would leave your computer.
Who Should Use It
AIRI fits people who want a free, open-source AI character they can configure themselves.
- Hobbyists who want an AI character with voice and a Live2D or VRM body.
- Developers and streamers who want to try game or chat-app integrations and accept early-stage software.
- People who already run Ollama or another provider and want a character front end for it.
What We Could Not Verify
AIRI is a poor fit in some situations, and several details remain unverified.
- Not a fit: users who want a finished, stable product; the README says the project is at an early stage of development.
- Unverified: any privacy policy, terms of service or imprint. None was found on the website, in the docs or in the repository.
- Unverified: the maintainer's location. No source read names a country.
- Unverified: hardware requirements, disk use and performance. None are stated.
- Unverified: the browser version. The page at airi.moeru.ai showed "AIRI could not start" when PromptQuorum fetched it, so it was not inspected.
- Unverified: mobile support. The release lists an Android APK and the README mentions an experimental mobile app, but the docs describe mobile use through the web version.
- Unverified: subscription prices and what the official AIRI provider includes. The release notes mention a subscription and Flux but give no prices.
- Unverified: whether the installers are code-signed.
- Unverified: voice quality, latency, character behavior and game-playing results. All feature descriptions are the maker's own.
- Unverified: whether AIRI works fully offline. The sources do not say so.
- Unverified: independent coverage. No third-party review was read for this entry.
AIRI vs Alternatives
AIRI combines voice chat, character models and game integrations in one project. PromptQuorum has not compared it with the alternatives hands-on. The alternatives below come from AIRI's own "Similar Projects" list and PromptQuorum's reviews.
AIRI
- Noted in sources:
- MIT; browser and desktop; voice, Live2D/VRM, games
Open-LLM-VTuber
- Noted in sources:
- Self-hosted voice chat with Live2D; PromptQuorum review
Articles about Open-LLM-VTuber (1)
- Open-LLM-VTuber Review: Self-Hosted Voice Chat with a Live2D AvatarUpdated October 24, 2026
SillyTavern
- Noted in sources:
- Character-chat frontend with lorebooks; PromptQuorum review
Articles about SillyTavern (10)
- SillyTavern in 2026: The Free Frontend for AI Character Roleplay and Long-Form ChatUpdated November 14, 2026
- RisuAI Review: Open-Source Roleplay Chat ClientUpdated November 14, 2026
- Best Uncensored Local LLMs for Erotica & Creative Writing 2026: Hermes 3 vs DolphinUpdated November 13, 2026
- Agnai Review: Self-Hostable AI Roleplay ChatUpdated October 30, 2026
- Castmates Review: Offline AI Character Chat for iPhoneUpdated October 26, 2026
- text-generation-webui in 2026: How Oobabooga's Local LLM UI Became "TextGen"Updated September 5, 2026
- LibreChat Review: Self-Hosted ChatGPT Alternative for Local LLMsUpdated September 3, 2026
- Backyard AI Review: Local AI Character Chat and RoleplayUpdated September 2, 2026
- Open WebUI vs SillyTavern: Best Chat UI for Local LLMsUpdated August 30, 2026
- SillyTavern vs Agnai vs RisuAI: Best Local Roleplay Frontend (2026)Updated August 29, 2026
+10 more not shown
Amica
- Noted in sources:
- In AIRI's list: "Great at VRM, WebXR"
Neuro
- Noted in sources:
- In AIRI's list: a Neuro-sama recreation
FAQ
Is AIRI free?
The app is free and MIT-licensed. The v0.15.4 release notes describe an optional subscription and Flux balance for the official AIRI provider, but no prices were found, so costs depend on whether you use that provider or your own provider accounts.
Does AIRI run models on my computer?
Not by itself, per the docs: you connect providers, and Ollama is one of the listed chat providers for local models. The README also says the desktop version can use NVIDIA CUDA and Apple Metal natively but gives no details on which models.
Which platforms does AIRI support?
A browser version at airi.moeru.ai and desktop builds for Windows, macOS and Linux, per the release files. The docs describe mobile use as going through the web version.
Can AIRI play games?
The README says it can play Minecraft and lists Factorio as work in progress with a proof of concept. These are maker statements that PromptQuorum did not test.
Is there an official AIRI cryptocurrency?
No, per the README: it says the project has no officially minted cryptocurrency or token and asks users to proceed with caution.
Verdict
AIRI is worth a look if you want a free, open-source AI character with voice and a Live2D or VRM body and can accept early-stage software.
It has an MIT license, installers for three desktop systems plus a browser version, a long list of selectable providers including Ollama, and active development: a stable release on 12 November 2026 and a commit on 18 November 2026.
The trade-offs are real. The README calls the project early-stage, privacy depends on the providers you pick, no privacy policy was found, and hardware needs and quality are unstated. PromptQuorum has not tested the app. Try it first with a local Ollama model or a throwaway provider key, and check which services receive your voice and screen data.
Sources
- AIRI website — web app entry point; the page did not load when checked on 18 November 2026.
- AIRI documentation — project description and navigation, checked 18 November 2026.
- Docs overview — web and desktop versions and their intended uses, checked 18 November 2026.
- Docs versions page — stable, pre-release and nightly channels, checked 18 November 2026.
- Desktop manual — provider setup, Ollama and sign-in, checked 18 November 2026.
- Web manual — providers, microphone and desktop-only screen capture, checked 18 November 2026.
- AIRI GitHub repository — repository description, license and last commit, checked 18 November 2026.
- AIRI README — features, roadmap, install commands, provider list and similar projects, checked 18 November 2026.
- LICENSE — MIT License and copyright line, checked 18 November 2026.
- Release v0.15.4 — release files, subscription and error-report notes, checked 18 November 2026.
