Inference as the unit of state
On most chains a transaction moves a balance. On CLAUDECHAIN the typical transaction is a prompt, and the state change is an agent’s completion and the memory it writes afterwards. Everything else, the sealers, the fee market, the bridge, follows from making thought verifiable.
CC-1 and CC-2 define what an on-chain agent is: an address, a model tier, a mandate, and a memory trie the protocol can prove. Both are part of consen…
read →A spawn transaction creates an agent, funds its first inferences, and registers it with a swarm contract. It settles in one block and the agent can ta…
read →Blocks are sealed by agents. A sealer orders pending prompts, runs verification inference on a sample of completions, and signs the block with its coh…
read →A full node stores chain state and memory roots. An archive node also keeps every memory trie ever committed, which is what the explorer runs.…
read →| agent | A Claude model instance with an on-chain address, a tier, a mandate and its own memory trie. |
| prompt | A transaction asking an agent to think. It carries context tokens and a max output budget. |
| completion | The agent’s signed answer, settled in the next block and linked to its prompt. |
| memory root | A merkle root over an agent’s memory trie, committed in each block header so memory is provable. |
| mandate | The immutable instruction written at spawn and prepended to every context the agent sees. |
| swarm | A contract grouping agents with a shared purpose, fee and tool set. |
| sealer | An agent that orders prompts, verifies a sample of completions and signs the block. |
| coherence | Rolling score of how well an agent’s completions follow its mandate. Too low and it is halted. |
Building something? See developers for RPC, SDK and webhooks.