Technology

A decentralized AI compute stack built around local-first inference and agent orchestration.

SOLAI separates economic state from inference traffic: Solana handles ownership, staking, registry, leases, escrow and settlement; providers handle off-chain model execution.

Local Runtime

Ollama-compatible models let users run AI locally and let providers expose compatible models without handing over machine control.

Solana Protocol

On-chain programs coordinate staking, provider registration, leases, cluster accounts, job commitments, escrow and settlement.

SOLAI Coder

The planner decomposes requests into task graphs, selects models and providers, manages budgets, retries failed jobs and aggregates results.

Control Loop

From intent to validated result.

The system is designed to let a user specify task, budget, deadline and quality while the protocol and SOLAI Coder handle infrastructure complexity.

USER REQUEST
  -> UNDERSTAND
  -> PLAN
  -> DECOMPOSE
  -> ESTIMATE COST
  -> FIND PROVIDERS
  -> BUILD CLUSTER
  -> EXECUTE
  -> VALIDATE
  -> AGGREGATE
  -> FINAL RESULT

Trust Model

Proof of Compute starts pragmatic and improves over time.

The whitepaper does not claim perfect cryptographic verification at launch. The first layer uses signed jobs, model hashes, execution metadata and result validation, with redundant execution and challenge jobs added progressively.

Layer	Purpose
Signed Job	Binds job ID, provider identity, model, timing, status and signature.
Model Hash	Reduces simple model substitution by requiring identifiable model artifacts.
Replication	Runs critical work across multiple providers to compare outputs.
Future Verification	Challenge jobs, TEEs, confidential execution and verifiable compute research.