Get started
NeoSyntropy documentation
Order in the AI field: models propose only where you allow them; a deterministic graph decides what may happen.Bring order to AI workflows
Most AI frameworks push the agent to do everything — plan, branch, call tools, and move state in one open loop. That is expensive, hard to audit, and easy to lose control of.
NeoSyntropy takes the invertible option. The workflow is an explicit finite-state graph you author. AI is injected only on the routes that need it — a SchemaNode for typed extraction, a ReasoningNode for tool-backed evidence, a SemanticRouter for soft intent. Everywhere else stays deterministic code, rules, and edges.
That design cuts unit cost of AI (models run only on scoped steps), lets you enforce axioms as hard gates and guards, and gives the developer complete control over where intelligence may act and where it may not.
Lower AI cost
Call models only on the nodes and routers that need them — not on every hop of the workflow.
Enforce axioms
Auth, policy, eligibility, and money stay on DeterministicRouter rules, edge guards, and fail-closed validation.
Developer control
You decide the graph, the allow-listed tools, and which steps may use a model. Proposal is never permission.
How it works
Models propose what should happen next; the graph defines what is allowed to happen. You author nodes and edges in the Python SDK. When backend credentials are set, the NeoSyntropy backend owns selection, routing, plan validation, and state commits. The client only executes local handlers and submits results.
Explicit graph
Nodes describe capabilities. Edges describe permitted movement.
Fail-closed gates
Transition legality and guards reject illegal steps before commit.
Typed tools
Nodes declare allow-listed tools with pydantic schemas; undeclared calls never run.
Two execution modes
With backend credentials, ControlManager runs an opaque execute/commit loop against POST /api/v1/control/runs. Without credentials, the SDK falls back to a local deterministic router for offline development.
Grounded in neosyntropy-framework README and docs/concepts-explained.md.