YouTube Transcript:
Chris Laffra - PyScript - Python in the Browser | PyData London 25

Skip watching entire videos - get the full transcript, search for keywords, and copy with one click.

Video Transcript

Welcome everybody to the afternoon

session by data London day two after

tutorials yesterday or the first day of

the actual conference. I'd like to talk

to you today about running Python in the

browser. I like running Python in weird

places and I think the browser is

probably the weirdest where you can run Python.

Python.

Uh this is not a project that I came up

with but I picked it up when it was

announced and I built some applications

on top of it and I think it's a really

fun environment to play with Python and

I try to persuade you of that. So that's

my goal for

today. So who am

I? I grew up in the Netherlands. I got

my master's degree and PhD there. Then I

moved to the US. I worked at technology

banks, technology and banks. So IBM

research was the beginning of my career.

Then Morgan Stanley, then back to the

Netherlands for a bit. Then I joined

Bank of America where we built a large

trading system or platform in Python

that ran all of the trading in Bank of

America or Meil Lynch.

So this is an investment bank where you

want to have a single representation of

what is called an instrument. Uh so that

is an asset or like a stock or a bond or

an interest rate. And these are things

that you trade. So you sell and buy from other

other

people. So at some point you want to

know how much is it worth. So you need a

function on that is called price or and

you have some attributes that you pass

in. So if you have different

implementations in different parts of

the bank then predictability becomes a

problem and your numbers cannot be

compared or compatible and you cannot

calculate your risk reliably. So there's

a lot of efforts in large banks to come

up with single instrument

representation and the project I joined

in bank of America is called quartz

where it tried to do that in

Python. Later on, if you step forward a

little bit, I'm at JP Morgan again,

which is Athena. So, this is sort of um

u a reverse time series. Okay, I I

joined iteration three of this project

and then later on I joined iteration

two. So, I saw like it was like an

alternate reality. Um but there's only

really one way in which you can solve

this problem and all these banks are

trying it on their own way.

I also worked at Google and Uber. So I

have a mix in both technology and

finance but I'm not a command or

something like that. Today I work at AIA

which is based in Abu Dhabi in the

Emirates and that's a sovereign wealth

fund where we manage money by trying to

do systematic trading and make more.

So you can understand that that requires

a lot of data science because you have

to like gather a lot of data from

outside make decisions based on it and

then come up with trading strategies

that you

apply. In all of these worlds there are

two different kinds of people. Let's

just like just make it black and white.

So there's two kinds of people. There's

like uh econ economists, they're like

data scientists, they're like uh

quantitative, they're mathemat

mathematicians, and they like to use um

Jupyter notebooks. On the other side,

you have people in production that like

to have predictability of the results.

They want to have standardized Python.

They want to have reuse. They want to

have not copies of things. And they want

to run at yeah production scale because

you have to manage a lot of

data and these are like aliens. They

cannot talk to each other. So that is my

goal in life is to like sort of get them closer

closer

together. In parallel to yeah doing work

I also like communicating. So I've

written quite a few books. Uh the most

recent ones are not about technology but

more about how do you communicate better

as an engineer. So there are a lot of

channels that we use to communicate with

as engineers. I'm assuming that most of

you are engineers here. Um and we

engineer we we communicate with people

that don't understand us. like we want

to talk about register allocation and

not about what is the value of the

software to the bottom line whatever

right so there's a lot of opportunity

for engineers to become happier more

successful richer

um if you can actually talk to these

people so that's at some point I started

like collecting all these tips and ended

up in a book the other book is written

by an AI so you should maybe take a look

at it. I'm not trying to make any money

on this on the second book. So, it's

sold for yeah

um printing cost because this was an

experiment. When Chad GPT just came out,

I wrote a book together with it. So, I

asked all the questions and all the

answers were given by Chad

GPT. We'll use AI in this presentation

as well. Uh not just for creating the

slides, but we'll use it to generate

code and I'll explain how that works.

because I think that's the future where

we collaborate with

AIS to add value quicker maybe bring

these two worlds

together. So why pycript? So I'm showing

you some applications of pycript. So

what is pycript actually? So pycript is

coming out of a need is that we want to

have an easier way to develop web applications.

Python is very popular. Um, who here

uses Python? Yeah, everybody, of course.

Yes, because this is a Python

conference. But, uh, if you look at

things like LinkedIn, uh, and you you

scan all the jobs and you see what

languages do people require, Python is

on the top. And there's many ways in

which you can do these kinds of

calculations and Python often comes to

the top. So, it's a very popular language.

language.

Um, also it comes with a lot of

batteries. So there's a lot of

development over the last 30 years or so

that Python has been around specifically

targeted towards data analytics. So data

manipulation, data visualization,

natural learning processing where

everything has been solved. So it's a

lot easier for you to build value and

add things on top because there's so

But there are also too many batteries.

Uh it can be overwhelming to pick the

right one. There's more than half a

million projects on Pi right now. And if

you see the download stats, it's like I

don't know 30 billion a month.

So this means that people download

Python packages a lot. And why is that?

because they have to create an

environment where

um all the versions of all the packages

that they require recursively have a

predictable outcome so that the next

time you run it it's reproducible

result. So this is an AI generated

picture. I asked it like why is Python

installation a pain and it came up with

this picture and it's apparently such a

pain that this is probably XKCD

reference where Python can fly somewhere

XKCD and it's such a pain that it

actually writes your code on the wrong

like but it's the reality like

installation of Python packages is hard

and when there's a problem you solve it

you make another tool right so UV solves

everything now uh not really it still

requires if you want to have your

application delivered to people who are

not Python developers and you say well

it's easy you just do UV in run and okay

they give up

So this is too hard to develop your

applications and deliver it in one

button press and this is what pycript

tries to achieve. So we want to have

Python available everywhere without any

installation. So how do you do

that? You make a web application. So the

web application is a link. People click

on the link and it works. So how does

that work? Uh there's a little bit of

JavaScript that bootstraps

uh the download of the web assembly

containers for the two Python VMs that

are currently supported in PIScript

which are MicroPython and

Piodite. And these are then bootstrapped

with an empty container and they say

like well I need my uh all the basic

libraries from Python like definition of

string and all of that and then it

builds that up and you say well I also

want Panda so I want this and also

downloads that from

Pippi and then installs it on demand in that

that

container and at that moment your Python

code is then executed and runs in that

container. it can talk to the JavaScript

engine to access the DOM and in that way

you can actually render things in the

browser. So this is the basic process

without any manual installation. People

just click on the link behind the

scenes. The PIScript runtime the

JavaScript bootstrapper will kick off

the right VM the one you chose and then

it will bootstrap your application. So

why do we have two VMs?

Um there's actually you can do any uh

but there's two popular ones and

microython and podite. So microython is

for embedded devices. It's really tiny

python VM and it measures lab libraries

included like

240k. Piodite is much bigger and it has

everything in the language. So to get it

working on an embedded device you say

like well um for string there are like

70 functions right 70 methods you can do

on a string we don't need all of them on

embedded only maybe 30 right so you

selectively cut down the size of the

library and therefore also the

capabilities of the VM to make it as

small as possible in pi says like no no

right we're supporting all of Python so

this is a standard C Python VM [Music]

[Music]

and that has been compiled to web

assembly. So normally Python runs on

Linux, on Mac OS, on Windows and

somebody wrote the make file to say like

how do you compile into a native

application, right? It's the CPython VM

is a C program and you can compile it to

native instructions and bundle it up and

then make an XC out of it or an

executable format for Linux. And this is

what somebody has done. So they have

written the make file basically for web

assembly for the piodite VM. So your

Python code is not converted into web

assembly that just runs as before and

it's a standard Python VM. So

compatibility is

oneonone but it comes at a price. It's

much bigger and I said in the beginning

uh you start with nothing and it has to

like import everything and build in

there because for security reasons you

start with an empty sandbox.

So that's the trade-off startup time and

later. So to get started with PIScript u

we always recommend people to go to

piscript.net which is a starting point.

It gets you to discard. It gets you to

And I'll show you two examples of uh

standard examples that come with um

PIScript that show you the power of

running Python in the

browser. The first application you've

already been watching this slideshow is

done as a PIScript application.

So, I wrote some Python code that

downloads a

Google presentation, extracts out the

pictures, puts them in the file system,

and then I have Python logic that says

like, oh, you press the right key, I go

to the next one. Oh, you press the left

key, I go to the and it even works with

this remote. This is really interesting,

too. So, I can go and back and forth.

So, it's almost like Google Slides, but

then all written in Python,

right? Why would you do that? Well, it's

just fun. [Music]

[Music]

Uh, so the, uh, first example, you go to

pcript.net and it will link to the

documentation as you see in the bottom

here, but it also has a link to examples

and this is what I did. So, I opened the

first one and this is a yeah, basically

a a poor person's IDE. You can edit

Python code in here.

Um, so you can see it here. It's in an

editor and when you make changes and you

save and you run, it runs here in the

browser. Now this is using WebGL and

it's really like I mentioned earlier

from the Python code you can get to the

JavaScript runtime and then to the DOM.

So here we're using a JavaScript library

to do that for us. So from Python we

basically just instantiate that WebGL

library and you can see how it works

here in the top. we say import tree and

then from tree we say get a webgl

renderer and we basically you know write

the same kind of code that we would do

in JavaScript but now we write it in

Python but it's really executing

JavaScript so there's a proxy between

the two languages between the two

runtimes and anytime you make a call

from Python to JavaScript we create a

proxy and then we say to the other side

like I want to have the WebGL renderer

on three and it will create a proxy on

the other side for us to hold on to that

result of that execution and send back

the proxy so that now we can say I want

to call new on it and then so we go back

and forth between these two runtimes all

the time so anytime you see a dot here

that's basically going between these two

runtimes and once it runs uh it runs at

normal yeah speed, maximum JavaScript

speed basically, which is pretty fast

these days actually because V8 is

unbelievable. So this is the basics.

This is one thing you can do. You can

very easily use JavaScript libraries and

script them from Python in this way. So

this works because you have an

index.html HTML that says at some point

says I have a script tag

here. Um, can I zoom in a bit? Yeah. So,

the script tag looks like this. So, it

says like what I want to run is piodite.

Here's my source file and here's my configuration.

configuration.

So this is like a JavaScript script tag

but in this case we say the type is

Python and then it just bootstraps on

top. So this is one example. The other

example is using a Python library called

Folium and this is often used in Jupyter

notebooks for rendering the result of

geographical information system related data.

data.

So in this case we have the map of the

US and we overlay it with some data. Uh

so this is this whole script here but

this one doesn't go to JavaScript but it

keeps talking to Python libraries. So I

installed another package in my

piscripttoml. You can see that one of

the packages I

install is called Folium and the other

one is pandas because those are the two

packages that my code needs and my code

imports pandas and it imports volume and

it basically downloads this data from

this location turns it into a pandas

data frame and then it gives that to

folium to render a map and it basically

extracts the data out. So now when you

look at this normally in Jupyter

notebook what it generates is HTML. When

you say

display this generates HTML gives it

back to the Jupyter notebook and it

renders it in the document in place but

what is rendered there is HTML that is

generated by the Python code that was

executed and this is the same way. So

from inside your web application, you

can run Python code that generates HTML

and then you basically host that back

into the DOM. So when I interact with

it, this is again JavaScript. So this is

not using

Python event handling and Python isn't

drawing these these rectangles whatever

at at this moment it's all done by

JavaScript the same way that you would

have in a Jupyter notebook.

So you can combine these things. You can

write pure Python. You can

write Python that uses packages that

generates JavaScript code. Or you can

directly go to JavaScript and you can

mix and match at any point which you like.

Yes, a question. Jun has a question.

Wait a second for the microphone, please.

Yes. Could you please go back to the

previous Yeah,

Yes. JavaScript. Oh yes. Yes. At the

top. Um that's a good question. At the

top there's some initialization

happening. So you need to

install uh PIScript and that is done

with this URL and this registers itself

as a content handler for the page. So it

gets notified when a script tag is in

included in the page. So that JavaScript

code then takes out that main.py and it

starts up the VM and bootstraps the VM

with that. Yeah. So there's a little bit

of setup at the top here.

Yep. If you're interested in doing this

yourself, you just go to

piscript.com. So this is running on

here and this is from the from the

examples folder but you can also create

your own examples very easily. So you

just press the new button and it will

create a new example and so you don't

have to know how to get strap

Another example is um so I showed you a

screenshot of this but this is actually

my homepage and this is also written as

a pi script application. So there's some

animation that is happening here. Um

when you move the mouse it's uh allowing

you to yeah click on

things. So why would you write this in

Python? Because I could. I thought it

was really cool. And um you could have

done this in JavaScript of course but

hey come on. Uh if you look at inspect

and then at the source uh maybe I should

show the page for source. Uh you can see

that I use some JavaScript and then I

use main.py. So this is my code for my personal

personal

awesome. Another example is based on

work I did at JP Morgan when I was

employed there. We had uh I mentioned

these large trading platforms and these

large applications and we start millions

of them every

day and they take a long time to start

up and I wanted to save money with the

bank money and I thought like how can I

make this faster so I wrote a profiler

uh it's a sampling profiler that runs in

production on these on these apps and it

records whatever is happening and then

there was a UI written in Typescript

that renders

it and then pycript came out. I left JP

Morgan and I thought like hm I want to

do that thing again but not use

TypeScript but I want to do it in

Python. So this was my first experiment

that I wrote with

PIScript. So you run the sampler and

that's already installed actually on my

platform on my laptop here. So anytime

Python runs on my laptop, it rec it

creates a recording of that execution

and they automatically show up in this

table on the left and then I can

basically browse them back. So this

whole thing is written in Python. So

there's no JavaScript here. This is all

real Python. So this is an example of

the other extreme. I'm not leveraging

any JavaScript libraries because I

wanted to see how far can I push it and

this was horribly slow. Because I

mentioned before like whenever you have

a dot you go from the Python VM to the

JavaScript VM you have to create a proxy

to so that they can talk with each other

in the future but I also have to

marshall all the arguments like if you

send a string from here to there uh we

have to encode it in a format that both

understand and JSON is the best one. So

there's a lot of conversions happening.

So if you imagine you draw this whole

diagram, there's a lot of draw line

instructions, a fill box, draw text, and

each single one is not that expensive to

go across, maybe 50

nconds. But if you have a thousand of

them or

10,000, it adds up and it's more than a

second. So this thing was horribly slow,

like boom boom boom. So I had to make

this faster. So the current speed is

it's actually I cannot like disconnect

it from the mouse. This is how fast it

renders and every time I move the mouse

it rerenders the whole page. So this is

u and you zoom in it does the same

thing. So this is more than acceptable

for the performance that I was aiming

for. So the way this works it collects

all the draws. It collects all the draw

strings and all the draw lines and it

basically wraps those as one

instruction. And I say, I want you to

draw purple boxes in these locations.

integers. So now I need only maybe like

five or six round trips between these

two VMs to render it fast. So I'm

cheating. I have a little bit of

JavaScript in the back end. So it's like

my native code, right? In in languages

like Java, you write C code. In Python,

you write C code to make that little bit

faster. In Pycript or Podite in the

browser, you would use JavaScript as

your native code to make it

faster. Now, what can you do in this

thing? Like, yeah, it records

everything. It also records the logs.

So, this is interesting. You can combine

the two. So, you can look at the

timeline. What happened here? So, this

is like a log entry. You can screw. So

this is all Java. This is not

JavaScript. This is Python. So this UI

is drawn by Python and it's at this

moment this is all in the Python memory

whether I'm like looking over this this historical

historical

data and you can do things like what

this what does this program do if I ask

you right you don't know. Uh it's called

images. So maybe it does something with

images. You don't know. Um but it it

runs for a while. You can see here it

does pipot for a while and then it runs

process image for 5 seconds. So this is

coming back to that example that I

mentioned earlier at JP Morgan where we

have startup took so long and I didn't

know why. In Python uh we pay the price

for imports and they're often hidden. So

if you look at this example dataf

frames, this is an example where I

wanted to do something with pandas dataf

frames and it took like 15 seconds to

just import

pandas. At the top you can see CPU was

zero. So this is just waiting for IO. Um

the yellow line shows you the number of

packages loaded so far. So it goes

eventually to something like

626. So in order to do one thing with

pandas, you need to load 600 modules and

each one is a Python file and each one

has to come from the file system. And

this is all IO. I'm just waiting for the

bytes to come in. Now why it was so slow

this time, I have no idea. It's not

normal to take 15 seconds, but at least

I know now that it comes from IO and I

need to have better caching on my local

file system or

something. But if you want to know what

this thing does like the images example,

um I can look at the design and this is

quite interesting. So what I do here is

this is drawn with D3JS. So I'm

cheating. I'm using JavaScript but this

is this is the possibility right I can

switch between Python code and

JavaScript whatever is the most

practical at that moment like Python

doesn't try to solve solved problems uh

we leverage what is available anywhere

but from this I can now very easy yeah I

I can sort of get an an idea what this

program is

doing and if I don't then I can always

ask the AI so open AI what is this thing

doing and it looked at that trace of the

execution and it just gave me a

does. So this was my first experiment.

Um I did a tutorial in uh December on

PIScript online and there I wanted to

also use AI. So you always you have to

use AI right? So this UI is is written

in Python and it goes out to OpenAI to

basically yeah help me with setting my goals.

goals.

So it says here PI data London but let's

So, it's trying to find a a picture for

trying to get promoted. Takes a little

longer. It's like 15 seconds on average.

The text is easy to generate

apparently. So, it gives me some idea of

what I can do to get promoted. I have to

understand the expectations,

requirements. Oh, yeah. That makes

sense. Uh have to research. Yeah. Yeah.

Yeah. This all makes sense. And then [Music]

[Music]

uh yeah the picture uh still has to be

worked on but your goal has to be

achievable and

renovate and miserable but anyway you

get the feeling. Uh so how do you do

this? This is like 30 lines of Python

this whole app. So it's very easy to construct

construct

um because you don't have to learn any

JavaScript. This is if you know Python,

this will feel like very familiar

territory. So to build that last app, I

wrote a UI toolkit because it's really

hard to write UIs in a very low level on

the DOM level like create DOM entries

and whatever. I wanted to have a higher

level abstractions. I wanted to have

reactivity. Uh I want to have a model

and then have widgets update

automatically when the model changes.

And because we're writing Python code,

this is actually not that not that hard.

Um, so on the right here, you see I

introduce a class. It's a Python class.

And then I basically write a UI. So it's

it says like I need a vertical box.

Inside I have a label and then I have an

input which refers to the name of the

product. So this is the first line and

you can sort of see the structure of the

code already. It's a vertical box with

little things in there. And now when I

make changes in this model, so if the

name changes, all the UI elements are

dependent like ReactJS um and then this

string updates. You can also do this

from code. So you can like make some

changes in the model and then directly

and then the UI gets updated. Now this

is easy to write was very easy to do.

This library is called LTK. It's the

little toolkit. So if you go here, it

shows you

the repository. So you can you can

actually look at the code and it's not

that hard. It's

u this is the biggest file. This has a

definition of all the widgets in LTK and it's

it's

like let me see 1700 lines and this is

the most complex one. The rest are much

lines. So that allows me now to build up

more interesting

abstractions in Python to build yeah bigger

bigger

applications. I also mentioned that we

have MicroPython and we have Piodite and

this is an example where you can do

that. So here I'm running with Piodite.

Now I'm switching to Piodite and you see

it takes a little longer. So it has to

construct the Python VM, bring in more stuff.

stuff.

trade-off. I've also built uh Python

visualization in the past. This was like

2015. I did it on

pspot and this is a version that runs

entirely in the browser. So, you can

also do set cis set trace on your

application. So, anything you can do in

Python, you can actually do within the

browser. So, then you can build very

interesting visualizations like this one.

And this one is calculating the golden

ratio. And it sets a break point on this

line. And each time the break point is

hit, we run the visualization script at the

the

bottom. And this is yeah updating the

view on the top. So I can go back in time

time

and basically see the picture. So if you

ever want to know what the difference is

between breath first search and depth

first search. So this is breath first

search. So you can see it right? It's

it. So think about how this here like

sounds. And now I'm going to switch to

So this helps me understand how the program

program

works. And again, this is a little bit

more useful way of running Python in the

browser. Uh because I don't need a

server to run the algorithms and

visualize them. Um it would have been

quite hard to do all of that in

JavaScript. So this is a very good

example of why you want to do it there.

Okay, I want to get promoted.

Uh this is another example of an

application that I thought like I was

going to get really rich with. Nobody

bought it. Uh it's called

lifesaver.coach. So if you're

interested, go there, buy a

subscription. Um to get promoted, it

gives you advice here and so on. Um I

can click this thing like and this will

make a a task out of it. So now I know I

have to work on this and then I'm done

with it. Yay. Um, next slide is I

already got the notification of the

time. So, the next one is um, Jupyter

notebooks allow you to write Python code

and run it sort of in the browser but it

runs in a kernel. So, this is uh, if

you've heard of Python for Excel, Python

in Excel, this is like Python in Excel

but this is Excel in Python. So I've

written a sheet a spreadsheet here in

Python using LTK and it's basically

building the dependency graph of

execution but all the Python code runs

in the browser in web assembly

here. Um why would you do that? Because

I wanted to prove that I could. And it's

using two VMs. It has MicroPython for

the UI and it has Pyodite for the

worker. And it can do anything you can

do with pandas, polars, whatever. So

here I'm loading uh URL. So if I go here

and basically grab that one

one

quickly. So if I have this

URL and I go to a new file for instance,

let's start with an empty

sheet. I can say import and then pass in

that URL.

and it will do a preview of that and

then I can import it into the sheet. I

can do a pandas or polars. So we got the

po polar guide book authors here sitting

in the first row. So this is for you

guys. This is polars. Here we go.

Um you can also use um duct db data

frames and here I experimented with some

interesting filtering. So here we have

like the coordinates are filtered by

14,000. So let's change that. So I say

instead of 14 I want to have

15,000. So when I say enter it actually

updated the source code

here and then redrrew the map. So now I

have fewer

ones. So

um that is all cool but it it gets even

better. So I I gave a demo of this to

Guido and then he said like uh yeah it

would be really cool if you can actually

interact with the sheet from the cell

right and he said like I I tried to like

negotiate that in in Microsoft but they

wouldn't let me do this in Excel. Um so

I said oh yeah that's possible let's do

that. So what we have here is a cell

function and here's the code and what it

can do it can escape its sandbox and it

can talk to the sheet and it can update

DOM entries in the sheet. So this is

what I'm doing here. Can also set event

handlers. So I can say move the piece

from here to there and then move from

here to there and then move my queen

here. And this thing is really dumb

because it moves randomly. So, it's easy

to win, but

uh try and do this in Excel, right? Try

to do this in I try to actually um hook

up a Python engine as well, a game

engine, but those are mostly written in

C. And they're not pure Python, so

they're hard to import. So here you can

see I import chess as a module and there

are chess engines that are compatible

with chess the the Python

module but they wouldn't run in web

assembly. So I'd have to compile the C

code to web assembly first and that that

was a little bit too much work. Now so

that is some of the things that are

possible but people were complaining to

me like oh that looked like Excel. This

is can you do all the key bindings for

Excel? No no please not. uh do you have

all the imported uh default functions in

X? No, let's let's remove the sheet.

Right? So this is an experiment that I

did is similar. So this is again micro

Python for the UI. It's using piodite to

run the code and it's using another

piodite for finding out all these

components. So this pallet is done by

one VM and then it runs and why you have

to talk to me afterwards. We can talk

about that. So what I want to do is I

want to draw a picture. So I want to

draw a candlestick of some historical

data. So what I need is a data frame. So

I click on data frame. I want to connect

it to something. And there are two

potential candidates. So history is one.

So I create history. So now I can

connect up my history to my data

frame. And this history object needs a

symbol and from date and to date. So I

luckily have components hanging around

that match those things. So I can

connect them. So Tesla symbol from date

and then to date. And when I click it

now has all the three inputs and it can

actually run. So you will see it

run that cell. So it ran the pandas data

frame creation code for that and then

then it gave the output to the input of

the candlestick and it could draw could

draw itself again. This runs up to

March. So let's fix March and make it a

little bit later. Select today is 06

07 and when I change the input changed

and then the data frame will be run

again and then it's updated again. So

there was some recovery and now we're

Um so this is what is possible uh with

Python in the browser. These are some

examples that I gave you. If you're

interested in trying them out yourself,

uh, this is all open source. Py Sheets

is open source. The flow example is open

source. So, just head down to the GitHub

logo and you'll see the repositories there.

there.

And my plan was to or my goal for today

was to get you excited about running

Python in the browser um in a in a like

as a replacement for JavaScript because

I think Python is a much nicer language.

We all agree,

right? So with that, I want to thank you

and if you have any more questions, we I

guess we almost ran out of time or we

ran out of time. So come after me, come

to me in the in the hallway and we have

Click on any text or timestamp to jump to that moment in the video

Most transcripts ready in under 5 seconds

One-Click Copy125+ LanguagesSearch ContentJump to Timestamps

Paste YouTube URL

Enter any YouTube video link to get the full transcript

Most transcripts ready in under 5 seconds

Get Our Chrome Extension

Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.

Add to Chrome — Free

Works with YouTube, Coursera, Udemy and more educational platforms

Get Instant Transcripts: Just Edit the Domain in Your Address Bar!

YouTube

←

→

↻

https://www.youtube.com/watch?v=UF8uR6Z6KLc

YoutubeToText

←

→

↻

https://youtubetotext.net/watch?v=UF8uR6Z6KLc

YouTube Transcript:Chris Laffra - PyScript - Python in the Browser | PyData London 25

Video Transcript

Paste YouTube URL

Transcript Extraction Form

Get Our Chrome Extension

Get Instant Transcripts: Just Edit the Domain in Your Address Bar!

YouTube Transcript:
Chris Laffra - PyScript - Python in the Browser | PyData London 25