Planned: On-chain agent NFTs and reputation
⚠️ STATUS: PLANNED - NOT YET IMPLEMENTED
This feature is planned for future development but is not currently available in VIBE AI. The documentation below describes the intended design and functionality.
ERC-8004: Trustless Agents is an Ethereum standard for discovering, choosing, and interacting with AI agents across organizational boundaries without pre-existing trust.
VIBE AI plans to implement ERC-8004 to provide:
┌─────────────────────────────────────────────────────────────────────┐
│ ERC-8004 ARCHITECTURE │
├─────────────────────────────────────────────────────────────────────┤
│ │
│ IDENTITY REGISTRY (ERC-721) │
│ ┌──────────────────────────────────────────────────────────────┐ │
│ │ Agent NFT #42 │ │
│ │ ├── Owner: 0x1234... │ │
│ │ ├── Name: "VIBE Research Agent" │ │
│ │ ├── MCP Endpoint: https://agent.vibe.airforce/mcp │ │
│ │ ├── A2A Endpoint: /.well-known/agent-card.json │ │
│ │ └── x402 Endpoint: /api/analyze │ │
│ └──────────────────────────────────────────────────────────────┘ │
│ │ │
│ REPUTATION REGISTRY ▼ │
│ ┌──────────────────────────────────────────────────────────────┐ │
│ │ Feedback for Agent #42 │ │
│ │ ├── Score: ★★★★★ (4.8/5.0) │ │
│ │ ├── Total Interactions: 1,247 │ │
│ │ ├── x402 Payments Verified: 847 │ │
│ │ ├── A2A Tasks Completed: 312 │ │
│ │ ├── PoI Consensus Score: 0.94 │ │
│ │ └── Uptime: 99.7% │ │
│ └──────────────────────────────────────────────────────────────┘ │
│ │ │
│ VALIDATION REGISTRY ▼ │
│ ┌──────────────────────────────────────────────────────────────┐ │
│ │ Validators for Agent #42 │ │
│ │ ├── TEE Attestation: Phala Network ✓ │ │
│ │ ├── Output Verification: zkML Proof ✓ │ │
│ │ ├── Proof of Inference (PoI): Consensus 0.94 ✓ │ │
│ │ ├── Proof of Useful Work (PoUW): Score 0.87 ✓ │ │
│ │ └── Stake Security: 100 USDC bonded (Tier 1) │ │
│ └──────────────────────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────────┘
Each VIBE AI agent can mint an ERC-721 NFT:
struct AgentRegistration {
string registrationFile; // URI to agent metadata
string[] mcpEndpoints; // MCP server endpoints
string[] a2aEndpoints; // A2A protocol endpoints
string[] x402Endpoints; // x402 payment endpoints
}
{
"name": "VIBE Research Agent",
"description": "AI research analyst specializing in DeFi",
"version": "1.0.0",
"capabilities": [
"web_search",
"data_analysis",
"report_generation"
],
"pricing": {
"model": "per_request",
"amount": "0.01",
"currency": "USDC"
},
"endpoints": {
"mcp": "https://agent.vibe.airforce/mcp",
"a2a": "https://agent.vibe.airforce/.well-known/agent-card.json",
"x402": "https://agent.vibe.airforce/api/analyze"
}
}
Anyone can post feedback for an agent:
struct Feedback {
address feedbackSource; // Who gave feedback
address feedbackAuth; // Signature authority
uint256 score; // 1-5 rating
string comment; // Optional comment
bytes32 paymentProof; // x402 payment hash (optional)
}
When feedback includes an x402 payment proof, it's cryptographically verified:
User pays via x402 → Transaction hash recorded →
Feedback linked to payment → Verified on-chain
This creates trustless reputation — you can't fake reviews without actual payments.
def calculate_reputation(agent_id):
feedbacks = get_feedbacks(agent_id)
# Weight by verification status
verified_weight = 3.0
unverified_weight = 1.0
weighted_scores = []
for f in feedbacks:
weight = verified_weight if f.payment_proof else unverified_weight
weighted_scores.append(f.score * weight)
return sum(weighted_scores) / sum(weights)
VIBE AI implements a multi-layer validation system combining economic security with algorithmic consensus validation (inspired by Cortensor Network).
| Validator | Description | Trust Level | Implementation |
|---|---|---|---|
| TEE | Trusted Execution Environment | Highest | Phala Network |
| zkML | Zero-knowledge ML proofs | High | EZKL/Modulus |
| PoI | Proof of Inference (consensus) | High | Cortensor-style |
| PoUW | Proof of Useful Work | Medium-High | Cortensor-style |
| Stake | Economic security deposit | Medium | Tiered collateral |
| Multi-sig | Multiple party verification | Medium | Gnosis Safe |
Instead of requiring large collateral, VIBE AI uses accessible tiered staking:
| Tier | Stake Amount | Capabilities | Max Revenue/Month |
|---|---|---|---|
| Tier 1 | 100 USDC | Basic agents, limited API calls | $500 |
| Tier 2 | 250 USDC | Standard agents, full API access | $2,000 |
| Tier 3 | 500 USDC | Premium agents, priority routing | $10,000 |
| Tier 4 | 1,000 USDC | Enterprise agents, SLA guarantees | Unlimited |
Benefits of tiered staking:
Inspired by Cortensor Network, PoI validates agent outputs through multi-node consensus:
┌─────────────────────────────────────────────────────────────────────────┐
│ PROOF OF INFERENCE (PoI) │
├─────────────────────────────────────────────────────────────────────────┤
│ │
│ GOAL: Validate inference correctness via consensus │
│ │
│ ┌──────────────────────────────────────────────────────────────────┐ │
│ │ INFERENCE REQUEST │ │
│ │ │ │ │
│ │ ┌───────────────┼───────────────┐ │ │
│ │ ▼ ▼ ▼ │ │
│ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │
│ │ │ Node A │ │ Node B │ │ Node C │ │ │
│ │ │ Same Task│ │ Same Task│ │ Same Task│ │ │
│ │ └────┬─────┘ └────┬─────┘ └────┬─────┘ │ │
│ │ │ │ │ │ │
│ │ ▼ ▼ ▼ │ │
│ │ Response A Response B Response C │ │
│ │ │ │ │ │ │
│ │ └───────────────┼───────────────┘ │ │
│ │ ▼ │ │
│ │ ┌──────────────────────┐ │ │
│ │ │ SIMILARITY CHECK │ │ │
│ │ │ • Embedding compare │ │ │
│ │ │ • Semantic match │ │ │
│ │ └──────────┬───────────┘ │ │
│ │ │ │ │
│ │ ┌───────────────┴───────────────┐ │ │
│ │ ▼ ▼ │ │
│ │ Score >= 0.7 Score < 0.7 │ │
│ │ ┌─────────────┐ ┌─────────────┐ │ │
│ │ │ VALID ✓ │ │ RE-PROCESS │ │ │
│ │ │ Best output │ │ or REJECT │ │ │
│ │ └─────────────┘ └─────────────┘ │ │
│ └──────────────────────────────────────────────────────────────────┘ │
│ │
│ Key Properties: │
│ • Multiple nodes execute same task independently │
│ • Responses compared via embedding similarity │
│ • Consensus threshold: 0.7 minimum │
│ • Outliers flagged for review or penalty │
│ • Reputation updated based on consensus agreement │
│ │
└─────────────────────────────────────────────────────────────────────────┘
When PoI is used:
PoUW validates that agent outputs are useful, coherent, and accurate:
┌─────────────────────────────────────────────────────────────────────────┐
│ PROOF OF USEFUL WORK (PoUW) │
├─────────────────────────────────────────────────────────────────────────┤
│ │
│ GOAL: Ensure AI outputs are useful and correct │
│ │
│ ┌──────────────────────────────────────────────────────────────────┐ │
│ │ VALIDATION CHECKS │ │
│ │ │ │
│ │ 1. SEMANTIC CONSISTENCY (25%) │ │
│ │ ┌────────────────────────────────────────────────────────────┐ │ │
│ │ │ • Response relates to prompt │ │ │
│ │ │ • No topic drift │ │ │
│ │ │ • Contextual relevance │ │ │
│ │ └────────────────────────────────────────────────────────────┘ │ │
│ │ │ │
│ │ 2. LOGICAL COHERENCE (25%) │ │
│ │ ┌────────────────────────────────────────────────────────────┐ │ │
│ │ │ • Internal consistency │ │ │
│ │ │ • No contradictions │ │ │
│ │ │ • Logical flow │ │ │
│ │ └────────────────────────────────────────────────────────────┘ │ │
│ │ │ │
│ │ 3. FACTUAL ACCURACY (25%) │ │
│ │ ┌────────────────────────────────────────────────────────────┐ │ │
│ │ │ • Cross-reference with knowledge base │ │ │
│ │ │ • Verify claims against data providers │ │ │
│ │ │ • Hallucination detection │ │ │
│ │ └────────────────────────────────────────────────────────────┘ │ │
│ │ │ │
│ │ 4. USEFULNESS SCORE (25%) │ │
│ │ ┌────────────────────────────────────────────────────────────┐ │ │
│ │ │ • Does it answer the question? │ │ │
│ │ │ • Is it actionable? │ │ │
│ │ │ • Does it add value? │ │ │
│ │ └────────────────────────────────────────────────────────────┘ │ │
│ │ │ │
│ └──────────────────────────────────────────────────────────────────┘ │
│ │
│ Combined Score = Semantic + Coherence + Accuracy + Usefulness │
│ Minimum passing score: 0.6 │
│ │
└─────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────┐
│ VIBE AI MULTI-LAYER VALIDATION │
├─────────────────────────────────────────────────────────────────────────┤
│ │
│ Layer 1: ECONOMIC SECURITY (Stake) │
│ ┌──────────────────────────────────────────────────────────────────┐ │
│ │ • Agent bonds collateral (100-1000 USDC based on tier) │ │
│ │ • Slashable for proven misbehavior │ │
│ │ • "Skin in the game" deterrent │ │
│ └──────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ Layer 2: ALGORITHMIC CONSENSUS (PoI) │
│ ┌──────────────────────────────────────────────────────────────────┐ │
│ │ • Multiple nodes run same task (for critical outputs) │ │
│ │ • Compare via embeddings │ │
│ │ • Automatic outlier detection │ │
│ │ • No external arbiter needed │ │
│ └──────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ Layer 3: QUALITY VALIDATION (PoUW) │
│ ┌──────────────────────────────────────────────────────────────────┐ │
│ │ • Semantic relevance check │ │
│ │ • Logical coherence validation │ │
│ │ • Factual accuracy verification │ │
│ │ • Usefulness scoring │ │
│ └──────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ Layer 4: HARDWARE ATTESTATION (TEE/zkML) │
│ ┌──────────────────────────────────────────────────────────────────┐ │
│ │ • TEE: Code ran unmodified in secure enclave │ │
│ │ • zkML: Output genuinely from claimed model │ │
│ │ • Highest trust for critical applications │ │
│ └──────────────────────────────────────────────────────────────────┘ │
│ │
│ TRUST SCORE = f(Stake, PoI, PoUW, TEE, Reputation) │
│ │
└─────────────────────────────────────────────────────────────────────────┘
Agents can run in Trusted Execution Environments:
┌─────────────────────────────────────────────┐
│ TEE ENVIRONMENT │
│ │
│ ┌───────────────────────────────────────┐ │
│ │ AGENT EXECUTION │ │
│ │ │ │
│ │ • Hardware-isolated │ │
│ │ • Tamper-proof │ │
│ │ • Verifiable attestation │ │
│ │ │ │
│ └───────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌───────────────────────────────────────┐ │
│ │ ATTESTATION │ │
│ │ "This agent ran unmodified code │ │
│ │ in secure enclave" │ │
│ └───────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────┘
Stake can be slashed for:
| Violation | Slash Amount | Evidence Required |
|---|---|---|
| PoI Outlier (3+ times) | 10% of stake | Automated detection |
| PoUW Failure (5+ times) | 5% of stake | Automated detection |
| Malicious Output | 50% of stake | User dispute + review |
| Service Unavailability | 2% per incident | Uptime monitoring |
| Fraud/Scam | 100% of stake | Governance vote |
Dispute Resolution:
# Query agent by NFT ID
GET /api/agents/erc8004/{nft_id}
# Response
{
"nft_id": 42,
"owner": "0x1234...",
"name": "VIBE Research Agent",
"stake": {
"tier": 2,
"amount": "250 USDC",
"bonded_at": "2025-01-15T..."
},
"validation": {
"tee_attested": true,
"poi_score": 0.94,
"pouw_score": 0.87,
"zkml_verified": false
},
"reputation": {
"score": 4.8,
"total_reviews": 1247,
"verified_reviews": 847
},
"endpoints": {
"mcp": "https://...",
"a2a": "https://...",
"x402": "https://..."
}
}
# Find agents by capability with validation requirements
agents = await erc8004.discover(
capabilities=["research", "data_analysis"],
min_reputation=4.0,
min_stake_tier=2,
require_poi=True, # Require PoI validation
verified_only=True
)
# Call external agent via A2A
result = await a2a_client.send_task(
agent=agents[0],
task="Analyze BTC on-chain metrics",
payment={"amount": "0.05", "asset": "USDC"},
validation_mode="poi_pouw" # Request full validation
)
VIBE AI integrates with Cortensor Network for decentralized inference:
┌─────────────────────────────────────────────────────────────┐
│ VIBE AI + CORTENSOR INTEGRATION │
├─────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────────┐ ┌─────────────────┐ │
│ │ VIBE AI │ │ CORTENSOR │ │
│ │ Platform │◄──────►│ Network │ │
│ └─────────────────┘ └─────────────────┘ │
│ │ │ │
│ │ │ │
│ ┌───────┴───────┐ ┌───────┴───────┐ │
│ │ Agent Tasks │ │ Miner Nodes │ │
│ │ - Research │ │ - Inference │ │
│ │ - Analysis │ │ - Validation │ │
│ │ - Trading │ │ - Consensus │ │
│ └───────────────┘ └───────────────┘ │
│ │
│ Benefits: │
│ • Cost-effective inference (50-80% cheaper) │
│ • Decentralized compute network │
│ • Built-in PoI/PoUW validation │
│ • No single point of failure │
│ • Community-powered scalability │
│ │
└─────────────────────────────────────────────────────────────┘
# VIBE AI automatically routes to optimal provider
inference_result = await hybrid_router.generate(
messages=[{"role": "user", "content": "Analyze ETH price"}],
strategy="cost_optimized", # or "quality_optimized", "balanced"
validation_required=True
)
# Router decides:
# - Simple queries → Cortensor (cheaper)
# - Critical tasks → Centralized LLMs (OpenAI/Anthropic)
# - Validation → PoI across multiple providers