Patented context-aware voice AI, tokenized on Virtuals.

Axtora Labs builds context-aware voice agents: voice-to-voice AI that retrieves curated context before generating a reply, so the response is tied to a knowledge base the customer controls — not to whatever the underlying LLM happened to be trained on. The result is voice AI that stays inside its domain envelope: a toy stays age-appropriate, a clinic-aide stays medically careful, an industrial robot stays safety-first.

The technology is protected by a granted Hong Kong patent (HK30101316), with applications pending in the US, India, China, and PCT international, and EU / Japan / Korea filings due by July 2026. The Hong Kong Productivity Council’s Patent Application Grant covers ~90% of the global patent costs — a government vetting step that required a positive worldwide novelty search.

Access to the technology is licensed via the $AXTORA token, launching on Virtuals Protocol on May 18, 2026.

Voice AI today fails the moment the conversation has stakes. Five concrete failure modes:

1. 1

   No real context awareness

   Most assistants don't know the deployment they're in. A toy talking to a 6-year-old, a clinic aide talking to a patient, and a hotel concierge talking to a guest get treated identically.

2. 2

   Safety drift

   Open-ended LLMs hallucinate, stray off-policy, or produce inappropriate content — especially around children, patients, and other vulnerable users.

3. 3

   Generic personas

   Voice assistants can't reliably hold a brand voice, a domain expert role, or a curriculum-aligned tutor over a long conversation.

4. 4

   No knowledge-base control

   Customers can't pin the conversation to their own facts (curriculum, product catalog, SOPs, clinical protocols). The model decides what's true.

5. 5

   High build cost

   Rolling your own context-grounded voice stack means stitching together STT, retrieval, LLM, TTS, latency engineering, and safety eval — months of work per deployment.

A patented five-stage pipeline that makes context retrieval a first-class step before the LLM ever responds:

[Voice Input] → [Voice 2 Text] → [Context Search] → [Context Database]
                                                         ↓
[Context + Query] → [Large Language Model] → [Text 2 Voice] → [Voice Output]

Five things this gives a deployment that an off-the-shelf voice agent can’t:

• Context-aware processing — retrieves relevant knowledge before generating, so answers are grounded.
• Admin / parent control — the knowledge base is curated by the deploying party.
• Trained voice personas — multiple consistent voices and personalities per deployment.
• Modular architecture — drop-in for toys, apps, kiosks, robots, enterprise.
• Token-licensed — $AXTORA gates access, making integration a transparent, on-chain transaction.

The patented pipeline (HK30101316) is composed of five stages:

1. 1

   Voice → Text

   Multilingual, multi-accent speech recognition optimized for low latency.

2. 2

   Context Search

   Retrieval over the curated knowledge base, conditioned on the query, the deployment context, and prior turns.

3. 3

   Context Database

   Customer-controlled knowledge: curriculum, SKUs, SOPs, safety policies, persona definitions, brand voice, parental rules.

4. 4

   Large Language Model

   Best-in-class LLM, but constrained: it sees the retrieved context plus the user's query and is steered by the persona system prompt.

5. 5

   Text → Voice

   Trained voice persona synthesis, maintaining consistent character voice and emotional tone across the session.

Technical advantages

• Real-time-grade latency (target sub-300ms first audio).
• Multilingual STT and TTS with accent coverage.
• Cloud-first with edge deployment options for privacy-sensitive workloads.
• Privacy-first: customer-owned context databases, no required telemetry of conversation content.
• Customizable personas and knowledge per deployment.

Granted

• Hong Kong — HK30101316 · granted May 10, 2024 · priority Feb 1, 2024 · enforceable.

Pending

• United States — application filed (pending).
• India — application filed (pending).
• China — application 2024800631123 filed (pending).
• PCT International — application filed (pending).
• Europe — filing deadline July 2, 2026.
• Japan — filing deadline July 2, 2026.
• Korea — filing deadline July 2, 2026.

Combined TAM across our five segments is $500B+:

Healthcare and enterprise voice are additive on top — they aren’t in the table because they overlap with Conversational AI accounting.

Token utility

1. 1

   Licensing access

   $AXTORA is required to license the Axtora pipeline for commercial deployments. Pricing is transparent and on-chain.

2. 2

   Governance

   Holders vote on feature priorities, patent portfolio expansion, partnerships, and protocol upgrades.

3. 3

   Revenue sharing

   A portion of licensing revenue, patent royalties, API fees, and premium subscriptions is distributed to holders.

4. 4

   Priority access

   Early features, priority support, beta program, and exclusive content.

Distribution

Participation

1. 1

   Visit app.virtuals.io

   Open the Virtuals Protocol app in a Web3-enabled browser.

2. 2

   Connect a Web3 wallet

   Any wallet compatible with Virtuals Protocol on Base.

3. 3

   Hold $VIRTUAL

   $VIRTUAL is the participation currency for the IAO.

4. 4

   Open the $AXTORA listing

   Navigate to the $AXTORA project page on May 18, 2026.

5. 5

   Participate

   Commit your desired amount of $VIRTUAL.

• AI-agent native — the platform is purpose-built for tokenizing AI agents, not generic tokens.
• Proven launches — multiple successful AI agent IAOs.
• Liquidity — automatic DEX integration and deep AI-agent liquidity.
• Community — engaged audience already attentive to AI-agent narratives.
• Infra — staking, governance, and cross-chain support out of the box.

Phase 1 — Token launch & narrative (Q2 2026)

• IAO on Virtuals (May 18, 2026) is the headline event — narrative is `patented voice AI now tokenized`, not `another AI agent token`.
• Lean into the patent + HKPC government vetting as the credibility moat in every piece of comms.
• Whitepaper v1.0 + this TL;DR + the live /demo route are the primary content surfaces.
• Community building on X, Telegram, Discord — co-marketing with Virtuals.

Phase 2 — Pilot & design partners (Q3 2026)

• Beta program for early adopters — 5-10 design partners across toys, edtech, eldercare, hospitality concierge, retail.
• First commercial licensing agreements signed using $AXTORA.
• Mobile SDK and developer-preview API.
• Patent filings extended to Europe, Japan, Korea (deadline: Jul 2, 2026).

Phase 3 — Commercial GA (Q4 2026)

• Self-serve licensing platform live — any developer can spin up a context-aware voice agent and pay in $AXTORA.
• Public API + first hardware integrations (toy, robotics) shipping.
• Staking goes live; revenue-sharing mechanism in production.

Phase 4 — Marketplace & scale (2027)

• AI agent persona marketplace — anyone can publish a context-grounded voice persona, monetized in $AXTORA.
• Third-party SDK and verticalized templates (clinic aide, retail concierge, factory engineer).
• Educational toy partnerships, healthcare pilots, additional language support.
• Cross-chain expansion beyond Base.

Channels

• Direct enterprise — outbound to toy, eldercare, hospitality, and retail concierge buyers; founder-led for the first 5 design partners.
• Developer self-serve — docs, SDK, sandbox; growth via the Virtuals + AI-agent dev community.
• Hardware partners — embed the pipeline in toys, robots, kiosks; partner pays in $AXTORA per device or per session.
• Token holders as evangelists — governance gives them a real seat at the table on roadmap and partnerships.

Pricing model (initial)

• Per-session licensing for low-volume / pilot customers — pay-per-conversation in $AXTORA.
• Per-device flat-rate for hardware OEMs (toys, robots).
• Enterprise commit + revenue share for high-volume deployments (call centers, retail).
• Marketplace fee on persona sales (Phase 4).

North-star metrics

• Sessions / week running on the pipeline (proxy: real usage).
• Active licensed deployments (proxy: enterprise traction).
• $AXTORA burnt per session (proxy: token utility velocity).
• Patent jurisdictions granted (proxy: defensive moat).

2026

2027

• Founder & CEO — voice tech and NLP background; AI + blockchain.
• CTO — ML architect; scalable cloud infra.
• Patent counsel — IP specialist running the global patent strategy.
• Advisors — AI/ML researchers, DeFi/protocol engineers, edu/toy/robotics industry vets, regulatory counsel.

• $AXTORA is a utility token, not a security. No claim of investment return is made.
• Participation in the IAO involves risk; conduct independent research before participating.
• Patent applications outside Hong Kong are pending and may be modified or rejected.
• Roadmap dates are best-effort and subject to engineering and market conditions.
• This page is a project brief, not financial, legal, or tax advice.