Reasoning state that the model
cannot rewrite by narration
Rulith is a rule-driven working board and execution protocol. The model proposes; the board decides what is supported and what may run.
The execution substrate
Claims retain their origin and support. Model assertions remain distinct from source-attested observations and execution results.
Rules derive consequences, maintain support, detect conflicts, and retract unsupported conclusions without asking the model to remember the chain.
Actions declare inputs, preconditions and execution requirements. The service checks current evidence before admitting governed work.
Each Case identifies a piece of work with acceptance conditions and a recorded outcome. Cases share the Agent graph; capability rules must distinguish business objects explicitly.
Designed for ordinary agent clients
MCP is the primary integration surface. The Agent can discover available Case Types and Actions, inspect bounded views of evidence and request governed work. Account administration and Worker credentials stay outside the model conversation.