Skip to contents

Configure the generative UI canvas

Source: R/genui_components.R
canvas_options.Rd

Choose which components the canvas in codeagent_app() offers and how strictly model-authored text is checked. Components are the same objects shinygenui uses (shinygenui::genui_component()); build your own with canvas_component_output() or canvas_component_widget(), or pass any existing genui_component.

Usage

canvas_options(
  components = list(),
  starter = TRUE,
  data_env = globalenv(),
  csp = TRUE,
  validate_text = TRUE
)

Arguments

components

A list of genui_component objects supplied by the host.

starter

Which of shinygenui's own components to offer: TRUE (all of them, the default), FALSE (none), or a character vector of their names ("value_box", "data_table", "scatter_plot", "histogram", "markdown_card", "card_row").

data_env

Environment in which dataset names are looked up.

csp

Content Security Policy for the page while the canvas is available: TRUE (default; images, media, frames and plugins only from the app itself), FALSE (none), or a policy string of your own – for example when a component loads map tiles from another origin. The policy applies to the whole host page, including the host's own content.

validate_text

If TRUE (default), every string the model passes to a canvas tool is refused when it contains markup, links, URLs or control characters, before anything reaches the browser. This guards against the model (or a prompt injection) placing data in a URL the browser would fetch. Set FALSE when your components need the model to supply URLs.

Value

A codeagent_canvas_options object for codeagent_app(canvas = ).

Details

Host components are trusted code: codeagent registers them as canvas tools (read-only, behind the central permission gate), runs them through the same transactional executor, Data Shield scanning and strict session restore as the built-in ones, and leaves what they render to you.

See also

canvas_component_output(), canvas_component_widget()

Examples

if (FALSE) { # \dontrun{
boxplot_card <- canvas_component_output(
  "boxplot", "Box plot of a numeric column by group.",
  args = list(y = ellmer::type_string("Numeric column."),
              by = ellmer::type_string("Grouping column.")),
  output = shiny::plotOutput,
  render = function(args, data) shiny::renderPlot(
    boxplot(reformulate(args$by, args$y), data = data())))
codeagent_app(chat, canvas = canvas_options(
  components = list(boxplot_card), starter = c("data_table", "histogram")))
} # }