Overview
Mirai is a Live2D AI companion that reacts during streaming LLM chat. The character is the interface, with expressions changing mid-sentence, the mouth following the real audio waveform, and the interface color reflecting the character's mood. The model marks up its own answer with inline [[emotion]] directives, which are parsed as they stream and dispatched at the exact character offset they were written for. Speech audio is routed through an AnalyserNode and the loudness drives the Cubism mouth parameter every frame, with a noise gate, asymmetric attack/release, and a decaying peak ceiling. The app boots in Rehearsal Mode with a built-in script but live rig, streaming, directive parser, expression engine, lip sync, and browser voice. To connect a real model, add one Vercel AI Gateway key. It supports nine emotions, real lip sync, voice in and out, four personas, five rigs with two runtimes, reactive theming, on-device memory, and zero-config fallbacks.
Pricing
Free
Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.
Visit WebsitePrices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.
Use Cases
- Developers use Mirai to run a Live2D AI companion locally without any API key or account.
- Users use Mirai to connect a real LLM model via a single Vercel AI Gateway key for chat, neural speech, and transcription.
- Developers use Mirai to switch between different LLM providers at runtime through a dropdown.
- Users use Mirai to interact with a character that expresses nine emotions mapped to rig expression files or hand-tuned Cubism parameter poses.
- Users use Mirai to speak to the character via browser SpeechRecognition or server-side Whisper transcription.
- Developers use Mirai to add their own Live2D character by dropping it in public/models/ and adding an entry to the catalog.
- Users use Mirai to have on-device memory where facts are sent with each request and stored only in the browser.
Features
Streaming LLM chat responses
Chat interface with Cubism character
Expressions change mid sentence
Mouth follows actual audio waveform
Interface colors reflect character emotions
Reviews
No reviews yet. Be the first to share your experience!