README.md

---
title: README
emoji: 🧭
colorFrom: purple
colorTo: blue
sdk: static
pinned: false
---

# NeoSyntropy

### Deterministic control for AI workflows

**Models interpret. Your application decides and executes.** NeoSyntropy turns
AI workflows into controlled graphs with application-owned schemas, tools,
validation, routing, state transitions, and business policy.

[Website](https://neosyntropy.com) ·
[Framework](https://github.com/NeoSyntropy/neosyntropy-framework) ·
[Documentation](https://neosyntropy.com/docs)

---

## Focused role models

| Model | Responsibility |
| :-- | :-- |
| [Runtime Structure](https://huggingface.co/Neosyntropy/runtime-structure) | Structured extraction into application-owned schemas |
| [Runtime Route](https://huggingface.co/Neosyntropy/runtime-route) | Valid workflow topology and candidate selection |
| [Runtime Deterministic Reasoning](https://huggingface.co/Neosyntropy/runtime-deterministic-reasoning) | Schema-constrained reasoning and declared tool calls |
| [Runtime Stochastic Reasoning](https://huggingface.co/Neosyntropy/runtime-stochastic-reasoning) | Open reasoning with controlled tool requests |
| [Runtime Guard](https://huggingface.co/Neosyntropy/runtime-guard) | Evidence-based validation against supplied rules |
| [Runtime Score](https://huggingface.co/Neosyntropy/runtime-score) | Rubric-grounded scoring and measurement |

## One contract from data to inference

**Workflow contract → diverse scenarios → teacher gold → schema and semantic
validation → critic gate → frozen evaluation → role adapter**

Every accepted sample preserves prompt version, source lineage, teacher
provenance, validation checks, and a canonical content hash. Training artifacts
and metrics are published only after evaluation against frozen test sets. Each
model card links to its matching private, versioned pilot dataset for inspection.