CaptureSuite

Windows-first multimodal research capture platform: a daemon, versioned worker plugins, sealed session packages, and a PySide6 operator UI.

Platform: Windows 10 21H2+ x64 only for v1.0. The plugin contract is process + named-pipe based so other OSes can be added later without breaking plugins.

Who it's for

Researchers and lab engineers who need structured multimodal capture on Windows: a long-running daemon, drop-in worker plugins (C++ or Python), and sealed session packages for downstream QC / features / pose / ML — without rewriting the core capture path for every device.

What you get

• Daemon — QPC clock, session FSM, recovery, named-pipe control plane
• Plugins — drop in a plugin.json + worker executable (C++ or Python SDK)
• Sim — develop and demo without hardware
• LSL bridge — record any Lab Streaming Layer outlet with zero vendor code
• Analysis — QC, features, pose/kinematics, ML bundle jobs on sealed packages

Architecture (sketch)

flowchart LR
  UI[PySide6 operator UI] -->|named pipe| D[capture_daemon]
  D --> P1[Worker plugins]
  D --> LSL[LSL bridge]
  D --> SIM[Sim workers]
  D --> PKG[Sealed session packages]
  PKG --> AN[Analysis / QC / ML jobs]

Loading

Inspect a synthetic QC result first

After setting up the documented Python 3.12 analysis dependencies, run from the repository root (the parent directory must exist; choose a NEW output path):

py -3.12 tools/demo_qc.py ../capturesuite-qc-demo

This runs the real offline QC engine against a copy of the bundled synthetic EMG/IMU mini-session. It needs no capture daemon, GUI, physical device, account, or patient data. The output contains DEMO-RECEIPT.json and an inspectable HTML and JSON report under synthetic-demo.mmsession/processing/jobs/demo-qc/reports/.

The example checks original-fixture and copied raw-source hashes, rejects an existing output directory, and preserves failure output for diagnosis. It does not establish hardware performance, clinical validity, or full product readiness. Tests: tests/analysis/test_offline_demo.py.

Simulation walkthrough (after build prerequisites)

# Terminal 1 — build + daemon (once per machine: copy CMakeUserPresets.example.json)
. .\scripts\dev-env.ps1
cmake --preset windows-release
cmake --build build/windows-release --target capture_daemon
.\build\windows-release\daemon\capture_daemon.exe

# Terminal 2 — desktop
cd desktop
& "$env:LOCALAPPDATA\Programs\Python\Python312\python.exe" -m capture_desktop

Or use tools/run-demo.ps1 after a release build.

In the UI: status bar shows Connected → Create Session → Rehearse or Start Selected. Greyed buttons usually mean the daemon is not connected.

Bring your own device

1. Implement the worker protocol (or subclass capture_worker.Worker in Python).
2. Drop plugins/<your_id>/plugin.json next to the daemon (or under %LOCALAPPDATA%\CaptureSuite\plugins).
3. Restart the daemon — sources appear in the Capture rail.

Fast path:

.\tools\new_plugin.ps1 -PluginId "lab.force" -DisplayName "Force plate" -Family numeric -Modality force

Then paste a template from docs/prompts/ into Cursor. See docs/plugins/ and docs/design/PLUGIN_REGISTRY.md.

Docs

Audience	Location
Product specification	docs/spec/
Implementation decisions	docs/design/
Plugin authors	docs/plugins/
Operators	docs/operator/
Contributor rules	AGENTS.md, CONTRIBUTING.md

License

CaptureSuite application code is GPL-3.0. Schemas, the wire protocol library, the Python worker SDK, and the C++ stub template are Apache-2.0 so you can wrap proprietary vendor SDKs in plugins without viral licensing. Details: LICENSING.md.

How to cite

See CITATION.cff. A Zenodo DOI will be added on the first tagged release.

Maintainer

Jacob Metoyer (@Jacob-Met) — architecture and product direction. Independent research software (not an official product of any university lab unless separately stated).