ChatStep#

Open this notebook in Jupyterlite | Download this notebook from GitHub (right-click to download).

import time
import panel as pn

pn.extension()

The ChatStep opens up the possibility to display / hide intermediate steps prior to the final result.

Check out the panel-chat-examples docs to see applicable examples related to LangChain, OpenAI, Mistral, Llama, etc. If you have an example to demo, we’d love to add it to the panel-chat-examples gallery!

Parameters:#

Core#

collapsed_on_success (bool): Whether to collapse the card on completion. Defaults to True.
context_exception (str): How to handle exceptions raised upon exiting the context manager. If “raise”, the exception will be raised. If “summary”, a summary will be sent to the chat step. If “verbose”, the full traceback will be sent to the chat step. If “ignore”, the exception will be ignored.
success_title (str): Title to display when status is success; if not provided and collapsed_on_success uses the last object’s string. Defaults to None.
default_title (str): The default title to display if the other title params are unset. Defaults to an empty string "".
failed_title (str): Title to display when status is failed. Defaults to None.
margin (tuple): Allows creating additional space around the component. May be specified as a two-tuple of the form (vertical, horizontal) or a four-tuple (top, right, bottom, left). Defaults to (5, 5, 5, 10).
objects (list): The list of objects to display in the ChatStep, which will be formatted like a Column. Should not generally be modified directly except when replaced in its entirety.
pending_title (str): Title to display when status is pending. Defaults to None.
running_title (str): Title to display when status is running. Defaults to None.
status (str): The status of the chat step. Must be one of [“pending”, “running”, “success”, “failed”]. Defaults to "pending".

Styling#

collapsed (bool): Whether the contents of the ChatStep are collapsed. Defaults to False.
default_badges (dict): Mapping from status to default status badge. Defaults to DEFAULT_STATUS_BADGES; keys must be one of ‘pending’, ‘running’, ‘success’, ‘failed’.

Methods#

Core#

stream: Stream a token to the last available string-like object.
stream_title: Stream a token to the title header.
serialize: Format the object to a string.

Basics#

ChatStep can be initialized without any arguments.

chat_step = pn.chat.ChatStep()
chat_step

Append Markdown objects to the chat step by calling stream.

chat_step.stream("Just thinking...")

Calling it again will concatenate the text to the last available Markdown pane.

chat_step.stream(" about `ChatStep`!")

Alternatively, setting replace=True will override the existing message.

chat_step.stream("`ChatStep` can do a lot of things!", replace=True)

It’s possible to also append any objects, like images.

chat_step.append(pn.pane.Image("https://assets.holoviz.org/panel/samples/png_sample.png", width=50, height=50))

Calling stream afterwards will append a new Markdown pane below it.

chat_step.stream("Like it can support images too! Above is a picture of dices.")
print(chat_step.objects)

[Markdown(str, css_classes=['step-message']), Image(str, height=50, sizing_mode='fixed', width=50), Markdown(str, css_classes=['step-message'])]

To convert the objects into a string, call serialize, or simply use str().

chat_step.serialize()

ChatStep(
    Markdown='`ChatStep` can do a lot of things!',
    Image='https://assets.holoviz.org/panel/samples/png_sample.png',
    Markdown='Like it can support images too! Above is a picture of dices.'
)

Badges#

The default avatars are BooleanStatus, but can be changed by providing default_badges. The values can be emojis, images, text, or Panel objects.

chat_step = pn.chat.ChatStep(
    default_badges={
        "pending": "🤔",
        "running": "🏃",
        "success": "🎉",
        "failed": "😞",
    },
    status="success",
)
chat_step

Status#

To show that the step is processing, you can set the status to running and provide a running_title.

chat_step = pn.chat.ChatStep(status="running", running_title="Processing this step...")
chat_step.stream("Pretending to do something.")
chat_step

Upon completion, set the status to success and it’ll automatically collapse the contents.

chat_step.status = "success"

To have the title update on success, either provide the success_title or default_title.

chat_step = pn.chat.ChatStep(status="running", running_title="Processing this step...", success_title="Pretend job done!")
chat_step.stream("Pretending to do something.")
chat_step.status = "success"
chat_step

To streamline this process, ChatStep may be used as a context manager.

Upon entry, the status will be changed from the default pending to running.

Exiting will change the status from running to completed if successful.

chat_step = pn.chat.ChatStep(running_title="Processing this step...", success_title="Pretend job done!")
with chat_step:
    chat_step.stream("Pretending to do something.")

chat_step

However, if an exception occurs, the status will be set to “failed” and the error message will be displayed on the title.

chat_step = pn.chat.ChatStep(running_title="Processing this step...", success_title="Pretend job done!")
chat_step

try:
    with chat_step:
        chat_step.stream("Breaking something")
        raise RuntimeError("Just demoing!")
except RuntimeError as e:
    print("Comment this try/except to see what happens!")

Comment this try/except to see what happens!

The title can also be streamed.

chat_step = pn.chat.ChatStep()
chat_step

chat_step.status = "running"
for char in "It's streaming a title!":
    time.sleep(0.02)
    chat_step.stream_title(char)

It can be used with any status, like setting status="pending".

chat_step = pn.chat.ChatStep()
chat_step

for char in "I'm deciding on a title... how about this one?":
    time.sleep(0.01)
    chat_step.stream_title(char, status="pending")

The title can also be completely replaced–equivalent to setting chat_step.pending_title = "Nevermind!

chat_step.stream_title("Nevermind!", replace=True, status="pending")

Be sure to check out ChatFeed to see it works with a ChatFeed or ChatInterface!

Open this notebook in Jupyterlite | Download this notebook from GitHub (right-click to download).