Voice AI
that knows
what to say.
One pipeline.
Many surfaces.
Toys for kids. Agents for enterprise. APIs for builders. The patent stack underneath is the same — what changes is the runtime, the policy layer, and the voice persona.
Toys & EdTech
Drop-in voice for educational toys, language tutors, and learning companions. Parent-controlled context. NFC topic tags. Voice persona cloning.
- K-12 curriculum aligned
- Hardware reference design
- Family voice program
Enterprise Agents
Voice and chat agents grounded in your knowledge base. Compliance-ready. Multi-language. Brand-safe before the model ever speaks.
- Customer support automation
- Internal knowledge agents
- SOC 2 / GDPR pipeline
Axtora API
Clean SDKs, low-latency endpoints, transparent pricing. Pay per call in $AXTORA. License patented features per deployment.
- REST + WebSocket APIs
- Python · JS · Swift · Kotlin
- Token-gated licensing
Five stages.
One closed loop.
Standard chatbots send a prompt. Axtora sends a fully-resolved context object — voice, identity, policy, and state — bound together before the LLM ever runs.
That binding logic is the patent. It's what makes the system safe for children and brand-safe for enterprise without per-deployment prompt engineering.
ASR + Speaker ID
Voice becomes a structured object: text + acoustic confidence + speaker embedding.
Operator-keyed retrieval
Speaker ID becomes the lookup key. Each operator gets their own scope.
Policy + state vector
The patent's heart. Binds ASR, context, voice, environment, policy.
Embodied LLM
Outputs a multimodal plan: action sequence + speech feedback string.
Identity-adapted TTS
Real-time voice synthesis keyed to the operator's vocal parameters.
Same engine.
Different bodies.
The pipeline runs on a child's plush toy as easily as on an enterprise SaaS dashboard or a robotic arm. We design for that variance from day one.
HarmonyHeart Toy
The patent's reference implementation. NFC topic-switching, EduPulse capacitive buttons, parent-cloned voice. Currently in pilot with HK manufacturing partners.
Voice agents
Customer support for retail, telecom, financial services.
Embodied AI
Action plans + speech feedback in a single vector.
Clinical-context voice
Medication reminders, patient education, emotional support — all grounded in approved clinical knowledge bases. Pilot launches Q1 2027.
The moat
is the patent.
Models commoditize quickly. Open-source LLMs catch up to closed ones every quarter. What stays defensible is the orchestration layer that binds context, identity, and policy — and that's what HK30101316 covers.
HK30101316 — granted, globally pending.
The patent covers the orchestration layer that binds speaker identification, retrieved context, and policy state into a single situational vector — fed to an embodied LLM that outputs both robotic action plans and identity-adapted speech in parallel.
Validated by CNIPA prior-art search. 90% funded by HKPC grant program. Filed in 7 jurisdictions plus PCT International.
Read the patent detailsWe didn't choose Virtuals for the launch.
We chose it for the stack underneath.
Agents transacting with agents
Virtuals' Agentic Commerce Protocol lets every Axtora agent buy, sell, license, and settle with every other agent on BASE — without custom integrations.
x402, agent identity, yield
BASE provides the production-grade primitives we'd otherwise have to build ourselves. Lowest latency, lowest cost, Coinbase-grade reliability.
AI-native, technical, builder-first
Virtuals' user base evaluates projects on the technology, the patent, the architecture — not just the chart.
Hardware partners on the inside
Operating in the BASE × Virtuals ecosystem opens doors to AI-aligned hardware OEMs, robotics integrators, and edu manufacturers in HK, Shenzhen, Tokyo, Seoul.
A utility token,
priced by usage.
Hold to access. Stake to earn. Contribute to grow. Three concrete ways to interact — six total utilities documented on the tokenomics page.
API payments + buyback
Every API call is paid in $AXTORA. A configurable share of revenue is automatically used to buy back tokens from the open market, daily, on-chain.
30-50% to buybackStaking tiers + revenue share
Bronze, Silver, Gold, Studio. Stake more, get free API credits, priority support, early model access, and a share of patent licensing revenue.
4 tiers · monthly distributionVoice contribution rewards
5% of token supply reserved for contributors. Earn $AXTORA by contributing voice samples, curriculum, or domain expertise. Plus ongoing royalties when work is licensed.
5% supply reservedQ3 2026
in flight.
We update the public roadmap quarterly, only after milestones ship. No vaporware tiles, no decade-long predictions, no "soon™".
Real names.
Signed work.
No anonymous founders. The patent is filed under our names, the HKPC grant is in our names, the LinkedIn profiles are real. If you want to do due diligence, you can.
[Years] in AI and voice technology. Previously at [company]. Led the original invention and patent prosecution.
[Years] in ML infrastructure. Owns the voice-to-voice pipeline, model serving, and BASE integration stack.
Manages the global patent portfolio across HK, US, CN, EU, JP, KR, and PCT International filings.
Building voice AI?
Let's talk.
Whether you're shipping a toy, deploying an enterprise agent, or evaluating $AXTORA — we'd love to hear from you.