OrogenVerifiable LLM inference, anchored to physical work.

OpenAI-compatible gateway routed to TEE-attested GPU operators, with validator replay, opML challenges, and burn-and-mint settlement on a Substrate-based chain.

Start here

Protocol RFCs

orogen.network

Inference customers, operators, and the chain can all check.

Every response ships a signed receipt (RFC-0001). Validators replay ≥10% of jobs on identical hardware (RFC-0006). Slashing extrinsics are on-chain with reason codes (RFC-0005). Verifiability is pallet-enforced, not a marketing claim.

How receipts work

Drop-in OpenAI compatibility — your code works unchanged.

Install `orogen-sdk` (Python) or `@orogen/sdk` (TypeScript) — from source for now, registry publish is coming soon — and point `base_url` at the gateway. The wire shape is OpenAI Chat Completions plus four optional `useful_*` fields (nonce, tier, region, price cap). A LiteLLM provider ships the same surface inside existing stacks.

Customer guide

Burn-and-mint settlement accounting.

Customers pay USD. The gateway burns OROG at the oracle TWAP (RFC-0008) and credits the customer in non-transferable CUC. Operators serve inference and the runtime applies the protocol settlement split after verification.

Tokenomics

Demand-elastic emission, pallet-enforced — no foundation mint discretion.

Per-epoch mint is bounded by a 180-day rolling burn cap with a 5%/year supply ceiling and a 0.5%/year floor. Year 1 bootstrap is 8%, Year 2 4%, Year 3+ demand-elastic. Mint outside the rule is impossible at the runtime layer.

Emission rule

What lives in these docs

These docs cover the four audiences identified in the marketing brief. If a section is short, that is intentional — the canonical specs live in the RFCs and the per-pallet code under pallet-suite/; prose here exists to orient, not to duplicate.

Audience	Job to be done	Where to start
Customers — app builders, LLM consumers	Send OpenAI-shaped requests, verify receipts, estimate per-tier price	/customers/
Operators — GPU owners across six tiers	Run a node, pass attestation, stake, route to demand	/operators/
Validators — replay node operators and proof workers	Receive the verification allocation by performing assigned replay or proof work; no passive staking yield	/validators/
Researchers and governance participants	Read the protocol spec, the RFCs, and the tokenomics	/protocol/ · /tokenomics/ · /governance/

Status

Forge testnet is live and open to outside operators. Public endpoints (chain RPC, gateway, attestation, indexer, faucet) are listed on the Forge testnet page. It is a test-mode preview: the gateway runs in test mode, attestation uses mock quotes, and the faucet is low-cap — not a production release.
SDKs are development packages. They are not on PyPI/npm yet — install from source. See Downloads.
Mainnet TGE remains gated. Multi-validator Forge operation, independent validators, real audits, live chaos drills, and release CI all need to close before any production launch claim.

If something here contradicts the canonical RFC, the RFC wins and this page is the bug. Open an issue or edit on GitHub.