Canonical model
One responsibility. One owner.
The object model keeps the system honest.
AgentKit only works if each commercial responsibility has exactly one canonical owner. This keeps identity, policy, objectives, search, ranking, decisions, execution, evidence, and learning distinct and composable.
Canonical objects
CommercialIdentity
Immutable facts about the business.
Owner: companyCommercialPolicy
Approval, compliance, brand, and operating rules.
Owner: companyObjective
The commercial decision to be made.
Owner: founderSearch
Discovery of reachable opportunity space.
Owner: engineRankedOpportunity
Transparent ordering of the options.
Owner: engineDecision
The chosen commercial outcome.
Owner: engineExecutionPlan
Instructions for getting the work done.
Owner: engineExecutionSurface
Where the decision becomes visible.
Owner: platformConnector
How the system reaches each surface.
Owner: platformOwnership rules
- Canonical objects do not share responsibility.
- Temporary artifacts can be produced by a stage, but permanent truth belongs to one object.
- Execution surfaces consume decisions; they do not redefine them.
- Connectors describe capabilities, metadata requirements, disclosure rules, and upload methods.
Why this matters
- It keeps the architecture extendable without drift.
- It makes future platforms pluggable.
- It allows the engine to reason from shared truth instead of surface-specific assumptions.
Primary CTA
One model, many surfaces, no ambiguity.
Use the canonical model to keep the product coherent as new decision types and execution surfaces are added.