Desktop App¶

The IINTS-AF desktop app is a graphical workbench for the same Python SDK used by the CLI. It helps users run curated workflows, inspect artifacts, certify data, use local AI review, and explore research evidence without memorising every command.

The app does not contain a second physiology or safety engine.

For a visual, step-by-step walkthrough, start with the Research Workbench User Guide.

flowchart LR
    A["Desktop interface"] --> B["Audited app bridge"]
    B --> C["Python SDK"]
    C --> D["Simulation and validation"]
    D --> E["Normal run bundle"]

Scope

The desktop app is research software, not a medical device. It must not be used for dosing, diagnosis, treatment decisions, or real-time patient care.

Install Or Download¶

Packaged beta downloads use the stable desktop-beta-latest release tag:

Platform	Direct download
Windows	IINTS-AF-Desktop-Beta-windows-x64.exe
macOS	IINTS-AF-Desktop-Beta-macos.dmg
Linux	IINTS-AF-Desktop-Beta-linux-x64

For platform-specific startup and security guidance, use Desktop App Installation.

For a Python installation:

python -m pip install --upgrade "iints-sdk-python35[desktop-all]"
iints-desktop

The desktop-all extra installs PySide6, reports, MDMP, research/AI libraries, edge/serial support, libRoadRunner, and FMPy. Packaged .exe, .dmg, and Linux builds contain the supported Python-side runtime already; do not install PySide6 or these libraries into a packaged app.

COPASI, OpenCOR, and Ollama are separate system applications. Their connectors, validation boundaries, and readiness checks ship with the app, but their binaries and model files are not silently downloaded. This keeps licences, native-code trust, model size, and reproducibility decisions visible to the researcher.

Workbench Areas¶

Area	Purpose	Output or effect
Simulation	choose a curated protocol, output folder, and seed	normal SDK run bundle
Results	open a run CSV, graph glucose, and inspect a bounded table preview	read-only review of generated artifacts
Reproducibility	create RO-Crate metadata, checksums, a source snapshot, and an academic audit	local FAIR-oriented package beside the run
AI Review	start/check local Ollama and ask about a loaded result	non-authoritative review text
Run Archive	find recent local runs and reopen their artifacts	local history index
Data/MDMP	apply an explicit certification contract to a CSV	certification and validation artifacts
Biology/Research	view structures and run independent SBML, COPASI, CellML, FMI, and BindingDB workflows	explanatory, mechanistic, device-physics, and assay-evidence artifacts
Methods/Updates	inspect versions, diagnostics, logs, and update routes	transparent maintenance output

The exact tab names can differ between the current Qt beta and the newer Tauri prototype. Both must call SDK functions or the fixed bridge rather than reimplementing formulas in the interface.

Run A Workflow¶

Select an output workspace you can find again.
Select a protocol and read its patient, scenario, duration, and expected output.
Set or record the seed.
Start the run and keep the execution log.
Open the Results area only after the run reports completion or a recorded failure.
Inspect the CSV, metadata, report, and safety events together.

Curated protocols include baseline, meal stress, delayed absorption, clinical discussion, jury, and public-demo routes. They are starting points, not claims that every virtual trajectory is clinically representative.

Results Viewer¶

The results viewer can load results.csv, calculate a compact preview, render a glucose graph, and show selected rows. Use it for orientation.

For formal analysis, use the generated report and study tools. The viewer must not silently change thresholds, columns, missing-data handling, or time intervals.

Read Understand A Run before comparing two runs.

Local AI Review¶

When Ollama is installed, the app can:

start or connect to the local server
list installed models
prepare a selected local model
ask a research question about a compact result summary

The app should show model name, backend status, and the evidence supplied to the model. AI responses are formatted for readability, but bullets and confident wording do not make them authoritative.

Use local AI only after deterministic metrics and certification checks. See AI Assistant.

Data Certification¶

The app can create MDMP certification artifacts for selected CSV files. The user still chooses the contract and reviews the grade details.

A certificate means the recorded checks completed under that contract. It does not certify a medical device, establish clinical safety, or make a dataset unbiased.

See Certification Quickstart.

Academic Reproducibility Package¶

After loading a completed run CSV, use Create Academic Package. The app runs the same exporter as:

iints research academic-bundle path/to/completed_run

The output records checksums, run/software metadata, selected references, and unresolved review checks. It stays local and never uploads the run. A ready result means the implemented metadata checklist passed; it does not establish scientific validity, privacy clearance, or clinical safety.

The artifact licence starts as NOASSERTION. Do not replace it with the SDK's Apache-2.0 code licence unless that licence genuinely applies to every exported artifact.

Read Academic Research Workbench for the artifact definitions and publication workflow.

Biology And Evidence Tools¶

The research area can display or open explanatory evidence such as:

AlphaFold structures and PAE confidence views
STRING or Reactome pathway context
GTEx or Human Protein Atlas expression context
ClinVar or Ensembl variant information
ChEMBL, RCSB PDB, UniProt, and Open Targets records
genomics and tissue-resistance stress-test outputs

These resources help investigate biological assumptions. They do not automatically calibrate the virtual patient and never enter dosing or safety logic.

Each evidence card identifies its integration maturity. Portal means the app only opens an official site; planned means no functioning import exists yet. AlphaFold pLDDT and PAE remain structural-confidence measures and are never translated automatically into mutation severity or physiological parameters.

The Tauri shell opens approved external evidence sites in the system browser through an allowlist. It does not embed arbitrary remote pages.

Mechanistic Reference Model Lab¶

The Tauri workbench can safely inspect a local SBML file without executing it. When the optional mechanistic extra is installed, the same panel can run that external model independently through libRoadRunner and write a CSV, model summary, provenance manifest, and review report.

This workflow complements AlphaFold: structure is inspected separately from dynamic equations. It does not convert external units to mg/dL or minutes, does not replace the Python patient models, and never tunes patient parameters automatically. Read Mechanistic Reference Models before interpreting cross-model plots.

Cross-scale Reference Labs¶

The Tauri workbench also exposes four collapsible labs:

COPASI task inspection and explicitly confirmed CopasiSE execution
CellML inspection and OpenCOR standards validation
safe FMU archive inspection and explicitly trusted FMPy execution
read-only BindingDB affinity evidence by UniProt accession

COPASI and FMI execution require separate review confirmations. Rust validates the request boundary, while the Python SDK remains the source of scientific logic and provenance artifacts. FMI execution can load native code and must be limited to a reviewed model hash from a trusted publisher.

Read Cross-scale Reference Labs before using these workflows.

Updates And Logs¶

The app exposes version checks, the latest download route, and the Python SDK updater. Before updating an environment used for a paper or benchmark:

record the current SDK and app versions
archive the environment or dependency lock
finish or freeze active studies
use a dry run where available
repeat smoke and reproducibility checks after updating

Read Update The SDK.

Desktop Implementations¶

Implementation	Status	Role
PySide6/Qt	current rich Python workbench	main results, AI, and research UI
Cocoa	compact macOS packaging fallback	reliable native beta shell
Tauri + Rust	next-generation prototype	smaller audited native boundary around the Python SDK
Tkinter	compatibility fallback	minimal launcher when Qt is unavailable

For implementation and security details, see Tauri Desktop Shell, Desktop Signing, and Developer Portal.

When To Use The CLI Instead¶

Prefer the CLI for:

scripted or batch studies
CI and reproducibility checks
long-running Jetson training
exact command capture for a paper
advanced options not exposed by the workbench

The desktop app and CLI should produce compatible SDK artifacts. If they disagree, treat that as a bug and preserve both logs.