Decentral is a decentralized AI inference network built on Solana. Anyone with a GPU can share compute and earn USDC. Anyone with a wallet can run uncensored AI. Every job, every payout, every buyback is a verifiable transaction on-chain.
The AI systems most people use are built on a fragile foundation. A handful of companies decide what questions you're allowed to ask, log everything you send them, and can change their terms overnight. GPU owners who want to monetize idle hardware have to go through corporate intermediaries that take most of the margin. The infrastructure for a permissionless inference market already exists. The coordination layer is what was missing.
That's what Decentral builds.
A private AI interface backed by the distributed compute network. No account required, no content filters, no logs. Pay per query with credits funded via a Solana wallet. Every session is fully ephemeral: prompts are encrypted client-side before they leave the browser and nothing is retained after the conversation ends.
A dashboard for GPU providers to register hardware, stake $DECENTRAL, and start fulfilling inference jobs. Browser workers can start earning through WebGPU with zero install. Native workers running the decentral-node daemon unlock higher earnings and larger model support across CUDA, Metal, and ROCm.
An OpenAI-compatible REST API backed by the decentralized network. Change one line of config in an existing integration and get private, uncensored, low-cost inference with per-token billing settled on-chain.
from openai import OpenAI
client = OpenAI(
base_url="https://api.usedecentral.org/v1",
api_key="dcntrl_live_..."
)A public, real-time view of the entire network. Live job feeds, provider leaderboards, token metrics, and treasury activity. Every buyback is a Solana transaction with a link.
Users submit inference jobs and lock credits in an on-chain escrow. The orchestrator network (a libp2p peer mesh, not a central server) routes each job to the best available worker based on model availability, latency, stake weight, and reputation. Workers run inference, return the output, and submit a cryptographic proof of completion. The settlement program verifies the proof and releases payment automatically: 75 to 85 percent goes to the worker in USDC and the rest flows to the protocol treasury.
Solana makes this practical. At roughly $0.00025 per transaction, settling a $0.08 inference job costs less than 0.3 percent overhead. 400ms finality means settlement often completes before the full token stream reaches the user.
Prompts are encrypted with AES-256-GCM before leaving the browser. Workers decrypt locally, in memory, for the duration of inference only. The orchestrator layer never sees plaintext content. On-chain records contain job metadata (model tier, credit amount, completion proof hash) and nothing else.
The only required identity is a Solana wallet address. No email, no phone, no KYC.
| Layer | Technology |
|---|---|
| Blockchain | Solana Mainnet |
| Smart contracts | Rust + Anchor |
| Settlement token | USDC (SPL) |
| Governance token | $DECENTRAL (SPL) |
| P2P networking | libp2p |
| Inference (native) | llama.cpp (CUDA / Metal / ROCm) |
| Inference (browser) | WebLLM + WebGPU |
| Indexing | Helius |
| API | TypeScript / Node.js |
| Dashboard | Next.js |
| SDKs | TypeScript + Python |
Open beta. All four products are live simultaneously: no waitlist, open access from day one.