01
Service Payment
Any PRC20 token pricing, routing, lock, and on-chain settlement
On-Chain AI Protocol
Blockchain as the economic, security, identity, verification, and settlement layer; globally distributed AI nodes as the execution layer.
Invokable
Priced
Verifiable
Settleable
Incentivized
01 · Definition
Protocol Definition
A decentralized AI Compute Network. The chain maintains identity, registry, pricing, verification, and settlement; global AI nodes execute inference, training, and agent workloads.
The chain governs rules and trusted state; AI Workers perform computation. Models, frameworks, hardware, and deployment join as open, independently evolving resources.
02 · Problem
Core Problems
AI services are highly centralized today. The protocol builds an open network where users freely choose providers and idle compute resources form an open market.
01
AI compute highly concentrated on few platforms
02
Users cannot freely choose compute providers
03
Idle GPUs, Macs, servers cannot join an open market
04
Pricing and allocation decided by centralized platforms
05
Users cannot verify providers executed as agreed
06
Lack of open economic incentives for AI compute and training
07
No unified open protocol across heterogeneous models and devices
03 · Architecture
Architecture Principle: Chain / Execution Separation
Blockchain Layer
OFF-CHAIN
AI Workers (Off-Chain)
Users → API / SDK / Web / Agent
↓
AI Gateway → Scheduler
↓
Worker A · Worker B · Worker C
↓
AI Execution → Usage Receipt
↓
Verification → Blockchain Layer
↓
Payment · Stake · Reputation → PRC20 Settlement → Research Treasury
04 · On-Chain
On-Chain Modules
02
Node Registry
Node identity, hardware declaration, model, pricing, stake, reputation
03
Model Registry
Model ID · Hash · Version · Architecture · Runtime · Capabilities · License · Benchmark · Owner
04
Task Registry
Task ID, input commitment, price cap, status, result commitment
05
Payment
Balance, lock, settlement, refund
06
Verification / Dispute
Task verification, dispute resolution, stake slashing
05 · Worker
AI Worker: Open Compute Node
Anyone with compute resources may run an AI Worker. No uniform hardware or deployment requirement; must implement AI Worker Protocol interfaces.
REGISTER → HEARTBEAT → OFFER → TASK_ACCEPT → TASK_START → TASK_STREAM → TASK_COMPLETE → PROOF_SUBMIT → SETTLE
Model / Protocol Decoupling
Models join as registered compute resources in Model Registry. Any model, framework, or hardware can evolve independently.
Model A · Model B · Custom · Specialized · Community · AI-generated
Model ID · Hash · Version · Architecture · Runtime · Capabilities · License · Benchmark · Owner
llama.cppvLLMOllamaTensorRTAgent Runtime
Supports home PCs, Macs, gaming PCs, GPU workstations, cloud servers, data centers, and future compute hardware.
06 · Gateway
Decentralized AI Gateway
Unified entry to the Pantanal AI network: chain-native AI API, SDK, Web, and Agent interfaces. AI nodes may run any registered model; the Gateway decentralizes as the network evolves.
Pantanal AI API
SDK
Web Interface
Agent Interface
API Key
Wallet Authentication
Rate Limit
Routing
User Experience: Buy AI Services
Users interact via AI API / SDK—choose model and service tier, submit tasks, and receive results. Scheduler routes by model, price, latency, reputation, availability, compute capability, and privacy level.
Client
Pantanal AI API
Gateway
Scheduler
AI Node
Response
Scheduler Routing Strategy
Model
Price
Latency
Reputation
Availability
Compute
Privacy
07 · Economy
Payment & Economic Model
Service payment: any on-chain PRC20 token for pricing and settlement; covers Inference · Agent · Training · Evaluation · Specialized AI Services
◇
Service Payment
Service payment: any on-chain PRC20 token for pricing and settlement; covers Inference · Agent · Training · Evaluation · Specialized AI Services
○
Payment Channel
Payment Channel: task-level spend cap, streaming usage receipts, on-chain final settlement
Inference
Agent Execution
Training
Evaluation
Specialized AI Services
- Compute nodes: PRC20 service revenue
- Validators: verification rewards
- Protocol: Research Treasury and ecosystem incentives
08 · Payment
Payment Channel
AI services are continuous computation. Payment Channel locks max_spend at task start, streams Signed Usage Receipts, and finalizes on-chain settlement at task end.
Avoids one on-chain transaction per generated token, reducing chain burden.
01
Lock
Lock max_spend
02
Stream
Signed usage receipts
03
Settle
On-chain final settlement
04
Cap
Loss ≤ authorization cap
Nonce 1 → cumulative usage = 0.3 Nonce 2 → cumulative usage = 1.0 Nonce 3 → cumulative usage = 1.8 → submit latest state only
09 · Price Lock
Task-Level Price Lock
All price parameters are locked at AI task start, ensuring economic security.
Model · Input Price · Output Price · Max Spend · Task ID · Nonce · Expiration
At task start: lock Model · Input Price · Output Price · Max Spend · Task ID · Nonce · Expiration. Workers cannot unilaterally change price after completion. User max loss = explicit authorization cap.
10 · Boundary
On-Chain / Off-Chain Boundary
Off-Chain
Prompt · Response · Streaming inference · Model execution
On-Chain
Task ID · User · Model ID · Model Hash · Price · Max Spend · Worker · Usage Commitment · Result Hash · Proof · Settlement
Lower on-chain storage cost
Avoid TPS bottlenecks
Protect user privacy
Scale with AI data volume
11 · API Key
API Key / Wallet Separation
API Key (access), Wallet (funds), Payment Channel (task spend cap), and Usage Receipt (actual usage) are managed in separate layers. AI Workers use short-lived, task-level credentials.
API Key
Access identity
Wallet
Funds
Payment Channel
Task spend cap
Usage Receipt
Actual usage
12 · Trust
Node Reputation
Reputation combines Task Success Rate, Latency, Availability, Verification Pass Rate, Dispute Rate, and history. Higher reputation earns more tasks and revenue.
Node A — Success: 99.9% · Latency: 100ms · Dispute: 0.01%
Stake & Slashing
Nodes must stake to bear misbehavior cost. Cheating flows through Dispute → Verification → Slashing, making cheating cost more than potential gain.
Fake Result
Fake Usage
Fake Capability
Fake Benchmark
Incorrect Execution
Repeated Failure
13 · Verifiable AI
Proof of Inference / Verifiable AI
How to prove a stranger node executed the agreed AI computation?
Phase 1
Stake + Reputation + Random Challenge + Signed Receipt + Dispute
Phase 2
Redundant Execution + Independent Verification
Phase 3
TEE Attestation
Long-term
Verifiable Computation · ZK · ZKML
AI computation results are verifiable.
Node Capability Verification
Nodes declare GPU, RAM, Model, Runtime, Performance, Price at registration—but declarations are untrusted. Network verifies via Benchmark, Challenge, Performance Test, Random Verification.
GPU · RAM · Model · Runtime · Performance · Price
Benchmark · Challenge · Performance Test · Random Verification
14 · Privacy
Privacy Levels
The network offers Standard · Private · Confidential · Advanced privacy tiers for different task requirements.
Standard
Standard AI node execution; transport encryption
Private
TLS · ephemeral identity · data lifecycle management · minimal logging
Confidential
TEE · confidential computing · attestation
Advanced
MPC · FHE · ZK-based · privacy-preserving inference (research)
15 · Marketplace
AI Compute Marketplace
Global AI compute marketplace. Nodes price by capability; users specify Model, Latency, Max Price, Privacy—Scheduler matches automatically.
Node A — Low cost / Low performance
Node B — Mid cost / Mid performance
Node C — High cost / High performance
Users purchase AI compute services.
Inference · Agent Execution
Fine-tuning · RL · Evaluation
Data Processing · Model Distillation · Batch Inference
AI Workers join as general AI Compute Providers in an open compute marketplace.
16 · Treasury
AI Research Treasury
Protocol allocates funds from AI compute economy for model training, AI research, benchmarks, security research, open models, and network infrastructure.
Model training
AI research
Benchmarks
Security research
Open models
Network infrastructure
AI Usage → Protocol Revenue → Research Treasury → Compute → Training → Better Models → Better Service → More Usage
17 · Competition
Model Competition & Evolutionary Selection
Different models compete on the network. Better performance and efficiency earn more users and tasks; weaker models lose market share—AI Model Marketplace + Evolutionary Selection.
18 · Governance
Governance Principles
AI proposes, network verifies, protocol governance authorizes.
AI Research Agent → Proposal → Training → Benchmark → Independent Validators → Governance → Protocol Activation
19 · AI-Geth
AI-Geth / EVM Role
AI-Geth is the bottom-layer blockchain for the AI Economy & Verification Layer—hosting EVM, identity, registries, payment, stake, verification, dispute, and governance; AI inference runs on off-chain Workers.
EVM · Token · Identity · Node Registry · Model Registry · Task Registry · Payment Channel · Stake · Reputation · Proof Verification · Dispute · Settlement · Governance
EVM-compatible Blockchain + Smart Contracts + AI Worker SDK + AI Gateway + Scheduler + Verification Network
20 · Economy Loop
Network Core Loop
AI Usage → PRC20 Payment → Compute Provider Revenue → More Compute Supply → More AI Services → More Network Usage → More Research Funding → Model Improvement → Better AI → More Users
21 · Comparison
vs. Traditional AI Cloud
Traditional
AI Company → Model · GPU · API · Billing · User Data
Pantanal
Protocol → Open Compute · Open Models · Open Marketplace · Open Verification · Token Economy · Decentralized Governance
Traditional AI cloud core asset: company-owned models and GPUs.
Pantanal network core asset: global AI Compute Supply organized by protocol.
22 · Abstraction
Core Technical Abstraction
Defined as an open protocol: AI Compute Protocol, decentralized compute network, payment protocol, verifiable AI, service marketplace, and AI evolution. Models, hardware, and Geth are components; the core organizes global heterogeneous compute with unified economics, verification, and settlement.
AI Compute Protocol
Decentralized Compute Network
AI Payment Protocol
Verifiable AI
AI Marketplace
AI Evolution
23 · Ecosystem
Participant Positioning
U
Users
Decentralized AI services similar to AI APIs
D
Developers
Open AI Compute Protocol
N
Compute Nodes
Connect idle compute resources to the network, provide AI services, earn revenue
M
Model Developers
Open network to deploy, test, distribute, and commercialize AI models
B
Blockchain Ecosystem
Settlement and verification infrastructure for AI Compute Economy
A
AI Systems
Open compute network for global resources, training, and continuous evolution
NEAR
Near-term: open decentralized AI inference network
MID
Mid-term: global AI Compute Marketplace where anyone with compute can provide AI services and earn
LONG
Long-term: open AI network that allocates compute, verifies AI computation, funds AI R&D, and enables continuous model competition and evolution
Global Compute → AI Compute Network → AI Services → AI Economy → AI Research → AI Evolution → Next-gen AI → Larger AI Compute Network
Gas Fee and AI Service Fee are settled independently at the protocol layer; both may use any on-chain PRC20 token.