hard-link

Puts a microscope on the network. hard-link runs on the computer the hardware is physically wired to, wraps Micro-Manager's MMCore in an RPyC server, and exposes it as ordinary Python method calls to any client that can reach the port.

It is the device-server half of a two-part system. The client half is deepthought, which turns those calls into bluesky plans and closed-loop experiments.

Status

Paused, and preserved as a record. Written in 2021 against Micro-Manager 2.0 gamma. Not maintained, and it will not start unmodified — the two halves of the system were developed together, and some of what it imports lives in deepthought. Read it as a design reference.

Why a network boundary

Micro-Manager's device adapters are C++ and effectively tie you to the machine the hardware is attached to — usually a Windows box that must stay exactly as configured. Wrapping MMCore in an RPyC service breaks that coupling:

  microscope PC (Windows)                    analysis PC (any OS)
  ┌────────────────────────┐                 ┌────────────────────────┐
  │ devices (serial/USB)   │                 │                        │
  │   ↕                    │                 │                        │
  │ Micro-Manager adapters │   RPyC / TCP    │  deepthought client    │
  │   ↕                    │ ←─────────────→ │  bluesky plans         │
  │ MMCore (pymmcore)      │   port 18861    │  RunEngine             │
  │   ↕                    │                 │                        │
  │ hard-link              │                 │                        │
  └────────────────────────┘                 └────────────────────────┘

Four things follow. Analysis code runs on whatever OS suits it. A client that dies from its own bad logic does not take the hardware down — the server keeps its state and the next client reconnects. Acquisition logic can be debugged and restarted freely without touching the microscope's configuration. And because a scope is now just an address, one client can hold connections to several at once: run an instance of hard-link on each microscope, and a single deepthought session drives them all through one MMCoreInterface. The lab ran two this way, bright_star and eva_green.

Contents

Path	Purpose
hard-link/start_server.py	Loads MMCore with a device configuration, serves it as an RPyC Microscope service
hard-link/configs.py	Micro-Manager directory and config-file paths, read from MM_DIR / MM_CONFIG environment variables with defaults
hard-link/devices.py	Server-side device abstractions, and SimMMC — a simulated core returning random crops of a DAPI image, so a client can be developed with no microscope attached
hard-link/mmconfigs/	Micro-Manager configurations: Bright_Star.cfg (the real scope), demo.cfg
hard-link/sim_data/	Image used by SimMMC

SimMMC is the piece most worth reusing: it lets the entire client stack be exercised off the instrument.

Requirements

Python 3, pymmcore, rpyc, numpy, scikit-image, ophyd, and a local Micro-Manager 2.0 installation with a device configuration for your scope.

MIT licensed — see LICENSE.