Skip to content

compMOR (v0.9 draft)

Tokenized AI inference credit on the Morpheus stack. Extends Venice DIEM.

Draft status

Pre-finalization draft (v0.9). Not yet deployed to mainnet. The full specification lives at COMPMOR_PROTOCOL_SPEC.md in the monorepo (~7,000 lines).

What is compMOR?

cMOR is a non-tradeable inference credit token. Holders burn cMOR to pay for AI inference; providers earn MOR (via settlement) for executing inference. The protocol:

  • Wraps Venice DIEM's subscription model as a programmatic, agent-friendly primitive
  • Routes inference across a ProviderRegistry of compute providers (Akash, Aethir, Render, Replicate, io.net, etc.)
  • Settles on-chain via Settlement contract, with attestations from a consensus forecasting pool

Components

ContractRoleStatus
AgentVaultERC-721 agent identity + per-agent cMOR ledgerv0.4 draft, 61/0 tests
AgentRefillRouterDEX-quote-based MOR → cMOR refillv0.2 draft, 34/0 tests
ConsensusForecastingPoolInference cost / quality oraclev0.x draft
CommitmentVaultPre-committed inference budgetsv0.x draft
ProviderRegistryOn-chain index of compute providersv0.x draft
SettlementMatches consumption events to debitsv0.x draft
FeeRouterSplits fees between recyclers + treasuryv0.x draft
CreditMinterThe cMOR ERC-20 itselfv0.x draft
GenesisLockPer-agent genesis-hash anchorv0.x draft
WorkingSoulPer-agent working-state hash + drift proofsv0.x draft
MergeAgentMulti-agent merge primitivev0.x draft
AgentGovernanceOn-chain governance of agent policyv0.x draft

Anchor demand: Hermes Agent

compMOR §22.10 designates Nous Research's Hermes Agent as the canonical anchor-demand "lighthouse" customer for cMOR burn. Hermes is an open-source self-improving AI agent with broad community adoption. The integration creates programmatic 24/7 cMOR demand that anchors the buyback-and-burn loop.

Reading order

For protocol implementers:

  1. Read COMPMOR_PROTOCOL_SPEC.md §§ 1–3 (motivation + token mechanics)
  2. Read §§ 4–10 (contracts + settlement)
  3. Read § 11 STRIDE threat model
  4. Read § 22 (Hermes integration)

The full spec is ~7,000 lines; treat the above as the minimum reading path.

See also

Released under the MIT License.