# CatOpt: Category-Theoretic Compositional Optimizer (MVP)

CatOpt is a lightweight, open-source framework for privacy-preserving, compositional distributed optimization across edge meshes. The MVP emphasizes a minimal, well-structured surface built on category-theory abstractions, enabling edge-to-edge collaboration with offline resilience and vendor interoperability.

What you get in this MVP
- Local problems expressed by agents (objects)
- Data exchange channels (morphisms) and problem transformers (functors)
- Global assembly via Limits/Colimits and a lightweight ADMM-like solver
- Privacy-by-design data contracts and modular adapters
- A tiny Python runtime surface suitable for rapid prototyping and CI validation

Why this matters
- Composability: add/remove agents without re-deriving the global problem
- Privacy: only exchanged, abstracted quantities per contracts
- Convergence: disciplined, modular solver structure with verifiable properties
- Interoperability: bridges to existing energy/robotics ecosystems via a common DSL

Usage (quick start)
- The core runtime primitives live in src/catopt_category_theoretic_compositional_/
- The tiny DSL bridge to contracts lives in src/catopt_category_theoretic_compositional_/dsl.py
- Importing and testing is kept simple to satisfy the CI gate (see tests/test_basic.py)

For more details, see tests, the MVP runtime, and the protocol DSL (ProtocolContract, build_minimal_contract).

This README is a starting point. As we evolve, we will add more examples, adapters, and end-to-end demos.