Package 'shinychat'

Description

Create a simple Shiny app for live chatting using an ellmer::Chat object. Note that these functions will mutate the input client object as you chat because your turns will be appended to the history.

The app created by chat_app() is suitable for interactive use by a single user. For multi-user Shiny apps, use chat_ui() and chat_server() and be sure to create a new chat client for each user session.

Usage

chat_app(client, ..., bookmark_store = "url", allow_attachments = TRUE)

chat_server(
  id,
  client,
  greeting = NULL,
  history = TRUE,
  bookmark_on_input = lifecycle::deprecated(),
  bookmark_on_response = lifecycle::deprecated(),
  session = shiny::getDefaultReactiveDomain()
)

Arguments

Value

• chat_app() returns a shiny::shinyApp() object.

• chat_server() includes the shinychat server logic, and returns an environment containing:

  • last_input: A reactive value containing the last user input (a string when attachments are disabled, a list of ellmer Content objects when enabled).

  • last_turn: A reactive value containing the last assistant turn.

  • update_user_input(): A function to update the chat input or submit a new user input. Takes the same arguments as update_chat_user_input(), except for id and session, which are supplied automatically.

  • append(): A function to append a new message to the chat UI. Takes the same arguments as chat_append(), except for id and session, which are supplied automatically.

  • clear(): A function to clear the chat history and the chat UI. clear() takes an optional list of messages used to initialize the chat after clearing. messages should be a list of messages, where each message is a list with role and content fields. The client_history argument controls how the chat client's history is updated after clearing. It can be one of: "clear" the chat history; "set" the chat history to messages; "append" messages to the existing chat history; or "keep" the existing chat history.

  • set_greeting(): A function to set, stream, or clear the chat greeting. Pass a chat_greeting() object, a plain string, or NULL to clear. Streaming greetings run inside an shiny::ExtendedTask so the session stays responsive; if called while a greeting is already streaming, the new greeting is queued. If the greeting has already been dismissed, calling set_greeting() updates the content but does not make it visible again; call clear(greeting = TRUE) first to show a new greeting after dismissal.

  • status: A reactive value indicating the current chat interaction state. Returns "idle" when no response is in progress, or "streaming" while a response is actively being received.

  • client: The current chat client object (an active binding that always reflects the latest client, even after set_client() is called).

  • set_client(new_client, sync = TRUE): Replace the chat client used by the module. When sync is TRUE (the default), the new client inherits conversation turns, system prompt, and tools from the previous client so the conversation continues seamlessly. Set sync = FALSE to use the new client as-is. If a response is currently streaming, the swap is deferred until the stream completes. If called multiple times while streaming, only the most recent new client is used.

  • slash_command(name, description, handler, ..., echo, force): Register a slash command. handler is required: pass a function (taking 0 or 1 argument), or NULL for a client-side command handled in JavaScript via the shiny:chat-slash-command DOM event. A handler that takes one argument receives a ContentSlashCommand object (not a plain string). See ContentSlashCommand for details on how to use this object to preserve the original command text across bookmarks. echo controls whether invoking the command is echoed as a user message and awaits a response; it defaults to TRUE when a handler is given and FALSE otherwise (set echo = FALSE for a handler that only performs side effects). Returns a function that removes the command. Errors if a command with the same name is already registered unless force = TRUE.

Functions

• chat_app(): A simple Shiny app for live chatting. Note that this app is suitable for interactive use by a single user; do not use chat_app() in a multi-user Shiny app context.

• chat_server(): Wire up batteries-included chat server logic in a Shiny session.

Greeting

When greeting is a function, it is called each time the greeting_requested event fires — on first view when the chat is empty, and again after clear(greeting = TRUE). The function should return a chat_greeting() (typically wrapping a stream). Static values (strings, chat_greeting() objects) are set once at init and do not regenerate.

The function signature determines what is passed. Currently the only recognized argument is client.

⁠function(client)⁠ (recommended). A clone of the client with its turn history wiped is passed as client. This avoids manually creating and configuring a separate client:

chat_server("chat", client, greeting = function(client) {
  stream <- client$stream_async("Generate a short welcome message.")
  chat_greeting(stream)
})

⁠function()⁠ (zero arguments). You create and manage your own client:

chat_server("chat", client, greeting = function() {
  greeter <- ellmer::chat_openai(model = "gpt-4o")
  stream <- greeter$stream_async("Generate a short welcome message.")
  chat_greeting(stream)
})

Static value. Set once; does not regenerate after clear():

chat_server("chat", client, greeting = "## Welcome!\n\nHow can I help?")

The returned set_greeting() helper is available for cases where you need to set a greeting outside the greeting lifecycle.

Examples

## Not run: 
# Interactive in the console ----
client <- ellmer::chat_anthropic()
chat_app(client)

# Inside a Shiny app ----
library(shiny)
library(bslib)
library(shinychat)

ui <- page_fillable(
  titlePanel("shinychat example"),

  layout_columns(
    card(
      card_header("Chat with Claude"),
      chat_ui(
        "claude",
        messages = list(
          "Hi! Use this chat interface to chat with Anthropic's `claude-3-5-sonnet`."
        )
      )
    ),
    card(
      card_header("Chat with ChatGPT"),
      chat_ui(
        "openai",
        messages = list(
          "Hi! Use this chat interface to chat with OpenAI's `gpt-4o`."
        )
      )
    )
  )
)

server <- function(input, output, session) {
  claude <- ellmer::chat_anthropic(model = "claude-3-5-sonnet-latest") # Requires ANTHROPIC_API_KEY
  openai <- ellmer::chat_openai(model = "gpt-4o") # Requires OPENAI_API_KEY

  chat_server("claude", claude)
  chat_server("openai", openai)
}

shinyApp(ui, server)

## End(Not run)

Description

The chat_append function appends a message to an existing chat_ui(). The response can be a string, string generator, string promise, or string promise generator (as returned by the 'ellmer' package's chat, stream, chat_async, and stream_async methods, respectively).

This function should be called from a Shiny app's server. It is generally used to append the client's response to the chat, while user messages are added to the chat UI automatically by the front-end. You'd only need to use chat_append(role="user") if you are programmatically generating queries from the server and sending them on behalf of the user, and want them to be reflected in the UI.

Usage

chat_append(
  id,
  response,
  role = c("assistant", "user"),
  icon = NULL,
  session = getDefaultReactiveDomain()
)

Arguments

Value

Returns a promise that resolves to the contents of the stream, or an error. This promise resolves when the message has been successfully sent to the client; note that it does not guarantee that the message was actually received or rendered by the client. The promise rejects if an error occurs while processing the response (see the "Error handling" section).

Error handling

If the response argument is a generator, promise, or promise generator, and an error occurs while producing the message (e.g., an iteration in stream_async fails), the promise returned by chat_append will reject with the error. If the chat_append call is the last expression in a Shiny observer, shinychat will log the error message and show a message that the error occurred in the chat UI.

Examples

library(shiny)
library(coro)
library(bslib)
library(shinychat)

# Dumbest chatbot in the world: ignores user input and chooses
# a random, vague response.
fake_chatbot <- async_generator(function(input) {
  responses <- c(
    "What does that suggest to you?",
    "I see.",
    "I'm not sure I understand you fully.",
    "What do you think?",
    "Can you elaborate on that?",
    "Interesting question! Let's examine thi... **See more**"
  )

  await(async_sleep(1))
  for (chunk in strsplit(sample(responses, 1), "")[[1]]) {
    yield(chunk)
    await(async_sleep(0.02))
  }
})

ui <- page_fillable(
  chat_ui("chat", fill = TRUE)
)

server <- function(input, output, session) {
  observeEvent(input$chat_user_input, {
    response <- fake_chatbot(input$chat_user_input)
    chat_append("chat", response)
  })
}

shinyApp(ui, server)

Description

For advanced users who want to control the message chunking behavior. Most users should use chat_append() instead.

Usage

chat_append_message(
  id,
  msg,
  chunk = TRUE,
  operation = c("append", "replace"),
  icon = NULL,
  session = getDefaultReactiveDomain()
)

Arguments

Value

Returns nothing (invisible(NULL)).

Examples

library(shiny)
library(coro)
library(bslib)
library(shinychat)

# Dumbest chatbot in the world: ignores user input and chooses
# a random, vague response.
fake_chatbot <- async_generator(function(id, input) {
  responses <- c(
    "What does that suggest to you?",
    "I see.",
    "I'm not sure I understand you fully.",
    "What do you think?",
    "Can you elaborate on that?",
    "Interesting question! Let's examine thi... **See more**"
  )

  # Use low-level chat_append_message() to temporarily set a progress message
  chat_append_message(id, list(role = "assistant", content = "_Thinking..._ "))
  await(async_sleep(1))
  # Clear the progress message
  chat_append_message(id, list(role = "assistant", content = ""), operation = "replace")

  for (chunk in strsplit(sample(responses, 1), "")[[1]]) {
    yield(chunk)
    await(async_sleep(0.02))
  }
})

ui <- page_fillable(
  chat_ui("chat", fill = TRUE)
)

server <- function(input, output, session) {
  observeEvent(input$chat_user_input, {
    response <- fake_chatbot("chat", input$chat_user_input)
    chat_append("chat", response)
  })
}

shinyApp(ui, server)

Description

Reads a file, base64-encodes its contents, and returns a list in the format expected by the attachments argument of update_chat_user_input().

Usage

chat_attachment(path, mime = NULL, name = NULL)

Arguments

Value

A list with elements mime, name, size, and data_url, ready to pass as an element of the attachments argument of update_chat_user_input().

Description

Removes all messages from the chat UI. Set greeting = TRUE to also clear the greeting, which re-triggers greeting_requested (see the Greeting section in chat_ui()).

Usage

chat_clear(id, greeting = FALSE, session = getDefaultReactiveDomain())

Arguments

Examples

library(shiny)
library(bslib)

ui <- page_fillable(
  chat_ui("chat", fill = TRUE),
  actionButton("clear", "Clear chat")
)

server <- function(input, output, session) {
  observeEvent(input$clear, {
    chat_clear("chat")
  })

  observeEvent(input$chat_user_input, {
    response <- paste0("You said: ", input$chat_user_input)
    chat_append("chat", response)
  })
}

shinyApp(ui, server)



library(shiny)
library(bslib)
library(shinychat)

# Regenerate greeting on clear
ui <- page_fillable(
  chat_ui("chat"),
  actionButton("new_chat", "New chat")
)

server <- function(input, output, session) {
  observeEvent(input$chat_greeting_requested, {
    chat_set_greeting("chat", "## Welcome!\n\nHow can I help?")
  })

  observeEvent(input$new_chat, {
    # Clearing with greeting = TRUE triggers greeting_requested again
    chat_clear("chat", greeting = TRUE)
  })

  observeEvent(input$chat_user_input, {
    chat_append("chat", paste0("You said: ", input$chat_user_input))
  })
}

shinyApp(ui, server)

Description

Enable conversation history for a chat

Usage

chat_enable_history(
  id,
  client,
  ...,
  on_save = NULL,
  on_restore = NULL,
  options = history_options(),
  restore_ui = TRUE,
  session = shiny::getDefaultReactiveDomain()
)

Arguments

Value

Invisibly, a function that cancels all history registrations.

Description

Get the current greeting content

Usage

chat_get_greeting(id, session = getDefaultReactiveDomain())

Arguments

Value

A character string with the current greeting content, or NULL if no greeting is set or has been cleared.

Description

Creates a greeting object for use with chat_ui() or chat_set_greeting(). A greeting is displayed when the chat first loads and is dismissed when the user sends their first message.

Usage

chat_greeting(
  content,
  ...,
  persistent = FALSE,
  dismissible = lifecycle::deprecated()
)

Arguments

Value

An S3 object of class "chat_greeting".

Patterns

Persistent greeting (stays visible after the user sends a message):

chat_greeting("Please read our [terms of service](https://example.com).", persistent = TRUE)

Greeting with suggestion cards (clickable chips that fill the input):

chat_greeting(paste(
  "## Welcome!\n\n",
  "Try one of these:\n\n",
  '<span class="suggestion">Summarize my data</span>\n',
  '<span class="suggestion">Create a plot</span>\n',
  '<span class="suggestion">Explain this code</span>'
))

Greeting with HTML tags (Shiny inputs/outputs):

chat_greeting(htmltools::tagList(
  htmltools::h2("Welcome!"),
  shiny::selectInput("model", "Choose a model:", c("gpt-4o", "claude-3"))
))

Examples

library(shiny)
library(bslib)
library(shinychat)

ui <- page_fillable(
  chat_ui(
    "chat",
    greeting = chat_greeting("## Welcome!\n\nHow can I help you today?")
  )
)

server <- function(input, output, session) {
  observeEvent(input$chat_user_input, {
    response <- paste0("You said: ", input$chat_user_input)
    chat_append("chat", response)
  })
}

shinyApp(ui, server)

Description

Adds Shiny bookmarking hooks to save and restore the ellmer chat client. Also restores chat messages from the history in the client.

If either bookmark_on_input or bookmark_on_response is TRUE, the Shiny App's bookmark will be automatically updated without showing a modal to the user.

Note: The client's chat state and the greeting content are both saved/restored automatically. If the client's state doesn't properly capture the chat's UI (i.e., a transformation is applied in-between receiving and displaying the message), you may need to implement your own session$onRestore() (and possibly session$onBookmark) handler to restore any additional state.

To avoid restoring chat history from the client, you can ensure that the history is empty by calling client$set_turns(list()) before passing the client to chat_restore().

chat_restore() bookmarks the whole session and doesn't know about multiple conversations. If you need per-conversation history (the chat history drawer, switching between saved conversations), use chat_enable_history() with history_options(restore_mode = "bookmark") instead — it replaces chat_restore()'s job for history-aware apps. The two are mutually exclusive; chat_app() picks one or the other based on whether history is set.

Usage

chat_restore(
  id,
  client,
  ...,
  bookmark_on_input = TRUE,
  bookmark_on_response = TRUE,
  restore_ui = TRUE,
  session = getDefaultReactiveDomain()
)

Arguments

Value

Invisibly returns a function that, when called, cancels all bookmark registrations made by this call. This is useful when swapping the chat client: cancel the previous bookmarks, then call chat_restore() again with the new client.

Examples

library(shiny)
library(bslib)
library(shinychat)

ui <- function(request) {
  page_fillable(
    chat_ui("chat", fill = TRUE)
  )
}

server <- function(input, output, session) {
  chat_client <- ellmer::chat_ollama(
    system_prompt = "Important: Always respond in a limerick",
    model = "qwen2.5-coder:1.5b",
    echo = TRUE
  )
  # Update bookmark to chat on user submission and completed response
  chat_restore("chat", chat_client)

  observeEvent(input$chat_user_input, {
    stream <- chat_client$stream_async(input$chat_user_input)
    chat_append("chat", stream)
  })
}

# Enable bookmarking!
shinyApp(ui, server, enableBookmarking = "server")

Description

Sets or clears the greeting displayed in an existing chat_ui(). The greeting is shown when the chat first loads and is dismissed when the user sends their first message.

Call chat_set_greeting() from the server to display a dynamic or streaming greeting. This is typically used inside an shiny::observeEvent() watching ⁠input$<id>_greeting_requested⁠ – an event that fires when the chat is visible, has no messages, and has no greeting set. See the Greeting section in chat_ui() for details on greeting_requested.

If the greeting has already been dismissed, calling this function updates the greeting content but does not make it visible again. To show a new greeting after dismissal, first clear the chat with chat_clear(id, greeting = TRUE).

Pass NULL to clear the current greeting entirely.

Usage

chat_set_greeting(id, greeting, session = getDefaultReactiveDomain())

Arguments

Value

Returns invisible(NULL) for static and NULL greetings, or a promise for streaming greetings (resolves when streaming is complete).

Examples

library(shiny)
library(bslib)
library(shinychat)

# Static greeting set from the server
ui <- page_fillable(chat_ui("chat"))

server <- function(input, output, session) {
  chat_set_greeting("chat", "## Welcome!\n\nHow can I help you today?")

  observeEvent(input$chat_user_input, {
    chat_append("chat", paste0("You said: ", input$chat_user_input))
  })
}

shinyApp(ui, server)


library(shiny)
library(bslib)
library(shinychat)
library(coro)

# Streaming greeting generated by a function
greeting_generator <- async_generator(function() {
  for (chunk in strsplit("## Hello!\n\nHow can I help you?", "")[[1]]) {
    yield(chunk)
    await(async_sleep(0.02))
  }
})

ui <- page_fillable(chat_ui("chat"))

server <- function(input, output, session) {
  chat_set_greeting("chat", chat_greeting(greeting_generator()))

  observeEvent(input$chat_user_input, {
    chat_append("chat", paste0("You said: ", input$chat_user_input))
  })
}

shinyApp(ui, server)


library(shiny)
library(bslib)
library(shinychat)

# LLM-generated greeting using greeting_requested
ui <- page_fillable(chat_ui("chat"))

server <- function(input, output, session) {
  chat_client <- ellmer::chat_openai(model = "gpt-4o")

  observeEvent(input$chat_greeting_requested, {
    stream <- chat_client$stream_async(
      "Generate a short, friendly welcome message."
    )
    chat_set_greeting("chat", chat_greeting(stream))
  })

  observeEvent(input$chat_user_input, {
    stream <- chat_client$stream_async(input$chat_user_input)
    chat_append("chat", stream)
  })
}

shinyApp(ui, server)


library(shiny)
library(bslib)
library(shinychat)

# Regenerate pattern: chat_clear(greeting = TRUE) triggers greeting_requested
ui <- page_fillable(
  chat_ui("chat"),
  actionButton("regenerate", "New greeting")
)

server <- function(input, output, session) {
  observeEvent(input$chat_greeting_requested, {
    chat_set_greeting(
      "chat",
      paste("## Welcome!\n\nGenerated at", Sys.time())
    )
  })

  observeEvent(input$regenerate, {
    chat_clear("chat", greeting = TRUE)
  })

  observeEvent(input$chat_user_input, {
    chat_append("chat", paste0("You said: ", input$chat_user_input))
  })
}

shinyApp(ui, server)

Description

Inserts a chat UI element into a Shiny UI, which includes a scrollable section for displaying chat messages, and an input field for the user to enter new messages.

To respond to user input, listen for input$ID_user_input (for example, if id="my_chat", user input will be at input$my_chat_user_input), and use chat_append() to append messages to the chat.

Usage

chat_ui(
  id,
  ...,
  messages = NULL,
  greeting = NULL,
  placeholder = "Enter a message...",
  width = "min(680px, 100%)",
  height = "auto",
  fill = TRUE,
  icon_assistant = NULL,
  enable_cancel = NULL,
  submit_key = c("enter", "enter+modifier"),
  allow_attachments = NULL,
  footer = NULL
)

Arguments

Value

A Shiny tag object, suitable for inclusion in a Shiny UI

Greeting

A greeting is an optional welcome message shown before any conversation messages. It is automatically dismissed when the user sends their first message (unless created with persistent = TRUE).

Static greeting. Pass a string or chat_greeting() to the greeting parameter:

chat_ui("chat", greeting = "## Hello!\n\nHow can I help you today?")

Dynamic greeting from the server. Leave greeting unset and use chat_set_greeting() from your server function. This is useful when the greeting depends on session state or is generated by a model.

greeting_requested input. When the chat is visible on the page, has no messages, and has no greeting set, Shiny fires ⁠input$<id>_greeting_requested⁠ (e.g. input$chat_greeting_requested for chat_ui("chat")). The value is an event counter suitable for shiny::observeEvent(). Use it to trigger server-side greeting generation:

observeEvent(input$chat_greeting_requested, {
  stream <- chat_client$stream_async("Generate a short welcome message.")
  chat_set_greeting("chat", chat_greeting(stream))
})

This input fires when the chat component is first viewed on the page and empty, and again after chat_clear() (greeting = TRUE), enabling a regenerate pattern where clearing the greeting automatically triggers a fresh one.

greeting_dismissed input. When the user dismisses the greeting, ⁠input$<id>_greeting_dismissed⁠ fires with a Date.now() timestamp. If the greeting is later cleared after being dismissed, the input resets to NULL. If you use chat_server(), you can access the greeting_dismissed reactive from the returned value instead of the raw namespaced input string.

Thinking display

When a model produces reasoning or "thinking" tokens, shinychat renders them in a collapsible panel above the response. The panel shows a live stream of the model's reasoning while it thinks, then auto-collapses when the response begins.

Thinking display works automatically with any model that supports it. Two paths are supported:

1. ellmer's ContentThinking objects. Models that provide a structured thinking API (e.g., Claude with extended thinking) emit ContentThinking objects when you stream with stream = "content". shinychat detects these and routes them to the thinking panel. This is what chat_append() uses internally when you pass it an ellmer content stream.

2. Raw ⁠<thinking>⁠ tags. Many open-source and local models (DeepSeek, QwQ, Qwen, etc.) emit ⁠<thinking>...</thinking>⁠ tags directly in their markdown output. shinychat detects these tags during streaming and renders the enclosed text in the thinking panel with no extra configuration.

Topic labels

You can optionally get labeled sub-sections within the thinking panel by asking the model to emit ⁠<topic>...</topic>⁠ tags in its reasoning. These are extracted and rendered as section headings inside the thinking display, and the current topic appears in the collapsed header as a live status.

To use topic labels, add something like this to your system prompt:

When thinking through a problem, wrap brief topic labels in <topic> tags
to indicate what you're currently reasoning about. For example:
<topic>parsing the input</topic>

Topic labels are entirely optional. Without them, the thinking panel still works – it just won't have sub-section headings.

Examples

library(shiny)
library(bslib)
library(shinychat)

ui <- page_fillable(
  chat_ui("chat", fill = TRUE)
)

server <- function(input, output, session) {
  observeEvent(input$chat_user_input, {
    # In a real app, this would call out to a chat client or API,
    # perhaps using the 'ellmer' package.
    response <- paste0(
      "You said:\n\n",
      "<blockquote>",
      htmltools::htmlEscape(input$chat_user_input),
      "</blockquote>"
    )
    chat_append("chat", response)
    chat_append("chat", stream)
  })
}

shinyApp(ui, server)

Description

Format ellmer content for shinychat

Usage

contents_shinychat(content)

Arguments

Value

Returns text, HTML, or web component tags formatted for use in chat_ui().

Extending contents_shinychat()

You can extend contents_shinychat() to handle custom content types in your application. contents_shinychat() is an S7 generic. If you haven't worked with S7 before, you can learn more about S7 classes, generics and methods in the S7 documentation.

We'll work through a short example creating a custom display for the results of a tool that gets local weather forecasts. We first need to create a custom class that extends ellmer::ContentToolResult.

library(ellmer)

WeatherToolResult <- S7::new_class(
  "WeatherToolResult",
  parent = ContentToolResult,
  properties = list(
    location_name = S7::class_character
  )
)

Next, we'll create a simple ellmer::tool() that gets the weather forecast for a location and returns our custom WeatherToolResult class. The custom class works just like a regular ContentToolResult, but it has an additional location_name property.

get_weather_forecast <- tool(
  function(lat, lon, location_name) {
    WeatherToolResult(
      weathR::point_tomorrow(lat, lon, short = FALSE),
      location_name = location_name
    )
  },
  name = "get_weather_forecast",
  description = "Get the weather forecast for a location.",
  arguments = list(
    lat = type_number("Latitude"),
    lon = type_number("Longitude"),
    location_name = type_string("Name of the location for display to the user")
  )
)

Finally, we can extend contents_shinychat() to render our custom content class for display in the chat interface. The basic process is to define a contents_shinychat() external generic and then implement a method for your custom class.

contents_shinychat <- S7::new_external_generic(
  package = "shinychat",
  name = "contents_shinychat",
  dispatch_args = "contents"
)

S7::method(contents_shinychat, WeatherToolResult) <- function(content) {
  # Your custom rendering logic here
}

You can use this pattern to completely customize how the content is displayed inside shinychat by returning HTML objects directly from this method.

You can also use this pattern to build upon the default shinychat display for tool requests and results. By using S7::super(), you can create the object shinychat uses for tool results (or tool requests), and then modify it to suit your needs.

S7::method(contents_shinychat, WeatherToolResult) <- function(content) {
  # Call the super method for ContentToolResult to get shinychat's defaults
  res <- contents_shinychat(S7::super(content, ContentToolResult))

  # Then update the result object with more specific content
  # In this case, we render the tool result dataframe as a {gt} table...
  res$value <- gt::as_raw_html(gt::gt(content@value))
  res$value_type <- "html"
  # ...and update the tool result title to include the location name
  res$title <- paste("Weather Forecast for", content@location_name)

  res
}

Note that you do not need to create a new class or extend contents_shinychat() to customize the tool display. Rather, you can use the strategies discussed in the Tool Calling UI article to customize the tool request and result display by providing a display list in the extra argument of the tool result.

Description

An ellmer::ContentText subclass that preserves the original slash command entered by the user. When the chat UI is restored from a bookmark (or pre-existing turns), the original ⁠/command user_text⁠ is shown in the UI instead of the (possibly transformed) text that was sent to the LLM.

Usage

ContentSlashCommand(
  text = stop("Required"),
  command = character(0),
  user_text = ""
)

Arguments

Value

A ContentSlashCommand object.

How it works

Slash command handlers that accept an argument receive a ContentSlashCommand object rather than a plain string. The object has three properties:

• command: the command name (e.g., "greet").

• user_text: the text the user typed after the command (e.g., "world").

• text: the text that will be sent to the LLM. This starts as a descriptive string like "The user entered the /greet slash command with arguments: world". Set it to whatever text the LLM should actually see.

Because ContentSlashCommand extends ellmer::ContentText, it works anywhere a ContentText does – including client$stream(), client$chat(), etc. LLM providers read the text property (inherited behavior), while contents_shinychat() reconstructs the original ⁠/command user_text⁠ for display in the chat UI.

Bookmark serialization via ellmer::contents_record() / ellmer::contents_replay() is automatic.

Example

chat$slash_command("greet", "Greet someone", function(content) {
  content@text <- paste("Say hello to", content@user_text)
  stream <- client$stream(content)
  chat_append("chat", stream)
})

The LLM sees "Say hello to world", but on restore the chat UI shows ⁠/greet world⁠.

See Also

chat_server() for registering slash commands via the slash_command() method on the returned object.

Description

Subclass this to plug a custom persistence backend into chat_enable_history() via history_options(store = ). All methods are partitioned by a conversation_partition() (chat id + owner scope); implementations should not need to know about users, sessions, or Shiny beyond that.

A conversation record is a list with fields schema_version, id, title, title_source ("llm", "user", or NULL), response_count, created_at, updated_at (ISO 8601 strings), client_info, nodes (a named list of turn nodes forming the conversation tree), current_leaf (id of the most recent node, or NULL), values (the app state dict captured by on_save), and bookmark_state_id. A conversation meta list is the lightweight summary returned by list(): id, title, created_at, updated_at, and size_bytes (the backend's storage footprint for that conversation, e.g. on-disk bytes).

Methods

Public methods

• ConversationStore$list()

• ConversationStore$get()

• ConversationStore$put()

• ConversationStore$delete()

• ConversationStore$search()

• ConversationStore$total_size()

• ConversationStore$clone()

ConversationStore$list()

Must be implemented by subclasses. All conversations in partition, newest-first by updated_at.

Usage

ConversationStore$list(partition)

Arguments

Returns

A list of conversation meta lists.

ConversationStore$get()

Must be implemented by subclasses. The full conversation record for id in partition.

Usage

ConversationStore$get(partition, id)

Arguments

Returns

The conversation record, or NULL if missing.

ConversationStore$put()

Must be implemented by subclasses. Upsert record into partition. A rename is just mutating record$title and calling put() again.

Usage

ConversationStore$put(partition, record)

Arguments

Returns

NULL, invisibly.

ConversationStore$delete()

Must be implemented by subclasses. Remove the conversation id from partition. Missing ids are a no-op.

Usage

ConversationStore$delete(partition, id)

Arguments

Returns

NULL, invisibly.

ConversationStore$search()

Case-insensitive substring match of query against title, over list(partition). Backends don't need to override this unless they have a more efficient search path.

Usage

ConversationStore$search(partition, query)

Arguments

Returns

A list of conversation meta lists whose title matches query.

ConversationStore$total_size()

Total bytes used by all conversations in partition, derived from list()'s per-record size_bytes. Backends don't need to override this unless they have a cheaper way to compute it.

Usage

ConversationStore$total_size(partition)

Arguments

Returns

The total size in bytes, as a double.

ConversationStore$clone()

The objects of this class are cloneable with this method.

Usage

ConversationStore$clone(deep = FALSE)

Arguments

Description

File-based conversation storage backend

Super class

ConversationStore -> FileConversationStore

Methods

Public methods

• FileConversationStore$new()

• FileConversationStore$list()

• FileConversationStore$get()

• FileConversationStore$put()

• FileConversationStore$delete()

• FileConversationStore$clone()

FileConversationStore$new()

Create a new file-based conversation store.

Usage

FileConversationStore$new(dir = NULL)

Arguments

FileConversationStore$list()

Usage

FileConversationStore$list(partition)

Arguments

FileConversationStore$get()

Usage

FileConversationStore$get(partition, id)

Arguments

FileConversationStore$put()

Usage

FileConversationStore$put(partition, record)

Arguments

FileConversationStore$delete()

Usage

FileConversationStore$delete(partition, id)

Arguments

FileConversationStore$clone()

The objects of this class are cloneable with this method.

Usage

FileConversationStore$clone(deep = FALSE)

Arguments

Description

Configure chat history options

Usage

history_options(
  restore_mode = c("browser", "url", "none", "bookmark"),
  store = "auto",
  scope = NULL,
  title = "auto",
  max_store_mb = 100
)

Arguments

Value

A configuration object for use with chat_enable_history().

Description

Streams markdown content into a output_markdown_stream() UI element. A markdown stream can be useful for displaying generative AI responses (outside of a chat interface), streaming logs, or other use cases where chunks of content are generated over time.

Usage

markdown_stream(
  id,
  content_stream,
  operation = c("replace", "append"),
  session = getDefaultReactiveDomain()
)

Arguments

Examples

library(shiny)
library(coro)
library(bslib)
library(shinychat)

# Define a generator that yields a random response
# (imagine this is a more sophisticated AI generator)
random_response_generator <- async_generator(function() {
  responses <- c(
    "What does that suggest to you?",
    "I see.",
    "I'm not sure I understand you fully.",
    "What do you think?",
    "Can you elaborate on that?",
    "Interesting question! Let's examine thi... **See more**"
  )

  await(async_sleep(1))
  for (chunk in strsplit(sample(responses, 1), "")[[1]]) {
    yield(chunk)
    await(async_sleep(0.02))
  }
})

ui <- page_fillable(
  actionButton("generate", "Generate response"),
  output_markdown_stream("stream")
)

server <- function(input, output, session) {
  observeEvent(input$generate, {
    markdown_stream("stream", random_response_generator())
  })
}

shinyApp(ui, server)

Description

Creates a UI element for a markdown_stream(). A markdown stream can be useful for displaying generative AI responses (outside of a chat interface), streaming logs, or other use cases where chunks of content are generated over time.

Usage

output_markdown_stream(
  id,
  ...,
  content = "",
  content_type = "markdown",
  auto_scroll = TRUE,
  width = "min(680px, 100%)",
  height = "auto"
)

Arguments

Value

A shiny tag object.

See Also

markdown_stream()

Description

Update the user input of a chat control

Usage

update_chat_user_input(
  id,
  ...,
  value = NULL,
  placeholder = NULL,
  submit = FALSE,
  focus = FALSE,
  attachments = NULL,
  attachment_mode = c("append", "set"),
  session = getDefaultReactiveDomain()
)

Arguments

Examples

library(shiny)
library(bslib)
library(shinychat)

ui <- page_fillable(
  chat_ui("chat"),
  layout_columns(
    fill = FALSE,
    actionButton("update_placeholder", "Update placeholder"),
    actionButton("update_value", "Update user input")
  )
)

server <- function(input, output, session) {
  observeEvent(input$update_placeholder, {
    update_chat_user_input("chat", placeholder = "New placeholder text")
  })

  observeEvent(input$update_value, {
    update_chat_user_input("chat", value = "New user input", focus = TRUE)
  })

  observeEvent(input$chat_user_input, {
    response <- paste0("You said: ", input$chat_user_input)
    chat_append("chat", response)
  })
}

shinyApp(ui, server)