FAQ

Common questions about NCN Network v2.

General Questions

What is NCN Network?

NCN Network v2 is a decentralized AI inference network that connects clients who need AI computation with compute providers who have GPU/CPU resources. It uses blockchain-based payments and validator consensus to ensure fair and secure computation.

How does NCN differ from centralized AI services?

Aspect
Centralized
NCN Network

Control

Single company

Decentralized

Pricing

Fixed by provider

Market-driven

Availability

Vendor dependent

Multiple providers

Censorship

Possible

Resistant

Payment

Subscription/API key

Pay-per-use crypto

What AI models are supported?

NCN supports:

  • TorchScript models (.pt, .pth)

  • ONNX models (.onnx)

  • Safetensors models

  • Custom models with executor scripts

Is NCN production-ready?

NCN Network v2 is approximately 85% complete. It's suitable for:

  • ✅ Development and testing

  • ✅ Beta deployments

  • ⚠️ Production (with caution)

See Project Status for current completion status.

Technical Questions

What programming languages are supported?

Backend Services: Rust Model Execution: Python (PyTorch, Transformers) Smart Contracts: Solidity Client Libraries: Currently Rust, others planned

What are the hardware requirements?

Gateway Node:

  • CPU: 2+ cores

  • RAM: 4 GB

  • Storage: 50 GB

Compute Node:

  • CPU: 4+ cores

  • RAM: 8+ GB (16 GB recommended)

  • GPU: Optional but recommended for ML workloads

  • Storage: 100+ GB

P2P Registry:

  • CPU: 2+ cores

  • RAM: 4 GB

  • Storage: 50 GB

How does the sandbox work?

The compute node sandbox uses multiple Linux security features:

  1. seccomp - Filters system calls

  2. Namespaces - Isolates processes, network, filesystem

  3. Landlock - Controls filesystem access

  4. Resource limits - Bounds CPU, memory, time

See Sandbox Security for details.

Can I run NCN on macOS or Windows?

  • Linux: Full support (recommended for production)

  • macOS: Development support (limited sandbox features)

  • Windows: Experimental (sandbox not available)

Payment Questions

How do payments work?

  1. Client submits request

  2. Validators sign preprocessing validation

  3. Client pays to smart contract (escrow)

  4. Compute node executes model

  5. Validators sign completion validation

  6. Smart contract distributes payment

See Payment Flow for details.

What cryptocurrencies are supported?

NCN uses ERC20 tokens on Ethereum-compatible chains:

  • Testnet: NCN Token on Forknet

  • Future: ETH, stablecoins planned

What happens if a request fails?

  • Compute failure: Request expires, automatic refund

  • Validator rejection: Request cannot complete, refund on expiry

  • Network failure: Request times out, automatic refund

How are fees calculated?

Typical distribution:

  • 80% to Compute Node

  • 10% to Gateway

  • 5% to Validators

  • 5% to Treasury

Operator Questions

How do I become a Gateway operator?

  1. Set up a Gateway node

  2. Create a subnet with models

  3. Pay subnet creation fee

  4. Attract compute nodes and clients

See Gateway Operator Guide.

How do I become a Compute provider?

  1. Set up a Compute node

  2. Join a subnet

  3. Download/configure models

  4. Start receiving tasks

See Compute Operator Guide.

How do I become a Validator?

  1. Stake required tokens

  2. Run P2P Registry node with validator key

  3. Participate in validation consensus

See Validator Operator Guide.

How much can I earn?

Earnings depend on:

  • Gateways: Routing fee per request

  • Compute: Execution fee per request

  • Validators: Validation reward per request

Typical fees: See Payment Flow.

Development Questions

How do I add a custom model?

  1. Export model to TorchScript/ONNX

  2. Create executor script

  3. Configure subnet with model

  4. Deploy to compute nodes

See Use Cases - Custom Models.

How do I run tests?

How do I debug issues?

See Debugging Guide.

Where can I find API documentation?

Security Questions

Is NCN secure?

NCN implements multiple security layers:

  • ✅ Sandboxed execution

  • ✅ Cryptographic signatures

  • ✅ Validator consensus

  • ✅ Escrow payments

  • ⚠️ External audit (recommended before production)

See Security Overview.

What if a malicious model is uploaded?

The sandbox prevents:

  • Network access

  • Filesystem access outside allowed paths

  • Dangerous system calls

  • Resource exhaustion

Malicious code is contained and cannot affect the host system.

What if validators collude?

The M-of-N consensus requires 3 of 5 validators to agree. Collusion requires:

  • Compromising multiple validators

  • Overcoming stake-based selection

  • Risking slashing penalties

Economic incentives discourage collusion.

How are private keys protected?

Private keys should be:

  • Never stored on-chain

  • Kept in secure storage (HSM for production)

  • Never committed to version control

  • Rotated regularly

See Best Practices.

Troubleshooting Questions

Why can't I connect to the Gateway?

Common causes:

  1. Gateway not running

  2. Wrong address/port

  3. Firewall blocking

  4. Network issues

Why is my inference request failing?

Check:

  1. Model exists on compute node

  2. Input format is correct

  3. Payment was submitted (if required)

  4. Compute node is not busy

See Common Issues.

Why is the sandbox failing?

Common causes:

  1. Running on non-Linux system

  2. Missing kernel features

  3. Incompatible model code

Try SANDBOX_MODE=permissive for debugging.

See Compute Issues.

Future Plans

What's on the roadmap?

  • Model auto-downloader (Phase 11)

  • Integration testing (Phase 12)

  • Production deployment (Phase 13)

  • Monitoring & analytics (Phase 14)

See Implementation Roadmap.

Will there be official SDKs?

Planned SDKs:

  • Python (in progress)

  • JavaScript/TypeScript

  • Go

Will NCN support other blockchains?

Currently focused on Ethereum/Arbitrum. Future chains may include:

  • Other EVM chains

  • Solana

  • Near

Still Have Questions?

PreviousOverviewNextContributing Guide

Last updated