Skip to main content
Join the official 2019 Python Developers SurveyStart the survey!

A player made and maintained cheat detection tool for osu!. Provides support for detecting replay stealing and remodding from a profile, map, or set of osr files.

Project description

PyPi version CodeFactor


Circlecore is the backend of the circleguard project, available as a pip module. If you are looking to download and start using the program circleguard yourself, see our frontend repository. If you would like to incorporate circleguard into your own projects, read on.

To clarify, this module is referred to internally as circlecore to differentiate it from the circleguard project as a whole, but is imported as circleguard, and referred to as circleguard in this overview.


First, install circleguard:

pip install circleguard

Circleguard can be run in two ways - through convenience methods such as circleguard.user_check() or by instantiating and passing a Check object to, the latter of which provides more control over how and what replays to compare. Both methods return a generator containing Result objects.

The following examples provide very simple uses of Result objects. For more detailed documentation of what variables are available to you through Result objects, refer to its documentation in the code.

Convenience Methods

For simple usage, you may only ever need to use convenience methods. These methods are used directly by the frontend of circleguard and are generally maintained on that basis, so methods useful in the most number of situations are used. Convenience methods are no different from running circleguard through Check objects - internally, all convenience methods do is create Check objects and run circleguard with them anyway. They simply provide easy usage for common use cases of circleguard, such as checking a specific map's leaderboard.

from circleguard import *

# replace the example api key with your own key - this key is invalid and will not work.
circleguard = Circleguard("5c626a85b077fac5d201565d5413de06b92382c4")

# screen a user's top plays for replay steals and remods.
for r in circleguard.user_check(12092800, num_top=10, num_users=2):
    if r.ischeat:
        # later_replay and earlier_replay provide a reference to either replay1 or replay2, depending on which one was set before the other.
        print("Found a cheater! {} vs {}, {} set later.".format(r.replay1.username, r.replay2.username, r.later_replay.username))

# compare the top 10 HDHR plays on a map for replay steals
# Mod to int documentation:
for r in circleguard.map_check(1005542, num=10, mods=24):
    if r.ischeat:
        print("Found a cheater on a map! {} vs {}, {} set later.".format(r.replay1.username, r.replay2.username, r.later_replay.username))

# compare local files for replay steals
for r in circleguard.local_check("/absolute/path/to/folder/containing/osr/files/"):
     if r.ischeat:
        print("Found a cheater locally! {} vs {}, {} set later.".format(r.replay1.path, r.replay2.path, r.later_replay.path))

# compare two specific users' plays on a map to check for a replay steal
for r in circleguard.verify(1699366, 12092800, 7477458):
    if r.ischeat:
        print("Confirmed that {} is cheating".format(r.later_replay.username))
        print("Neither of those two users appear to have stolen from each other")

More Generally

The more flexible way to use circleguard is to make your own Check object and run circleguard with that. This allows for mixing different types of Replay objects - comparing local .osr's to online replays - as well as the liberty to instantiate the Replay objects yourself and use your own Replay subclasses. See Advanced Usage for more on subclassing.

from circleguard import *
from pathlib import Path

circleguard = Circleguard("5c626a85b077fac5d201565d5413de06b92382c4")

# assuming you have your replays folder in ../replays, relative to your script. Adjust as necessary
PATH = Path(__file__).parent / "replays"
# assuming you have two files called woey.osr and ryuk.osr in the replays folder.
# This example uses python Paths, but strings representing the absolute file location will work just fine.
# Refer to the Pathlib documentation for reference on what constitutes a valid Path in string form.
replays = [ReplayPath(PATH / "woey.osr"), ReplayPath(PATH / "ryuk.osr")]
check = Check(replays)
for r in
    if r.ischeat:
        print("Found a cheater locally! {} vs {}, {} set later.".format(r.replay1.path, r.replay2.path, r.later_replay.path))

# Check objects allow mixing of Replay subclasses. circleguard only defines ReplayPath and ReplayMap,
# but as we will see under Advanced Usage, you can define your own subclasses to suit your needs.
replays = [ReplayPath(PATH / "woey.osr"), ReplayMap(map_id=1699366, user_id=12092800, mods=0)]
for r in
    if r.ischeat:
        # Replay subclasses have well defined __str__ and __repr__ methods, so we can print them directly to represent them in a human readable way if need be.
        print("Found a cheater! {} vs {}, {} set later.".format(r.replay1, r.replay2, r.later_replay))


Circleguard will cache downloaded replays if you give it the path to a database and set the cache option to True. This reduces download times, because replays are stored locally instead of waiting for the quite heavy api ratelimits. You can see more about setting options under Setting Options.

# if the database given doesn't exist, it will be created at the specified location.
cg = Circleguard("5c626a85b077fac5d201565d5413de06b92382c4", "/path/to/your/db/file/db.db")
cg.set_options(cache=True) # can also pass cache=True to a convenience method like map_check, but it will only apply for that single check. This will cache replays for all methods for this circleguard object.

# all 6 replays will be loaded from the api
for r in cg.map_check(221777, num=6):

# the first 6 replays will be loaded from the cache, and only 5 will be loaded from the api, avoiding the 10 replays/min ratelimit.
for r in cg.map_check(221777, num=11)

Caching persists across runs since it is stored on a file instead of in memory; just pass the path to the file when instantiating circleguard.

Advanced Usage

Setting Options

There are four tiers of options. The lowest option which is set takes priority for any given replay or comparison.

Options can be set at the highest level (global level) by using Circleguard.set_options. Options can be changed at the second highest level (instance level) using circleguard#set_options, which only affects the instance you call the method on. Be careful to use the static module method to change global settings and the instance method to change instance settings, as they share the same name and can be easy to confuse.

Options can be further specified at the second lowest level (Check level) by passing the appropriate argument when the Check is instantiated. Finally, options can be changed at the lowest level (Replay level) by passing the appropriate argument when the Replay is instantiated.

Settings affect all previously instantiated objects when they are changed. That is, if you change an option globally, it will change that setting for all past and future Circleguard instances.

Subclassing Replay

If you have needs that are not met by the provided implementations of Replay - ReplayPath and ReplayMap - you can subclass Replay (or one of its subclasses) yourself.

The following is a simple example of subclassing, where each Replay is given a unique id. If, for example, you want to distinguish between loading an otherwise identical replay at 12:05 and 12:07, giving each instance a unique id would help in differentiating them. This is a somewhat contrived example (comparing a replay against itself will always return a positive cheating result), but anytime you need to add extra attributes or methods to the classes for any reason, it's simple to subclass them.

from circleguard import *

class IdentifiableReplay(ReplayPath):
    def __init__(self, id, path): = id

circleguard = Circleguard("5c626a85b077fac5d201565d5413de06b92382c4")
check = Check(IdentifiableReplay(1, "/path/to/same/osr.osr"), IdentifiableReplay(2, "/path/to/same/osr.osr"))
for result in
    print("id {} vs {} - cheating? {}".format(,, result.ischeat))

Although Replay does not have the id attribute by default, because we gave our Check object IdentifiableReplays, it will spit back IdentifiableReplays back at us when we run the check, and we can access our id attribute.

Besides adding information to the Replay through the constructor, you can also control when and how it gets loaded by overloading its load method. The following example is again contrived, because we provide a database implementation for you (any replays you attempt to load through the api will be loaded from the database instead, if you had previously downloaded and cached them), but hopefully gets the point of overloading load across.

from circleguard import *

class ReplayDatabase(Replay):
    def __init__(self, map_id, user_id, mods, detect=Detect.ALL):
        self.map_id = map_id
        self.user_id = user_id
        self.mods = mods
        self.detect = detect
        self.loaded = False

    def load(self, loader, cache=None):
        # execute some sql (implementation not shown) to retrieve replay data from a local database. Assume the call returns a tuple of (replay_id, replay_data)
        result = load_replay_from_database(self.map_id, self.user_id, self.mods)
        replay_id = result[0]
        replay_data = result[1]

        Replay.__init__(self, self.user_id, self.mods, replay_id, replay_data, self.detect, loaded=True)

replays = [ReplayDatabase(1699366, 12092800, 4), ReplayDatabase(1005542, 7477458, 16)]
for replay in replays:
    print("loading replay from local database")

for result in

To get around the rather hairy problem of simultaneously allowing users to instantiate Replay subclasses at any point in their program and only loading them when necessary (when calling circleguard#run(check)), circleguard opts to wait to initialize the Replay superclass until the load method is called and we have all the necessary information that the Replay class requires, either from the api, a local osr file, or some other means.

This means that if you subclass Replay, you must make sure you do a couple of things that circleguard expects from any Replay subclass. Replay must be initialized in your load method (NOT in your __init__ method, as you would expect), and you must set self.weight to one of RatelimitWeight.HEAVY, RatelimitWeight.LIGHT, or RatelimitWeight.NONE in your __init__ method (NOT in your load method! Circleguard needs to know how much of a toll loading this replay will cause on the program before it is loaded). The documentation from the Ratelimit Enum follows, for your convenience:

How much it 'costs' to load a replay from the api. If the load method of a replay makes no api calls,
the corresponding value is RatelimitWeight.NONE. If it makes only light api calls (anything but get_replay),
the corresponding value is RatelimitWeight.LIGHT. If it makes any heavy api calls (get_replay), the
corresponding value is RatelimitWeight.HEAVY.

    This value currently has no effect on the program and is reserved for possible future functionality.


replay_data must be a list of circleparse.ReplayEvent like objects when passed to Replay.__init__. You can look at the circleparse repository for more information, but all that means is that each object must have the time_since_previous_action, x, y, and keys_pressed attributes.

Finally, the load method of the replay must accept one required argument and one positional argument, regardless of whether you use them - loader and cache=None, respectively. If you need to load some information from the api, use the passed Loader class to do so (see the Loader class for further documentation). Should you want to implement a caching system of your own, the cache argument takes care of all the nasty options hierarchy issues and delivers you the final result - should this singular replay be cached? If you choose to cache the replay, you will also have to implement the loading of the replay from the cache, by writing the corresponding logic in the load method. None of that is touched by circleguard - the caching of ReplayMaps happens in an entirely different location than replay#load. So long as you set self.loaded to True by initializing Replay in load, circleguard will respect your replay and assume you have loaded the data properly.

Loading Replays

Normally, all replays in a Check object are loaded when you call circleguard#run(check). However, if you require more control over when you load your replays (or which ones get loaded when you do), you can call circleguard.load(check, replay) to load an individual replay contained in the passed Check object. This is a shorthand method for calling replay#load(circleguard.loader, check.cache), and going through circleguard is always recommended, as not doing so can cause unexpected caching issues with the settings hierarchy not cascading down to the replay correctly. See the last section of Subclassing Replay for more on the optional cache option for replay#load.

There is no limitation on the order in which replays get loaded; when circleguard#run(check) is called, it first checks if check.loaded is True. If it is, it assumes all the replays in the check object are loaded as well and moves on to comparing them. Else, it checks if each replay in the check object have replay.loaded set to True - if so, it moves on to loading the next replay. Otherwise, it calls replay#load.

Modifying Convenience Method Check Before Loading

You may find yourself wishing to perform an action on the Check returned by a convenience method before running it. Although the standard convenience methods create the Check and immediately run it, Circleguard provides methods that only create the Check (circleguard#create_map_check, circleguard#create_user_check, etc).

For instance, the gui Circleguard takes advantage of these methods to load the replays one by one and increment a progress bar before running the check, something that would not be possible with the standard convenience methods.

You can also modify the Check by adding or removing replays before running it. You should see if the recommended approaches for dealing with this, such as the include argument for convenience methods and Check objects, satisfy your needs before resorting to modifying a returned Check.


If you would like to contribute to Circleguard, join our discord and ask what you can help with, or take a look at the open issues for circleguard and circlecore. We're happy to work with you if you have any questions!

You can also help out by opening issues for bugs or feature requests, which helps us and others keep track of what needs to be done next.


Whether you read through everything or scrolled down to the bottom, I hope this helped. If you have any questions, the link to our discord follows. We welcome any comments and are happy to answer questions.


Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for circleguard, version 2.4.0
Filename, size File type Python version Upload date Hashes
Filename, size circleguard-2.4.0-py3-none-any.whl (51.3 kB) File type Wheel Python version py3 Upload date Hashes View hashes
Filename, size circleguard-2.4.0.tar.gz (39.4 kB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page