Three-Tier Design¶
The Physics OS uses a three-tier architecture separating the open physics engine from the commercial execution fabric and the performance substrate.
Tier 1: ontic (Python)¶
The physics engine. Pure PyTorch tensor network library implementing MPS, MPO, QTT, and 168 physics domain packs. Fully open-source.
Install: pip install ontic-engine
Tier 2: physics_os (Python)¶
Licensed execution fabric providing REST API, Python SDK, MCP tool server, job scheduling, billing, and certificate generation. Never distributed as source.
Surfaces:
physics_os.api— FastAPI REST/WebSocket serverphysics_os.sdk— Python client libraryphysics_os.mcp— MCP tool server (agent-native workflows)physics_os.cli— Command-line interface
Tier 3: crates/ (Rust)¶
Performance substrate for GPU kernels, zero-knowledge proof generation, IPC bridges, and real-time visualization. Compiled to native binaries.
Key crates:
| Crate | Purpose |
|---|---|
ontic_core |
Physics engine core (QTT, MPO, CFD operators) |
ontic_bridge |
RAM bridge IPC (Python <-> Rust streaming) |
proof_bridge |
Trustless Physics: trace -> ZK proof pipeline |
ontic_gpu_py |
GPU compute kernels (PyO3 bindings) |
fluidelite_* |
FluidElite ZK proving family |
qtt_cem |
Maxwell FDTD solver |
qtt_fea |
Hex8 static elasticity solver |
qtt_opt |
SIMP topology optimization |