Balatrobot - Botting API Mod

[bold summit]

I am proud to introduce Balatrobot, a botting API for Balatro:
https://github.com/besteon/balatrobot

This mod requires Steamodded (tested on v0.9.3).

The bots are capable of running through an entire game of Balatro unassisted! Play hands, buy cards, and open boosters all using the API. The API will automatically query your bot for a choice of action when the time comes, no heavy lifting required.

To create a bot, just modify Bot.lua. There is an example on the github page to get you started.

I would love feedback - I am sure there are still bugs in the later stages of the run (I didn't have time to make a bot get that far yet ).

Currently requires some source code knowledge to get the relevant data for your botting decisions. A standardized gamestate data format coming soon. And if there is enough interest, perhaps a websocket/web API so that bots can be written in languages other than Lua.

Your bot can be as complex as you want it to be, but the API is designed to be easy to use! Here is an example of how a bot can choose between skipping or selecting a blind:

[pliant patrol]

I wonder how consistent the timing of this is with Ankh

[gray ferry]

without testing and thus without guarantee, I'd say this should work fine on latest steamodded

[pliant patrol]

would be huge for TASing

[gray ferry]

either way you should update... most mods will stop being compatible with 0.7.2 as they update

[pliant patrol]

or also how accurate the inputs are compared to frame-perfect inputs in RTA

[bold summit]

The API only uses button functions once they are actually active. It may not be 100% TAS ready yet from a "could a human do this?" perspective, but it is quite close already.

[placid vault]

So is this TAS or the game plays by itself

[bold summit]

This is high on my priority, will hopefully get to it later today.

[bold summit]

It is just an API. You must write the bot by yourself. The API tries to make it easy for you.

[gray ferry]

I don't think this is really using any of the Steamodded APIs? there's not much a Steamodded update can break in that case

[bold summit]

This is 100% from scratch, just uses Steamodded for loading the mod into the game.

Although I think Steamodded has changed how other .lua files are loaded in so that part might be broken in the latest release.

[gray ferry]

yeah should be fine then unless you're destructively overriding functions the loader uses

[gray ferry]

you might have to change love.filesystem to NFS

[bold summit]

Right now the API goes based off of if button.config.button is set, but timing can be adjusted for each action, although I haven't made that part easy to do yet. I am personally more interested in making a smart bot, not necessarily a TAS, so I would love collaboration on making it TAS ready.

[covert chasm]

nice work!

[bold summit]

Just tested it, didn't need to change anything at all. It loaded right away.

[gray ferry]

neat

[soft ermine]

now you dont even need to play the game, the game plays for you

[gray ferry]

The game plays ~~for ~~you

important distinction

[bold summit]

I also wrote a hooking library (hook.lua) that might be useful to other modders. Implements breakpoints and callbacks for functions and onread/write for tables. Breakpoints being functions that run right before the hooked function.

[soft ermine]

does that make me a card?

[gray ferry]

yes

[pseudo light]

It's just to make it compatible with lovely mod directory change. If it's activated it migth cause issue.

[pliant patrol]

that should be pretty close to what is needed for an input to happen
Even in RTA you can get some wild-looking early inputs

https://youtu.be/ftrjhg8AHpM?si=3wRpCwwnrCpx1vS4
Don't have much time now but would be cool to see this recreated with TAS

my sum of best segments was ~1:25.5 on this seed, wouldn't be surprised to see a low 1:24 TAS

[bold summit]

That would be a good exercise for getting far in a run to see interactions that may be buggy. I might give it a shot.

[keen haven]

cant you combine this with immolate to make a bot that can beat any seed automatically?

i might try to make a balatro bot that can complete any seed automatically

[pliant patrol]

You probably don’t even need Immolate

[keen haven]

I mean youd have to see into the future

[bold summit]

you could simulate the rng calls but I don't think that is as fun as making a smart bot that adapts on the fly. But go for it!

[keen haven]

ill try making a fun bot, but some seeds you cant win from fun

[pliant patrol]

btw you can't set the deck
that's an issue

[pliant patrol]

for k, v in pairs(G.P_CENTER_POOLS.Back) do
    if v.name == Bot.SETTINGS.deck then
        G.GAME.selected_back:change_to(v)
        G.GAME.viewed_back:change_to(v)
    end
end

This is how I fixed it

[bold summit]

I will add that

[pliant patrol]

trying just the first ante blind skips with action_delay=0 it's already way faster than a human can do it

there's some delay between skipping multiple blinds that must not be detected

[bold summit]

I currently don't check controller locks. But if you find the condition it should be easy to change it with a "firewhenready"

[forest dagger]

where did you end up putting this? i'm not nearly good enough with how lua is structured to intuit that LOL

i can tell that it would just be called with like deck == "Red Deck" or whatever, i just have no idea where i would actually put this to apply that

[pliant patrol]

I put this before whenever it calls start_run in middleware.lua

[bold summit]

I have pushed this change. I'm working on recreating your WR, going to implement a better way for the bot to choose when to rearrange/sell cards. I haven't implemented selling cards yet. I also want to adjust how cards are selected and various other timing bits. Right now, the cards are selected sequentially so that it looks nice to watch, but I will add some bot settings to adjust. Perhaps make a TAS bot mode

I also noticed that when opening a booster, the Skip button becomes available before all the cards are drawn. The bot will wait for G.STATE_COMPLETE before performing the action, I will work on that as well

[pliant patrol]

working on a wrapper that makes this more TAS-like

[pliant patrol]

well this is what happens when I try to buy a card

actually this appears to be my bad

Looking at it more, it’s only crashing because I’m having it buy the 2nd item

nvm I'm dumb

[pliant patrol]

Finished the TAS btw

The final time is 1:23.056

Biggest things I’m noticing from a TAS standpoint is skipping the spectral pack too late and skipping blinds too soon

For small optimizations you can also pre-select cards while they are being drawn

I used a backup strategy since Joker selling wasn’t implemented

(the inputs are in a separate file so that the wrapper can be used for more runs)

this is the splits against my PB

Why the discrepancies? In Ante 2, the bot’s advantage is lowered due to the money gaining animations, and a slightly faster strategy to buy the joker is possible (doing it during the tag animation)

In Ante 7, the bot waits for a foil tag animation that you can skip, and has to deal with the same time loss over its faster "strategy"

I’m going to mess around a bit with running the game at fixed FPS

At 100 FPS it yields 1:21.380

At 1,000 FPS it goes up to 1:21.917. Weird

At 10 FPS it’s 3:39.799 :/

Makes me wonder what the ideal FPS for the game is

At 60 FPS it’s 1:22.066

[keen haven]

bet animation speed changes very slightly which causes the difference

[bold summit]

How are TAS's normally scripted? Just a list of input sequences?

[keen haven]

yep, see tminterface and fceux

[pliant patrol]

Yeah, TMInterface was my guide, only other TAS tool I’ve used

And with that it’s literally like:
0 press up
10 steer right
20 steer left
etc…

The next step to making the TAS wrapper would be to do that with what I have so that it’s not as verbose

[keen haven]

@pliant patrol if you ever make a tas tool, PLEASE have it be more like fceux not tminterface. I dislike tminterface sooo much

[pliant patrol]

TASing Balatro is fundamentally different from both because its event based

[bold summit]

@pliant patrol I have posted a new update, it will function at 0 action speed now, which is 1 action per frame if the event condition is met.

Also added selling jokers. Bot.lua has the new TAS already in it. The blind and booster skip timing are still WIP. i am working on a TAS input system (which will double as saving a bot run). How does this format look to you for ease of TASing?:

ACTION,Args

For example

SELECT_BLIND
PLAY_HAND,1,2,3,4,5
PLAY_HAND,1
BUY_CARD,2
END_SHOP
SKIP_BLIND
SELECT_BLIND

feedback appreciated

[pliant patrol]

That format looks great

One thing I would recommend though is a default state just to make writing TASes a bit easier

For instance running SELECT_BLIND if nothing is satisfied

For the future, some good things to add related to TAS would be a UI for savestates/loadstates (autosaving the actions along the way) and undoing actions

And also a way to see future features of a seed, but that is beyond the scope of the tool

Another good thing to add to it would be a way to write comments

The TAS GUI and input tracker could probably be a separate mod as long as the input format remains standardized

[bold summit]

I was thinking of defaults already, I added more actions like rearrange_hand where you can return nothing and it's skipped. I can do that for all actions.

Yeah I hate designing GUIs so that probably won't happen unless someone makes a pull request

[pliant patrol]

There are some that wouldn’t make sense though (mainly some default for playing/discarding)

[bold summit]

The saving and loading of the TAS file format is the extent of my TAS related plans for this mod. My goal was really the botting API so I think I might make a websocket server and a standard gamestate format so I can write my bot in python

[pliant patrol]

I understand that 👍

[bold summit]

But the TAS file format would let me share the results of my bot without sharing my bot code

[pliant patrol]

I can make some better TAS tools when I get a chance, although I’ll probably put most of my modding efforts on Immolate and Ankh right now

[keen haven]

you might want timing of actions. e.g.:

frame num | action

[pliant patrol]

if anything a set FPS could be beneficial

because by default the FPS is dynamic and that affects the time of TASes

[keen haven]

oh yea totally

force 30 / 60 / 120 fps

and tas runs in 30

[bold summit]

SKIP_BLIND
DISCARD_HAND,2,3,6,7
PLAY_HAND,1,3,4,5,8
BUY_CARD,2
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND,1
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND,1
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND,1
BUY_CARD,2
END_SHOP
SKIP_BLIND
SKIP_BLIND
SELL_JOKER,2
PLAY_HAND,1
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND,1
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND,1
END_SHOP
SKIP_BLIND
SKIP_BLIND
SKIP_BOOSTER_PACK
PLAY_HAND,1```

[bold summit]

Logging and replaying are now implemented. Bot.SETTINGS.replay sets whether the Bot will read from a file or log actions as the bot plays (false will log the file)
I haven't gotten to file management really, the file format is just SEED_DECK_STAKE_CHALLENGE.run, and it is written to the Balatro.exe folder (not the mod's folder, need to fix that)

The filename is generated from Bot.SETTINGS

[pliant patrol]

logging is very powerful - I might use some of that code in Ankh (with credit, if you're okay with it)

^that tool is for speedrunning, so having a log of all the actions in a run or session is powerful for verification purposes

[bold summit]

Go for it! When replay is false, Botlogger just hooks each Bot function and gets what each decision function returns and logs it. When replay is true, Botlogger completely replaces the Bot functions and plays them back through a queue system

[pliant patrol]

ah, it hooks the bot functions

I'll probably use that as a reference but when I add it to Ankh it'll use player input

[bold summit]

I will probably be changing the format slightly and turning it into a websocket server so I can write my actual bot logic in Python

[bold summit]

If you haven't checked out my hook.lua, its a small hooking library for breakpoints, callbacks, onreads and onwrites of tables. You can modify return values with the callbacks, or just observe. You could use it to hook whatever receives player input in the engine

[pliant patrol]

ty

in fact, I would probably make that its own library, as it can be used for Ankh as well as an in-game TAS tool

[bold summit]

Put it in its own repo you mean?

Sure, I can do that. I can comment it better as well

[pliant patrol]

I was mainly referring to my own tool that would record player input

but with the context I didn't really clarify that well lol

[bold summit]

oh gotcha

[keen haven]

as said before, timing is probably important, for logging stuff timing can also help

[bold summit]

You could pretty easily rate limit the calls to G.update by hooking it. If someone wants to make a pull request with timing features, I'm all for it, I just have other priorities for the features I'm working on at the moment

Speedrunning isn't really my thing

[pliant patrol]

rate limiting isn't very important, the main idea I have is logging runs
Which would be useful both to check the legitimacy of a play session for Ankh and to record/playback inputs for a TAS tool

[bold summit]

I have the playback working, so you could just have people submit the actions list and then you play it back on your verified setup

[pliant patrol]

yep, I'll try and make something to log runs at some point in the near future

I want to make a website for seed analyzing first

[narrow ivy]

could Twitch Plays Balatro be done with this, or is this purely for tas

[pliant patrol]

The backend could be used for that

[bold summit]

Yes absolutely. You would need to integrate with twitch but you could control the game with this api

[idle spruce]

Is there a record/replay mod that works with this?

Also interested in trying out TASing with it

[keen haven]

not yet afaik

[bold summit]

Yes this works on a basic level at the moment. To replay, change bot.lua Bot.SETTINGS.replay to true.

The replay file will then be read from the folder where Balatro.exe is located. The file name is SEED_DECK_STAKE_CHALLENGE.run, replacing those values from Bot.SETTINGS. For example: "1OGB5WO_Plasma Deck_1_.run", and Bot.SETTINGS must match.

If replay is false, then the actions performed by bot.lua will be logged appropriately (it does not record actions performed by the player). Here is an example replay file:

SKIP_BLIND
SKIP_BLIND
DISCARD_HAND|2,3,6,7
PLAY_HAND|1,3,4,5,8
BUY_CARD|2
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND|1
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND|1
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND|1
BUY_CARD|2
END_SHOP
SKIP_BLIND
SKIP_BLIND
SELL_JOKER|2
PLAY_HAND|1
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND|1
END_SHOP
SKIP_BLIND
SKIP_BLIND
PLAY_HAND|1
END_SHOP
SKIP_BLIND
SKIP_BLIND
SKIP_BOOSTER_PACK
PLAY_HAND|1```

[idle spruce]

What do I use to generate the replay?

[pliant patrol]

It currently only records bot actions

[idle spruce]

Hmm, wonder if something like divvy's history could be easily converted

[idle spruce]

Do mods have file system write access?

[pliant patrol]

Yes

[idle spruce]

It's limited to the balatro folder right?

[pliant patrol]

Yes, unless it uses nativefs

[pseudo light]

Steamodded use Nativefs when used with lovely

And if Nativefs is available

[bold summit]

Yes they do, and it is not limited. Mods have code execution on your system (which means they can do anything a normal user can do). If you do not trust a mod, then do not use it, because it could have malware.

[idle spruce]

did not know that. Steammodded doesn't impose any sandbox limitations on exec?

[bold summit]

Steamodded is also just Lua as far as I know so any changes made could just be changed back. It's theoretically possible that Lovely could modify the Lua runtime to add sandboxing, but that is not an easy thing to do and I wouldn't trust it.

Best thing you can do is make sure you trust the mod author.

[idle spruce]

meh, mods are pretty small for the most part so it's not too hard to just read the code

[bold summit]

And mods also have access to the love2d environment so anything that can be done in love, mods can do.

[bold summit]

    mybot = Bot(deck ="Plasma Deck",
                stake = 1,
                seed = "1OGB5WO")

    mybot.skip_or_select_blind = skip_or_select_blind
    mybot.select_cards_from_hand = select_cards_from_hand
    mybot.select_shop_action = select_shop_action
    mybot.select_booster_action = select_booster_action
    mybot.sell_jokers = sell_jokers
    mybot.rearrange_jokers = rearrange_jokers
    mybot.use_or_sell_consumables = use_or_sell_consumables
    mybot.rearrange_consumables = rearrange_consumables
    mybot.rearrange_hand = rearrange_hand

    mybot.run()```

Here is an example of all that is needed to write your own bot using Python (bot logic not included). Just overwrite each of the functions from the base Bot class with your own decision making logic.

I am still working on sending the entire gamestate to the Python interface. That will be coming next!

[pseudo light]

Steamodded already do some kind of sandboxing, but that's simply to load profiles in an empty env, nothing fancy or sophisticated.

[pliant patrol]

@fallow pumice this might help you in creating a Twitch Plays for this game

[bold summit]

@fallow pumice It should be ready to go for this, since you don't need to read the data from the game. Let me know if you have any questions. And if you finish it let me know, I'd love to see it! Take a look at bot.py and what I'm replying to for the data format. Just pipe it to the lua socket and it should just work. You might need some extra sanity checking

The API already uses text so you wouldn't really need to do anything other than verify that a command is valid (my API kind of does that, but since I provide the client as well, it could use more checks)

[fallow pumice]

this looks awesome, might be a bit too advanced for me but I will check it out

[bold summit]

If you try it and get stuck lmk, I'll see if I can help. I would love to see a twitch plays balatro

[fallow pumice]

okay, ill spend some time today to figure it out. might be back later with some stupid questions.

[fallow pumice]

Ive installed Steamodded, to install Balatrobot do I just need to place the files from the git into the Mods folder?

looking at bot.py could do something like this?

def chooseaction(self):
if self.G['state'] == State.GAME_OVER:
self.running = False

    # Mapping keyboard inputs to class actions
    keyboard_inputs = {
        'b': Actions.SELECT_BLIND,
        's': Actions.SKIP_BLIND,
        'p': Actions.PLAY_HAND,

print("Enter your action: ")
key = input().lower()

    if key in keyboard_inputs:
        action = keyboard_inputs[key]
        return self.perform_action(action)
    else:
        print("Invalid input. Try again.")
        return None

def perform_action(self, action):
    match action:
        case Actions.SELECT_BLIND:
            return self.select_blind(action)
        case Actions.SKIP_BLIND:
            return self.skip_blind(action)
        # Add more cases 
        case _:
            print("Action not implemented.")
            return None

[bold summit]

Yes place the whole thing in the mods folder (unzipped). Structure should be Mods - balatrobot - main.lua (and the subfolders)

Bot_example.py has the bot functions you need to implement

[fallow pumice]

this look correct so far?

my bot works with vjoy to send inputs from twitch chat to the host machine

i can use another program like joytokey to convert the vjoy inputs to keyboard inputs

it a bit hacky but it tends to work well

so once i can have the entire game working via keyboard inputs the rest should be straightforward

[bold summit]

looks loaded to me

Ok so I think for what you are trying to do, you should modify run() of bot.py

        self.verifyimplemented()
        self.state = { }

        self.running = True

        self.sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
        self.sock.settimeout(1.0)

        while self.running:
            self.sendcmd('HELLO')

            jsondata = { }
            try:
                data = self.sock.recv(4096)
                jsondata = json.loads(data)

                if 'response' in jsondata:
                    print(jsondata['response'])
                else:
                    self.G = jsondata
                    if self.G['waitingForAction']:
                        # Choose next action
                        # YOU PROBABLY WANT TO CHANGE THIS SECTION
                        ##########################################################
                        action = self.chooseaction()
                        if action == None:
                            raise ValueError("All actions must return a value!")
                        
                        cmdstr = self.actionToCmd(action)
                        ##########################################################
                        # TO
                        cmdstr = GetNextTwitchChatMessage() # Or something
                        ##########################################################
                        print(f'CMD: {cmdstr}')
                        self.sendcmd(cmdstr)
            except socket.error:
                self.sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
                self.sock.settimeout(1.0)```

@fallow pumice ^

You will only want to send commands to the API when its actually waiting for an action, otherwise you will clog the queue when chat is spamming

The API should ignore all invalid commands

bot_example.py is really designed for implementing logic at each action, but that is the job of twitch chat for your use case, so modifying bot.py is probably better

and you can just get rid of the call to verifyimplemented()

[bold summit]

This API does not take key inputs, it takes decision commands like PLAY_HAND|1,2,3,4,5 so im not sure how helpful your bot would be. But if you modify it to just read twitch chat messages/commands and send it to the bot API, that is your best bet

[solemn trellis]

How do I run the python bot? "python bot_example.py" doesn't seem to do anything

[bold summit]

Did you install the mod with steamodded?

Please paste the error you see if you are getting one

[solemn trellis]

Yeah it's installed with steamodded. It just hangs like this when I run bot_example.py

[bold summit]

Open the game to the main menu first, then run the script

[solemn trellis]

I did, it's still not showing any output after opening the main menu and starting the script

[west spruce]

I've been messing with the python api version quite a bit, and made some modifications to fit my own use case (e.g. getting port from args, and adding the cards in hand to the gamestate)

But I'm currently having an issue where only one copy of the game seems to work at a time. If I run 3 copies configured to different sockets, only the first one will successfully link up with the api. Any idea why that might happen?

[west spruce]

Actually I'm a moron, run() runs synchronously, so the first bot is blocking the others

[west spruce]

Currently hitting about 3 hands per second across 5 instances of the game, so 15 hands played per second total. I noticed something above about 1 action per frame? How is that achieveable?

[bold summit]

The update function is hooked and the API only updates each frame

Each update, the action queue is checked and if there is an action waiting then it's popped and processed

[west spruce]

Gotcha, but that doesn't necessarily mean it's actually possible to take an action every frame

[bold summit]

Right, the animations have to play out, it's not TAS level of accuracy

But the actions will have more or less the same frame the action becomes available if the queue is populated

[west spruce]

Okay great. Yeah I've been working on cutting out/speeding up as many of the animations as possible. Kinda tricky, but definitely running a lot faster

Oh btw, for the server rules, if I make a mod forked off of yours for training an RL agent to play Balatro, is that cool to share here? (And/or do you want me to PR any of the changes I make)

[pliant patrol]

Unless you have a mod to fix the framerate (I made one quickly if anyone wants it), TASes will take slightly different amounts of time

[bold summit]

Yeah PR would be great!

[west spruce]

Opened a PR, gonna focus on setting up an RL gym proof of concept now

[bold summit]

Awesome, I will take a look this weekend

[bold summit]

Looks good, merged it in. Thanks!

[west spruce]

Put together a proof of concept RL training implementation, gonna polish that up and then push it tomorrow probably

[pliant patrol]

What’s the preferred comment format for TAS files? Going to start working on logging for Ankh sometime soon and want the files to be compatible with Balatrobot

[queen willow]

I know that there's the speedup/uncapped fps feature, but do you have a sense of how possible it would be to mod Balatro to run in a purely "headless" mode where it's not doing any rendering and purely doing card logic and communicating with the socket? Is that theoretically possible, or not even worth trying

[west spruce]

It's possible. I reduced rendering to 1/100 frames, and you could disable rendering entirely in a similar way

Unfortunately that doesn't do as much as you'd think to speed up the game without also ripping out a lot of other functionality to skip animations etc

Been working bit by bit to remove those, but it's tricky to do without causing instabilities

[bold summit]

The current bot API does not support comments. But as long as you don't use commas or | as a delimiter for comments then it should be easy to add them

[queen willow]

hm yeah. tricky. Is your fork of balatrobot where that work is going on?

[west spruce]

Kinda yeah. The initial stuff has been merged into Besteon's repo though, and I haven't pushed up the new stuff yet

Learning ray rllib and having some issues transitioning

[fringe bolt]

Is this working with the last version of Balatro and steamodded on Mac? I just tried and out of the box it crashed.

[west spruce]

Might have bricked something with my update, let me look at my changes

Rolling back a commit will probably fix it if that's the case

Seems like it's not getting the port for the socket setup

If you provide it as the first command line argument e.g. 12346 it should work

But it should also be grabbing it from the config file?

[fringe bolt]

Mm I think is just my setup that has something weird, working on windows

[bold summit]

I never tested it on Mac originally

[west spruce]

Adding a bunch more checks and sync logic for the python API and got significant stability and speed improvements

Those seem worth adding on your repo, but I've also been adding lots of experimental stuff that's useful to me, but maybe not for your other use cases

Not sure what the best way of syncing the repos is

[bold summit]

Feel free to make pull requests!

[solemn trellis]

I get this issue with the UI going really fast and crashing when I start a blind

[west spruce]

So for the sped up game logic you can adjust or disable that in config.lua

The crash it looks like a method in the flushbot class is being calling with the wrong number of arguments for some reason

Oh, yeah it looks like it's explicitly passing "self" as the first argument, but then python would add self again automatically

Try removing that first arf

Arg

[pliant patrol]

finally got balatrobot to work

I did one test run and it made it to ante 2 big blind somehow

now to get it to do the thing

[worthy belfry]

How do you check what is remaining in your deck? I couldn't find it in the data but I also could just be blind. Thanks!

[gray ferry]

wrong chat, but that would be the G.deck cardarea

[worthy belfry]

Thanks! Where is the correct chat to ask this question and is there a resource to understand what is contained in G? I asked it here cause I am using Balatrobot and it seems like G.deck is an empty list. Thanks for your help!

[gray ferry]

It's not specific to this mod or any other one mod, so it would probably be best to ask such questions in #💻・modding-dev. As for G.deck, it is an object of the game's CardArea class and not just a list, a list of its cards is kept at G.deck.cards

The best resource for such things would be Balatro's source code itself, which you can view simply by extracting the game executable with something like 7-zip

[worthy belfry]

Is it possible to have the bot give up on the current run and start a new run before getting a gameover?

[west spruce]

I don't think the currently published branch has that ability, but I have some modifications locally

My changes have gotten a bit squirrelly though

I can probably isolate the mod side changes and push those, but it might break some of the python stuff unless I make the corresponding changes there.

[slender shard]

could this be used for a "twitch plays" version of balatro where its soley chat inputs? Left, Right, Up, Down, A, B, X, Y?

[narrow ivy]

i think someone's working on that

[slender shard]

Sick