Welcome
Cortex is a staking protocol on Solana. SOL locked into compute clusters mines inference vouchers, and vouchers burn to pay for model API calls.
Overview
Cortex runs nine compute clusters, one per AI model, in three tiers. You stake SOL into a cluster to run a rig. A rig accrues inference vouchers, an SPL token denominated in output tokens: 1 voucher buys 1,000 output tokens at the cluster model's live API price. You either claim vouchers to your wallet or burn them at the gateway to run inference on that cluster's model.
The uncomfortable summary: emission is shared pro rata, unclaimed vouchers decay 5% at every daily cut, and a keeper you do not control has to call the settle instruction or nothing moves at all. Read Risks before you read anything else.
Where to start
- Clusters: the nine models and their tiers.
- Rigs and emission: how stake turns into vouchers.
- Epoch cut and decay: the daily 00:20 UTC cut.
- The gateway: where vouchers become model calls.
- Risks: read this first.