Unable to run QueryChat inside shinyModule

[jsinnett]

QueryChat doesn't appear to work properly when using it inside a module. The chat spinner just spins indefinitely.

In the examples I've found so far, the qc object is initialized at the global level. However, it works as expected when initialized in the server w/o shiny modules. My motivation for doing it this way is to be able to pass data.tables/data.frames that are created within an app. I also noticed that when using renderUI, the greeting message does not appear and am wondering if it's related to this issue. Ultimately, I think it's a name spacing issue with mod_ui/server but I can't track it down.

Really appreciate any help!

Reprex below

library(shiny)
library(bslib)
library(querychat)
library(ellmer)

options(duckdb.debug = T)

# Module UI
test_module_ui <- function(id) {
  ns <- NS(id)
  page_fluid(
    card(
      DT::DTOutput(ns("dt"))
    ),
    card(
      uiOutput(ns("querychat_sidebar"))
    )
  )
}

# Module Server
test_module_server <- function(id) {
  moduleServer(id, function(input, output, session) {
    ns <- session$ns

    message("Module ID: ", id)
    message("Session namespace: ", session$ns(NULL))

    # Create QueryChat instance
    qc <- QueryChat$new(
      mtcars,
      id = id,
      client = ellmer::chat_aws_bedrock(model = "us.anthropic.claude-sonnet-4-5-20250929-v1:0"),
      greeting = "Ask me about mtcars!"
    )

    message("QueryChat created with ID: ", qc$id)

    # Initialize the QueryChat server
    qc_vals <- qc$server()

    # Render the QueryChat UI
    output$querychat_sidebar <- renderUI({
      qc$ui()
    })

    # Render the data table
    output$dt <- DT::renderDT({
      DT::datatable(qc_vals$df())
    })
  })
}

# Main App UI
ui <- fluidPage(
  titlePanel("QueryChat Module Test"),
  test_module_ui("test_module")
)

# Main App Server
server <- function(input, output, session) {
  test_module_server("test_module")
}

shinyApp(ui, server)

Chat hangs

[cpsievert]

If you don't mind, it might help you could add a bit more context about why you're needing modules. Is it true that you're wanting each module to dynamically render multlple chat interfaces to the same dataset?

[jsinnett]

I'm working on a relatively complex shiny app and am interested in leveraging the latest LLM packages you all are building. I have two current use cases:

1. to give users the ability to get a high level summary of the data they have subsetted down via input selections / dimensions to group by etc. This is handled via ellmer where I effectively pass in summarized data as the context. I have accompanied plots to visualize their selections.

2. I am also working on a new chat feature that allows users to explore the original dataset pre input selection. Hopefully similar to what Databot does. My existing "chat bot" uses the same summarized data being passed into the high level summary module but it is quite limiting when users want to ask questions beyond the summarized data because claude has no context there. I have not tested sending a mid-sized table with ~50k rows / 10 cols through ellmer but I suspect it could be unwieldy. So I want to use querychat to automatically aggregate the data based on the users questions and send via ellmer either summarized or as is. Each dataset or data.table is held in a nested reactiveValues obj. Given that, I'm interested in building a querychat shinyModule for this feature that can be used across tabs or a collection of tabs with different datasets.

May have provided too much unrelated detail here but hopefully you get the gist of what I'm trying to do.

[cpsievert]

Thanks @jsinnett, I think it does indeed makes sense for us to have this scenario "just work". Here is a more minimal reprex:

library(shiny)
library(bslib)
library(querychat)

module_ui <- function(id) {
  ns <- NS(id)
  card(uiOutput(ns("chat_ui")))
}

module_server <- function(id) {
  moduleServer(id, function(input, output, session) {
    qc <- QueryChat$new(mtcars, greeting = "Ask me about mtcars!")
    qc$server()
    output$chat_ui <- renderUI({
      qc$ui()
    })
  })
}

ui <- page_fluid(
  titlePanel("QueryChat Module Test"),
  module_ui("module1")
)

server <- function(input, output, session) {
  module_server("module1")
}

shinyApp(ui, server)

The reason the above can "just work" is because $ui() is called within the session context (and thus, we can auto-detect the namespace).

If, however, you want to call $ui() outside of a session context, you'll need to pass along the namespace. Both the fix mentioned above and ns argument should come soon as a part of #172

library(shiny)
library(bslib)
library(querychat)

qc <- QueryChat$new(mtcars, greeting = "Ask me about mtcars!")

module_ui <- function(id, qc) {
  ns <- NS(id)
  card(qc$ui(ns = ns))
}

module_server <- function(id, qc) {
  moduleServer(id, function(input, output, session) {
    qc$server()
  })
}

ui <- page_fluid(
  titlePanel("QueryChat Module Test"),
  module_ui("module1", qc)
)

server <- function(input, output, session) {
  module_server("module1", qc)
}

shinyApp(ui, server)