Core concepts
What NeoSyntropy controls
A model may propose a bounded plan; the runtime checks that plan against graph structure and transition legality before anything commits.Proposal is not permission
The router does not perform business operations. It proposes candidate nodes and an execution topology. Validators and the transition table decide whether that proposal may run and whether state may advance.
This keeps language-model judgment useful without making model output the final authority for state movement or side effects.
System layers
Selection
Retrieve and rank likely capabilities for the current intent and state.
Control
Validate topology, prerequisites, fallback isolation, and edges.
Execution
Run local handlers (or provider-backed nodes), then submit results for commit.
Python SDK entry point
The public control surface is ControlManager. With a BackendClient it starts an opaque control run, executes local work units, and submits results until the backend completes or rejects the cycle.
from neosyntropy import ControlManager, FSM
manager = ControlManager(graph) # BackendClient.from_env() when credentials exist
result = manager.run({
"intent": "verify the customer and review refund eligibility",
"current_state": "Start",
"state": {"order_id": "ord_123"},
})
print(result.final_state)
print(result.completed)
print(result.rejected)Grounded in neosyntropy-framework/docs/concepts.md.