Users enter through AI Terminal, discover agents, create group workflows and use models or developer tools. Some infrastructure comes from TensorOpera/FedML, which the whitepaper identifies as a separate U.S. C-corp partner. The official application repository shows concrete check-in, agent/prompt records, memberships and a 2026 subscription collector; this is stronger evidence of shipped engineering than roadmap graphics, but it does not prove that every AI call or contribution is settled onchain.
COAI’s proxy delegates to 0xa9d0…a877. Initialization minted all one billion units to Safe 0x772f…98d4. That 2-of-3 Safe retains proposer, executor, canceller and default-admin roles over the owner TimelockController, whose minimum delay is 172,800 seconds. Managers can burn only their own balances; the Safe-controlled timelock can change managers and upgrade the implementation. The present absence of mint, pause, blacklist and tax functions is current code, not an immutable constitution.
The whitepaper assigns COAI to service access, publishing, resource registration, recognition and possible governance. It allocates 19.65% as available at TGE and describes a four-year release. No official wallet-to-category map or vesting-contract register was identified in the reviewed materials, so the schedule remains a disclosure rather than independently enforceable proof.
Proof of Intelligence proposes to measure training, inference, data and agent work and eventually coordinate a native L1. The roadmap itself allows changes. The reviewed materials did not establish public live-chain identifiers, validators, consensus code, slashing or governance deployments. COAI on BSC and the proposed PoI chain therefore remain distinct evidence scopes.