Launch Model Visualization Platform App

run_mvp() launches the MVP app. mvp_app() returns the app as a Shiny app object without running it, e.g. for deployment on Posit Connect or shinyapps.io (an app.R containing MVPapp::mvp_app() is sufficient).

The app is evaluated in its own environment (whose parent gives access to MVPapp and its key dependencies), so neither the global environment nor the search path is modified. Global options changed by the app are set when it starts and restored when it stops.

Usage

mvp_app(
  appDir = system.file("shiny", package = "MVPapp"),
  insert_watermark = TRUE,
  max_dataset_size = 200,
  authentication_code = NA_character_,
  internal_version = TRUE,
  pw_models_path = NA_character_,
  llm_choices = c("Claude", "Gemini", "OpenAI", "OpenRouter", "OpenAI-Compatible",
    "DeepSeek", "Azure OpenAI", "AWS Bedrock"),
  api_upload = NA_character_,
  api_chat = NA_character_,
  user_id = "MVP_user",
  reuse_context = FALSE,
  model_gemini = "gemini-3.8-flash",
  model_openai = "gpt-6.1-sol",
  model_anthropic = "claude-sonnet-5-5",
  model_openrouter = "arcee-ai/trinity-large-preview:free",
  model_openai_compatible = "gpt-5.2",
  model_deepseek = "deepseek-flash",
  model_apollo = "claude_4_6_sonnet",
  model_azure = "gpt-5.2",
  model_aws = "anthropic.claude-sonnet-5",
  temperature = 1,
  llm_seed = 42,
  model_lang = c("mrgsolve", "nonmem", "rxode2"),
  prompts_path = system.file("shiny/prompts.R", package = "MVPapp"),
  show_debugging_msg = FALSE,
  brand_color = NULL,
  brand_text_color = NULL,
  brand_logo = NULL,
  brand_favicon = NULL,
  page_title = NULL,
  ...
)

run_mvp(
  appDir = system.file("shiny", package = "MVPapp"),
  insert_watermark = TRUE,
  max_dataset_size = 200,
  authentication_code = NA_character_,
  internal_version = TRUE,
  pw_models_path = NA_character_,
  llm_choices = c("Claude", "Gemini", "OpenAI", "OpenRouter", "OpenAI-Compatible",
    "DeepSeek", "Azure OpenAI", "AWS Bedrock"),
  api_upload = NA_character_,
  api_chat = NA_character_,
  user_id = "MVP_user",
  reuse_context = FALSE,
  model_gemini = "gemini-3.8-flash",
  model_openai = "gpt-6.1-sol",
  model_anthropic = "claude-sonnet-5-5",
  model_openrouter = "arcee-ai/trinity-large-preview:free",
  model_openai_compatible = "gpt-5.2",
  model_deepseek = "deepseek-flash",
  model_apollo = "claude_4_6_sonnet",
  model_azure = "gpt-5.2",
  model_aws = "anthropic.claude-sonnet-5",
  temperature = 1,
  llm_seed = 42,
  model_lang = c("mrgsolve", "nonmem", "rxode2"),
  prompts_path = system.file("shiny/prompts.R", package = "MVPapp"),
  show_debugging_msg = FALSE,
  brand_color = NULL,
  brand_text_color = NULL,
  brand_logo = NULL,
  brand_favicon = NULL,
  page_title = NULL,
  ...
)

Arguments

appDir: Character. Directory containing the app's app.R. Defaults to the copy installed with MVPapp and rarely needs changing.
insert_watermark: Logical. Default TRUE. Set to FALSE to remove the "For Internal Use Only" watermark from simulation plots and NCA reports.
max_dataset_size: Numeric. Default 200. Maximum upload dataset size in MB.
authentication_code: Character. Default NA_character_. Provide a string (e.g., password) to password-lock the entire app.
internal_version: Logical. Default TRUE. Set to FALSE when hosting the app on a server (e.g. AWS) where the working directory is not writable: NCA reports are then written to the session's temporary directory, and the help text for locked models points to GitHub instead. For internal deployments, it also selects the external Apollo endpoint.
pw_models_path: Character. Default NA_character_. Path to an R file defining password-gated models (see system.file("shiny/passworded_models_example.R", package = "MVPapp") and vignette("supply-passwords")).
llm_choices: Character vector of LLM providers offered in the app; the first one is selected by default. Defaults to all supported providers: "Claude", "Gemini", "OpenAI", "OpenRouter", "OpenAI-Compatible", "DeepSeek", "Azure OpenAI" and "AWS Bedrock". Internal-deployment providers (e.g. "PMx Co-Modeler", "Apollo") are also accepted.
api_upload: Character. Default NA_character_. API upload URL when using Dify-based workflows (internal deployment only)
api_chat: Character. Default NA_character_. Base URL required when using OpenAI-Compatible providers, e.g. "https://api.moonshot.ai/v1" ("/chat/completions" is appended automatically)
user_id: Character. Default MVP_user. User ID for LLMs
reuse_context: Logical. Default FALSE. Set to TRUE to re-use same conversation during retries.
model_gemini: Character. Default gemini-3.8-flash. Model for Gemini.
model_openai: Character. Default gpt-6.1-sol. Model for OpenAI / ChatGPT.
model_anthropic: Character. Default claude-sonnet-5-5. Model for Anthropic / Claude.
model_openrouter: Character. Default arcee-ai/trinity-large-preview:free. Model for OpenRouter.
model_openai_compatible: Character. Default gpt-5.2. Model for OpenAI-compatible providers.
model_deepseek: Character. Default deepseek-flash. Model for DeepSeek.
model_apollo: Character. Default claude_4_6_sonnet. Model for Apollo (internal deployment only).
model_azure: Character. Default gpt-5.2. Model for Azure OpenAI.
model_aws: Character. Default anthropic.claude-sonnet-5. Model for AWS Bedrock.
temperature: Numeric. Default 1. Between 0 (more deterministic) and 1 (more creative) (not all models support this).
llm_seed: Numeric. Default 42. Seed number for LLMs (not all models support this).
model_lang: Character. Default "mrgsolve" (the first element). Model translation language: "mrgsolve", "nonmem" (for testing only) or "rxode2" (for testing only).
prompts_path: Character. Path to an R file defining the LLM prompts. Defaults to the prompts installed with MVPapp. See vignette("automatic-translation") for more information.
show_debugging_msg: Logical. Default FALSE. Set to TRUE to output verbose working messages in the console, useful for debugging. AI translation responses are then also kept for review (see Details).
brand_color: Character. Default NULL (the default navy). A colour, e.g. "#F36633" or "orange", used for the navbar, active tab, box headers, table headers and loading spinners. See "Branding" in Details.
brand_text_color: Character. Default NULL. Colour of the text on brand_color (navbar and box headers). By default white or black is chosen for readability. Ignored if brand_color is NULL.
brand_logo: Character. Default NULL (no logo). Path to an image file (png, jpg, svg, ...) or a web URL, shown in the navbar in front of the page title.
brand_favicon: Character. Default NULL (the MVP icon shipped with the package). Path to an image file (png, ico, svg) or a web URL, used as the browser tab icon.
page_title: Character. Default NULL ("Model Visualization Platform (MVP)"). Title shown in the navbar. When set, it is also used as the browser tab title.
...: For run_mvp(), shiny::runApp() arguments (e.g. launch.browser, port). launch.browser defaults to TRUE here, so the app opens in your web browser (also from RStudio, which would otherwise use its viewer pane); use launch.browser = FALSE to turn that off. Any other named arguments are passed to the app.

Value

run_mvp() is called for its side effect of launching the app and returns the value of shiny::runApp() invisibly once the app is closed. mvp_app() returns a Shiny app object (class shiny.appobj).

Details

Objects kept for debugging. MVPapp does not write to the global environment. Objects useful for debugging are kept in an internal environment, MVPapp:::.mvp_state, and remain available in the R session after the app is closed. List them with ls(MVPapp:::.mvp_state):

last_ev_df$model_1, last_ev_df$model_2: the dosing event tables used in each model's most recent simulation (also used for the reproducible bundle download).
deps_env: an internal cache used to run the app; not useful for debugging.
With show_debugging_msg = TRUE, AI translation responses: llm_result (initial translation), llm_refine (latest retry that still failed), llm_success (Dify's reply when told the code compiled), branch_model / branch_refine (raw answers before cleaning) and ellmer_chat_response / dify_chat_response (raw chat responses).

Branding. For a company deployment, the title, colour, logo and favicon can be set when launching, without editing any package files. Each is optional; with all of them left at NULL the app looks as it always did.

run_mvp(brand_color   = "#F36633",
        brand_logo    = "path/to/company_logo.png",
        brand_favicon = "path/to/favicon.png",
        page_title    = "Company Model Visualization Platform")

Image files are read from where they are and served by the app, so they do not need to be inside the package (a web URL also works). A logo with a transparent background and a height of about 30 pixels or more works best. brand_color replaces the navy in the navbar, the active tab, the headers and borders of the panels, the data table headers and the loading spinners. The same arguments work with mvp_app() for hosted deployments.

Hosting and security. MVPapp evaluates R and C++ model code supplied in the browser, so anyone who can open the app can run code on the server under the server account. authentication_code and the model passwords are a convenience to limit access to the interface and to hide models, not a security boundary: host the app only for trusted users, in a sandboxed account or container. The same applies to AI translation: its output is compiled automatically, and compiling mrgsolve code can itself run commands, so do not translate documents from untrusted sources on a shared server.

Without installing MVPapp (e.g. where package installation is restricted), the development version can also be launched from GitHub, provided its dependencies are installed: shiny::runGitHub("MVPapp", "stevechoy", subdir = "inst/shiny").

Examples

if (interactive()) {
  run_mvp(insert_watermark = FALSE) # remove watermarks

  # Other examples:
  # run_mvp(launch.browser = FALSE) # do not open a browser (argument passed to runApp)
  # run_mvp(authentication_code = "some_password") # Password-lock the site,
  #                                                # could be useful in deployment
  # run_mvp(pw_models_path = "path/to/your/private/models.R") # see
  #   system.file("shiny/passworded_models_example.R", package = "MVPapp")
  # run_mvp(llm_choices = "Claude", model_anthropic = "claude-sonnet-5-5",
  #         temperature = 0) # Uses Claude only with reproducible results
}

# App object, e.g. as the only line of an app.R file for deployment:
app <- mvp_app(insert_watermark = FALSE)