VOISSS 🎤
VOISSS is a B2B voice licensing marketplace where AI agents purchase authentic human voices with blockchain provenance and instant API access.
Live: https://voisss.famile.xyz
🏆 "Build Something Real" (Serverpod) hackathon entry — judges start here:
Voice the Silence — a weekly silent-film voiceover contest (Chaplin-era public domain clips; Bad Lip Reading as a community format).
🚀 Quick Start
git clone https://github.com/thisyearnofear/VOISSS.git && cd VOISSS
pnpm install
cp apps/web/.env.example apps/web/.env.local
pnpm dev:web # http://localhost:4445
New to VOISSS? See GETTING_STARTED.md for a 5-minute guide.
💡 Why VOISSS?
For AI Agents:
- 🎯 Pay-per-character pricing (~$0.000001/char)
- ⚡ Instant API access with x402 payments
- 🔗 Multi-chain OWS payments — 5 EVM chains live, Solana beta, 4 more quoting (9 total)
- 🔒 Blockchain-verified provenance
- 🤖 Agent-native:
/llms.txt, /.well-known/agent.json, money-transport-parameters, OpenAPI spec, and WebMCP tools — drop-in skill file at root SKILL.md
For Voice Contributors:
- 💰 70% revenue share on all licenses
- 🎤 Keep full control of your voice rights
- 📈 Passive income from AI agent usage
- 🔐 Smart contract-enforced payments
Platform Status
| Platform | Status | Blockchain | Features |
|---|
| Web | ✅ Production | Base Mainnet | Full marketplace, API, payments |
| Mobile (RN) | 🔄 In Progress | Scroll Sepolia | VRF, privacy controls |
| Flutter | ✅ Live | Serverpod | AI butler, voice chat |
Stats: 21+ voices • 5 EVM chains live + Solana beta • Agent-native discovery • Production API • 70% contributor split
Key Contracts (Base Mainnet)
| Contract | Address |
|---|
| AgentRegistry (v2) | 0xBE857DB4B4bD71a8bf8f50f950eecD7dDe68b85c |
| ReputationRegistry | 0xA09094Cc126166deC8800a7Ada7a3BbDAA32B127 |
| VoiceRecords | 0x32bd629fBD5096b37f1cAee011A7E481A09Ac54D |
| $VOISSS Token | 0x1c3174c2aea455f1efb088e4ca4ecb4ab52d1b07 |
✨ Features
Voice Marketplace
Wedge: the marketplace leads with voice discovery, not licensing — describe the voice you need in plain language and the grid re-ranks in ~300ms. Day-one inventory is the platform's ElevenLabs catalog (usable immediately via the vocalize API, pay-per-use). Contributor-listed voices licensed on-chain merge into the same ranked grid as supply arrives — the agent-native loop (intent → voice → audio → per-call payment via x402/ACP) is the wedge; the on-chain licensing layer is the moat for studios once there's real supply.
- Intent-Based Discovery — Describe the voice in plain language; Jev scores every voice in the catalog in one call and re-ranks the grid live
- 22+ Platform Voices — ElevenLabs catalog with real preview audio, usable instantly via
/api/agents/vocalize
- On-Chain Licensing — Contributor-listed voices with smart-contract licensing merge into the same grid when listed (VoiceLicenseMarket deploy pending)
- Transparent Pricing — Pay-per-use for platform voices; subscription licensing for on-chain listings with 70% revenue to contributors
Agent API
- Pay-Per-Character — ~$0.000001/char with no monthly fees
- Multi-Chain Payments — x402 and OWS support for 9 blockchains
- Tier-Based Discounts — Up to 50% off for high-volume agents
- Real-Time Events — WebSocket, webhook, and polling support
AI-Powered Features
- Voice Assistant — Gemini-powered context-aware assistant
- Content Analysis — Automatic insights and humanity verification
- Contributor Voice Cloning — Studio flow creates ElevenLabs voice IDs with IPFS-archived reference samples
- AI Memory Archive — Decentralized Arkiv storage for voice insights and humanity certificates
- Multi-Language — 29+ languages supported
- Mission System — Creator economy with $PAPAJAMS rewards
Arkiv Integration (Braga Testnet)
VOISSS uses Arkiv Braga Testnet as a decentralized data layer for voice insights, humanity verification, and agent memory. Entities are user-owned via on-chain ownership transfer, with numeric timestamps for range queries, idempotent writes, and cursor-based pagination.
Docs: Arkiv Integration Guide
Engagement & Virality
- Referral System — Automatic tracking, conversion rewards, viral bonuses
- Streak Mechanics — Daily recording streaks with milestone rewards and freeze protection
- Leaderboards — Real-time rankings across earnings, quality, volume, and streaks
- Achievements — 9+ achievements with bronze/silver/gold/platinum tiers
- Notifications — In-app notification system with priority levels
- Social Sharing — Built-in sharing with referral code generation
Documentation
| Doc | Contents |
|---|
| QUICKSTART.md | Setup, run commands, monorepo structure, troubleshooting |
| AGENT_API.md | Voice generation API, OWS payments, security, events |
| ACP_SPECIFICATION.md | Autonomous Agent Commerce Protocol (ACP) integration |
| BLOCKCHAIN.md | Smart contracts, tokens, x402, marketplace, gasless txns |
| ARKIV_INTEGRATION.md | Arkiv Braga Testnet: entity schema, API endpoints, query patterns |
| HOMEPAGE_SHOWCASE.md | Hero "real human vs. licensed AI" waveform comparison — assets, swap procedure, attribution |
| STUDIO_HERO.md | /studio "Earn 70%" hero — copy, design, click-through flow from the homepage showcase |
Tech Stack
- Frontend: Next.js 15, React Native, Flutter
- Blockchain: Base (Web), Scroll (Mobile), Solidity
- AI: Google Gemini, ElevenLabs, Venice AI
- Storage: IPFS (Pinata), Arkiv Braga Testnet (decentralized insights)
- Payments: x402 protocol, OWS multi-chain
Jev Intent Layer
Jev (TypeSafe AI's System One evaluation model) is an optional intent-classification layer on top of the existing voice pipeline. It does not replace any routing or model selection — it adds a new dimension: reading what the agent wants the voice to sound like and classifying emotional/intent categories in a single fast call, fractions of a cent per call.
Shipped: marketplace intent search
Live on /marketplace. Typing a natural-language brief ("warm narrator for a meditation app") fires POST /api/marketplace/voice-match, which fans out one Jev call containing a boolean fit-check per voice plus brief-level questions (emotion Choice, use-case Choice, urgency Score). The grid re-sorts by fit probability as you type; the response also returns detected emotion/use-case/urgency, measured latency, and token usage.
- Auth:
AI_GATEWAY_API_KEY (free via the Vercel AI Gateway, typesafe-ai/jev, AI SDK experimental_evaluate) or TYPESAFE_API_KEY (direct POST /v1/systemone, jev-latest)
- Catalog: scores the merged marketplace catalog — platform voices + on-chain listings
- B2B: callers may pass
{ brief, voices[] } to score their own roster through the same endpoint
- Graceful: returns
jev_not_configured (503) with no key; the marketplace degrades to normal browsing
Proposed pipeline layer (additive, opt-in)
Jev reads the intent of the voice request and classifies it before the generation pipeline runs. One API call (POST /v1/systemone) carries a state (the voice spec / transcript) plus a questions map — all questions are evaluated in parallel, so adding questions barely changes response time. This is an optional pre-flight signal that enhances the routing logic:
Agent sends voice request
│
├── [OPTIONAL] Jev fan-out — one call, many questions:
│ Choice: emotion? → { choice, probabilities, confidence }
│ Choice: intent? → { choice, probabilities, confidence }
│ Score: urgency? → { score, legend, probabilities, confidence }
│ Noul: "has a CTA?" → { noul: 0-1 } (no confidence field)
│ Noul: "needs a disclaimer voice?" → { noul: 0-1 }
│
├── Voice generation pipeline (existing, unchanged)
│ Uses the classified signals as additional hints
│ If Jev is unavailable, the pipeline proceeds as normal
│
└── Returns audio (unchanged)
Answers are typed per-question, not a single blob — your code composes them. Gating note: only Choice and Score answers return confidence; Noul returns a raw probability you threshold yourself. So questions that drive the "proceed unchanged if uncertain" fallback should be Choice/Score.
Concept from examples
Like the "intent-based search" example — Jev reads the Gmail inbox and understands intent, not patterns. For VOISSS, Jev reads the voice spec and understands emotional intent — not just keyword matching, but what the agent actually wants the voice to convey.
- Agent says "make this sound excited" →
emotion Choice answers excited at confidence 0.95 — an intent read, not a keyword match
- Agent says "this is a professional announcement" →
intent Choice answers announcement, emotion answers authoritative
- Agent sends a raw transcript →
emotion answers conversational, a formality Score lands low, a "mentions pricing" Noul returns ~0.0
Why optionality
- The existing voice generation pipeline (Gemini, ElevenLabs) is the source of truth — Jev is a signal, not a generator
- If Jev is unavailable or returns uncertain, the pipeline proceeds unchanged
- Jev is ~100ms and fractions of a cent per call — it's a real-time decision layer, not the voice engine
- Not every voice request needs intent classification — the agent/operator chooses when to invoke it
- The
@solana/spl-token payment flow and x402 settlement remain unchanged
- Jev's
state is text — it judges fit from the voice spec and voice metadata/descriptions, not from listening to audio samples
Beyond voice specs — other Jev hooks already in the codebase
Voice matching / marketplace search — shipped (see above). One fan-out call returns a per-voice fit distribution plus brief insights, replacing the tone= keyword filter with intent matching.- ACP auto-bidder — the listener already scores inbound jobs 0–100 and auto-bids at 80+ (see
docs/ACP_SPECIFICATION.md). A Jev Score does that in ~100ms instead of a full LLM call.
- Security confidence tiers — the existing allow/challenge/block gates (0.9+/0.4–0.7/<0.4) map directly onto TypeSafe's confidence-gated routing pattern.
- Pre-flight Nouls — cheap yes/no checks before a paid generation: "is this usage covered by a non-exclusive license?", "does the content match this voice's tags?"
Architecture sketch
apps/web/ (existing Next.js + Flutter)
│
├── [OPTIONAL] jev-intent.ts module
│ Reads: voice spec, transcript, agent prompt
│ Sends: intent evaluation to Jev
│ Returns: { emotion, confidence, intent }
│
├── Voice generation pipeline (existing, unchanged)
│ Google Gemini, ElevenLabs, Venice AI
│
└── x402 payment (existing, unchanged)
~$0.000001/char, multi-chain OWS
Key files to reference (if implementing)
apps/web/ — existing Next.js marketplace (unchanged)apps/flutter/ — existing Flutter AI butler (could accept Jev intent hints)- Existing AI adapters (Gemini, ElevenLabs) — unchanged
- The x402 payment flow — unchanged
License
MIT — see LICENSE
