Program real FPGAs
from Python.
manhattan-reasoning-gym is a Python SDK and CLI for hardware
design. Iterate locally with fast synthesis and place-and-route feedback,
run untrusted agents in a locked sandbox, and program real cloud FPGAs,
all from one package.
Three ways to use it
Cloud
Program a real Lattice ECP5 over the network with
mrg.cloud.App. Build, program, and drive live hardware in a few lines.
Sandbox
Run an untrusted agent in a locked container with mrg.Sandbox. It vets its own design, then promotes a candidate to silicon through a broker it never touches.
Local build
Get synthesis and place-and-route reports (utilization, Fmax, timing closure) with mrg.build. No board, no sandbox, just Docker.
Explore the docs
Guide
Install, log in, and program a board, then the full SDK and CLI across all three usage modes.
Quickstart · SDK usage · CLI usage
Examples
Complete, runnable designs, from a minimal smoke test to a real INT8 transformer accelerator on silicon.
Hello Wishbone · SAT solver · BERT FFN · FFN accelerator
Reference
The full public API. Every class, function, and signature, generated from the SDK's docstrings.
mrg.cloud · mrg.build · mrg.sandbox · mrg.bench