Orvix is a decentralized artificial intelligence (AI) compute network that coordinates supply and demand for AI inference without a central provider, connecting AI developers with a community of independent GPU operators through a single OpenAI-compatible application programming interface (API).[1][2] The protocol settles all payments in USDC, a US-dollar stablecoin, on the Solana blockchain, charging callers per token of output and paying providers per-token in USDC.[3] It positions itself as "The compute layer for the agentic web," aiming to replace centralized AI inference providers with an open market of independent providers, an orchestrator, and on-chain settlement.[3][1]
As of 2026, Orvix is in early development, described by its own team as "Alpha live with inference working" and warned in its repository as "not production-ready."[3][2] The project is licensed under the Apache License 2.0, and its associated token is ORVX, whose Solana contract address is 8TfggLjCG1ba17nBAjASVnRaWgwxaRaMmZZkXcy2pump.[3][4]
Orvix operates as a permissionless GPU network on which anyone with a modern GPU can run a node and earn USDC by serving inference requests.[3] The core design goal is to give AI developers access to distributed compute through one OpenAI-compatible API while letting providers run lightweight node software that converts idle GPUs into billable capacity, producing what the project describes as "open, community-owned inference with no vendor lock-in."[2]
A central selling point is drop-in compatibility: developers can use the network from any existing OpenAI client simply by swapping the base URL, keeping their existing code, and retaining support for streaming responses, tool calls, and the chat-completions surface.[3] The network meters every request on-chain and promises predictable per-token pricing denominated in USDC rather than fiat billing.[3]
The project distinguishes its model from both centralized incumbents and other decentralized compute markets. In a comparison the project draws between Orvix, OpenAI, and io.net, Orvix and OpenAI both expose an OpenAI v1 API surface while io.net uses a proprietary one; Orvix prices per token in USDC, OpenAI per token in US dollars, and io.net per hour of GPU time.[3] Orvix and io.net both allow permissionless providers, whereas OpenAI does not; and Orvix reports that it alone offers an open-source orchestrator and USDC settlement, while io.net operates an IO token economy and OpenAI has none.[3]
Orvix is built around three components: independent providers who contribute GPUs, a stateless orchestrator that routes work, and on-chain settlement on Solana.[1] The system architecture routes developer requests over an OpenAI-compatible HTTPS API to the orchestrator, a backend built with the FastAPI Python web framework, which in turn communicates over WebSocket connections to GPU nodes running the orvix-node agent.[2]
The orchestrator is stateless and accepts OpenAI-compatible requests, authenticates the caller, and dispatches each job to an eligible provider.[1] Provider selection is based on a provider's tier, its current load, and its locality.[1] The orchestrator handles authentication, API keys, routing, and node management.[2] Tokens are metered per job, the caller's USDC balance is charged, the serving provider is paid, and all of these actions are recorded on-chain.[1]
The orvix-node component is a Python agent that runs on GPU provider machines.[2] Real GPU inference has been implemented for chat, using the vLLM inference engine, and for image generation, and the project reports this has been verified end-to-end on a GPU node.[2] The codebase is organized as a monorepo whose top-level folders include the orchestrator, the orvix-node agent, a protocol package, and a staking program, alongside documentation and standard repository files such as ARCHITECTURE.md, CHANGELOG.md, and SECURITY.md.[2] Named technologies in the stack include solders and websockets.[2]
Provider onboarding follows a two-step sequence in which a provider must run a join command before a start command, because the start step has nothing to authenticate with until join has run.[2] Credentials are obtained by calling the POST /v1/provider/register endpoint with a session token, and the wallet key that authorizes the provider "never touches the GPU machine," an arrangement intended to keep signing keys off the serving hardware.[2] The project instructs providers simply to point their GPU at the orchestrator and begin serving requests, with per-token USDC payouts settled on Solana and "no lockups, no middlemen."[3]
Orvix supports several inference workloads through its OpenAI-compatible surface. Chat inference, marketed as "Chat Live," provides OpenAI-compatible completions including streaming and tool calls.[3] Image generation, marketed as "Image Live," is DALL·E-compatible and priced per megapixel.[3] Video generation, marketed as "Video Live," takes a text prompt as input and returns a short clip, with a model named orvix-video-1 serving clips through the network.[3] Among the specific technical attributes the project lists are Qwen-2.5-7B models, OpenAI v1 compatibility, USDC settlement, the Solana chain, and the Apache 2.0 license.[3]
The project also advertises a "Verifiable Inference" mechanism under which model outputs are hashed and providers are stake-slashable, so that, in the project's own words, "Every response is verifiable."[3] The site reports approximate performance and economic figures including roughly two-second first-token latency.[3]
The network's native token is ORVX, which has a total supply of one billion tokens.[3] The project states that participants can earn ORVX in three ways: by routing requests, by providing GPUs, or by holding through the buyback-and-burn mechanism.[3]
Revenue from each request is divided under a 70/30 split, with 70% of provider revenue going to the provider and 30% retained as the platform share.[3] The platform's 30% share is then routed through a 50/30/20 flow: 50% funds a market buyback of ORVX, 30% goes to the treasury, and 20% covers operations.[2][3] A portion of bought-back ORVX is periodically burned, which the whitepaper describes as tying token supply to real network usage.[1]
Beyond its role in the buyback loop, ORVX carries utility for both staking and pricing. Staking ORVX unlocks discount tiers and routing priority, so callers and providers who commit stake receive cheaper pricing and preferential job routing.[1] The project reports a provider staking requirement with a minimum of 25,000 ORVX and stake-based pricing tiers as implemented in its code.[2]
Several parts of the token machinery remain incomplete at the alpha stage. The project describes manual buyback executed through the Jupiter decentralized exchange aggregator together with monthly burn tooling, but reports that on-chain buyback and burn execution is "stub-gated pending devnet testing."[2] The repository lists placeholder flags such as PAYOUT_STUB, BUYBACK_STUB, and BURN_STUB, along with configuration variables including REQUIRE_STAKE_FOR_PROVIDER set to false, indicating that payout, buyback, burn, and staking-enforcement behavior is still stubbed.[2]
Orvix is governed through stake-weighted voting conducted on Snapshot, an off-chain voting platform. ORVX holders decide on network parameters, supported models, fee parameters, and treasury spending policy.[1][3] The whitepaper describes Snapshot voting as gasless and verifiable, meaning votes do not require transaction fees while remaining auditable.[1] Influence in governance scales with committed stake rather than with idle capital, so that voting power reflects tokens actively staked in the network.[1] Snapshot-based governance is also reported as implemented in the codebase.[2]
The project is explicitly labeled early development, or alpha. The repository states that the "backend MVP and node software are built and tested, but the project is not production-ready" and warns users to "Expect breaking changes."[2] It carries a direct caution: "This is alpha software. Do not use it for production workloads or with funds you cannot afford to lose."[2]
The project reports that its status is "Active development — backend MVP + tokenomics complete, image generation live, public testnet incoming."[2] Both the orchestrator and node packages are built and unit-tested, and a cross-process end-to-end flow between a node and the orchestrator over WebSocket has been verified.[2] The orchestrator and frontend are reported as deployed.[2] The project identifies keeping a node continuously online in production as its next milestone.[2]
Contributions of code, documentation, bug reports, and ideas are invited through the repository's CONTRIBUTING.md and its designated good first issues, and the project includes a CODE_OF_CONDUCT.md.[2] Security vulnerabilities are directed to the SECURITY.md file rather than public issues.[2]
The project maintains a narrow set of official channels and warns aggressively against impersonation. Its repository states that two named accounts and the repository itself "are the only official Orvix channels," that "Orvix has no Discord, and there is no other Telegram group," and that anything claiming otherwise "is not us, so treat it as a scam, especially if it mentions a token, an airdrop, or asks you to connect a wallet."[2] Consistent with this posture, the project's X account, @Orvixhq, repeatedly publishes the official ORVX contract address and urges users to "Always verify the contract address before trading or interacting with ORVX."[4] The @Orvixhq account, which lists a join date of February 2026, describes Orvix as "The AI Compute Marketplace" powering "AI models, AI agents & enterprise workloads."[4]
Through August 2026, the Orvix team self-reported a series of infrastructure and product updates on its X account. A "SCALING UPDATE" dated August 24 described changes intended to make the infrastructure "more scalable, reliable, and multi worker ready," listing Redis-backed rate limiting with an in-memory fallback, a Redis authentication challenge store with native time-to-live expiry, and a shared ORVIX Protocol package used by both the node and the orchestrator.[4]
On August 23, the team announced a product initiative called ORVIX Intelligence, stating it is "building more than another AI chatbot" and describing a workflow in which a token contract address can trigger analysis of holders, early buyers, accumulation, and social activity.[4] A companion "INTELLIGENCE UPDATE" the same day reported an expanded Token Intel API documentation giving developers access to an intelligence stack spanning token data, accumulation, holders, early buyers, social signals, wallet clusters, and AI intelligence.[4] A separate "ORVIX UPDATE" reported that the team was working on integrations with two projects it believed could create synergies with the ecosystem, adding that it was "focusing on real utility and meaningful integrations, not just partnership announcements."[4]
