Haruka

Haruka is a 3D AI companion with real-time voice, WebGPU accelerated local inference, and expressive Live2D motion. Connect 20+ language model providers, install it anywhere as a PWA, and let your agent pay for premium tools on its own through x402 on Robinhood Chain.

About Haruka

WearebuildingHaruka,a3DAIcompanion.Builtforreal-timevoice,privatelocalinference,andexpressivepresence.

Haruka runs the model, the voice pipeline, and the render loop as close to you as possible. Speech recognition and synthesis happen in real time, WebGPU keeps inference on your own device, and the character reacts to touch and tone instead of waiting for a text box. When your agent needs a paid tool, x402 on Robinhood Chain lets it settle per action without a subscription or an account.

Acompanionthatlistens,speaksandreacts.
Runsonyourdevice.Paysforitsowntools.

Real-time voice. (01)

  • Natural voice interaction
  • Advanced speech recognition
  • Low latency text to speech
  • Interrupt and talk over it

WebGPU acceleration. (02)

  • High speed local inference
  • Completely private on device
  • No prompt leaves your machine

Live2D expressions. (03)

  • Highly expressive 3D models
  • Reacts to touch and gaze
  • Motion driven by your voice

20+ LLM providers. (04)

Connect OpenAI, Claude, DeepSeek, Gemini, Mistral, Groq, Ollama and more with your own keys.

PWA and cross-platform. (05)

A responsive web interface you can install like a native app on desktop, tablet and phone.

x402 payments. (06)

Your agent gets a native pay-per-action path to premium tools, paid workflows and external APIs through the HTTP 402 standard on Robinhood Chain.

How it works

Three steps, no setup friction.

01

Pick your model.

Run a local model with WebGPU or connect any of 20+ providers such as OpenAI, Claude and DeepSeek with your own key.

02

Start talking.

Speech recognition and text to speech run in real time, so the conversation flows without typing or waiting.

03

Let it act and pay.

Haruka calls premium tools and paid APIs on demand, settling per action over x402 on Robinhood Chain.

Model providers

20+ providers, or nothing but your own device.

Cloud providers

OpenAIAnthropic ClaudeDeepSeekGoogle GeminiMistralGroqxAI GrokTogetherFireworksOpenRouterPerplexityCohere

Local and private

WebGPU in browserOllamaLM Studiollama.cpp servervLLM

By the numbers

Haruka in four figures.

20+

LLM providers ready to connect

100%

Local inference option with WebGPU

60fps

Expressive Live2D character loop

402

HTTP standard behind pay-per-action tools

Roadmap

A deliberate path to a living companion.

Phase 01

Companion core.

  • Real-time voice loop
  • Live2D character rig
  • Provider key management
Phase 02

Local first.

  • WebGPU inference
  • On device speech models
  • Offline PWA install
Phase 03

Agentic tools.

  • Tool calling
  • x402 pay-per-action
  • Robinhood Chain wallet link
Phase 04

Open ecosystem.

  • Character marketplace
  • Memory and long context
  • Third party tool registry

FAQ

Questions, answered.

Haruka is a 3D AI companion. It combines real-time voice interaction, an expressive Live2D character, WebGPU accelerated local inference, and connections to 20+ language model providers in one installable web app.

Meet your companion, out loud.

Open the Haruka app, pick a model, and start a real-time voice conversation in your browser.