Not a chatbot. A companion with real runtime, wallet-aware context, and developer-grade infrastructure behind her.
Haruka is rendered in real-time using the Live2D Cubism engine. Her expressions are not pre-animated loops - they are generated dynamically from your input, your voice, and the flow of conversation.
The physics engine runs at up to 60 FPS, calculating hair, ribbon movement, and fabric motion directly inside the browser.
Talk to Haruka and she talks back. Voice input is captured through the microphone, transcribed in real-time, and turned back into audio as her reply.
The full round-trip from mic input to voice output happens in under 2 seconds.
Haruka is built on the Open Souls Soul Engine. She uses WorkingMemory, MentalProcesses, and cognitiveSteps so she can think before she speaks and stay consistent across sessions.
She remembers you, stays in character, and keeps a more stable response posture over time.
Haruka is not locked to one AI model. Browser Local, OpenAI, OpenAI Compatible, Claude, DeepSeek, Gemini, Ollama, Aliyun NLS, Web Speech API, Comet API, and more all run through the same provider layer.
Switch anytime from the settings panel without breaking the companion experience.
Haruka is accessible via API. Any developer, any app, any platform can send a message and receive text, emotion, and memory context back.
POST /chat turns Haruka from a website into a service.
One line of code and Haruka shows up on another website with her Live2D character, voice, memory, and chat interface.
The current brief points to Atelier as the first live external integration.
HARUKA now uses x402 on Solana mainnet for `api-client` traffic. Developers can pay per call, settle on-chain, and unlock premium HARUKA responses without subscriptions or manual billing.
HARUKA can now read a connected wallet snapshot, surface SOL, USDC, and $HARUKA balances, and carry that context directly into chat.
The Utility page keeps the flow browser-native: connect wallet, capture snapshot, then open HARUKA in portfolio-aware chat mode.
Run AI inference entirely inside your browser using WebGPU. No server, no API key, and no conversation data leaving your device.
Your conversation stays private by default.
Haruka speaks English and Japanese. Toggle between EN and JP on the main screen, then practice naturally with a patient and encouraging companion.
Access Haruka from web, desktop, and mobile. She can be installed as a PWA for a native-app feel without going through an app store.
Haruka is planned to see your game in real-time, react, comment, and strategize alongside you using a screen-capture to vision-model pipeline.
Desktop mode is planned to capture the active window so Haruka can understand what you are working on and respond with that context in mind.
Say "swap 10 USDC to $HARUKA" inside the chat and Haruka can quote price via Jupiter, ask for confirmation, then execute the swap inside the conversation flow.
Phase 1 covers USDC and SOL into $HARUKA. Phase 2 expands to all Jupiter pairs.