Open-source project
vladelaina/BongoCat avatar
vladelaina/BongoCat

BongoCat: the default build is a diagnostic backend, and the offline build example uses options the table omits

🩷 💘C × SDL3 × OpenGL — Three Mysterious Forces, United as One! Bong~ Bongocat!!!Don't be shy, Join Here🧸: https://discord.gg/vf8jqnattk

4,092 stars108 forksCAGPL-3.0

At a glance

What is it?
BongoCat is a desktop application in C, built with CMake on SDL3 and OpenGL, distributed through a store listing and through GitHub releases. Its build documentation is unusually complete, with prerequisites per platform, three variants of the same configure command, and a table of options. Two things undercut it: the full runtime needs a proprietary SDK that is not in the repository, and the default when that SDK is missing is a build that configures successfully and renders nothing.
Who is it for?
Build it if you want to read the source or patch the input handling, and treat the option table as incomplete, since three include-directory options appear only in the offline example. Before you conclude the build is broken, check whether a Cubism SDK is present, because the documented default is a warning plus a diagnostic backend rather than a failure, and one flag turns that into a hard error.
Can I use it commercially?
Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
Is it still maintained?
Yes. The repository last received commits 36 days ago.
What is it written in?
Mainly C, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 13, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • vladelaina/BongoCat4,092 ★ · +2,085
Star history
Repository2026-09-212026-11-13
vladelaina/BongoCat2k4.1k

2,085 Stars in the last 53 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

The headline feature needs a proprietary SDK the repository does not contain

The application is a C project built with CMake on SDL3 and OpenGL, and its most visible capability, rendering Live2D models, comes from code that is not here.

The section on the model runtime is marked optional, and it describes a Cubism SDK for Native that you install yourself, either by placing it at a specific path inside the project or by passing its location to the configuration. The SDK has to contain its Core library, the Framework sources, and an OpenGL loader tree, in the layout a particular CMake module expects.

So there are two ways to obtain a working application, and only one of them is self-contained. With the SDK, you get model rendering. Without it, you get what the documentation calls the diagnostic backend.

That is the shape of the project in one sentence: an AGPL application whose interface is a small window, and whose content is a model format supplied by a separate vendor under its own terms. Anyone assessing what they can ship has to look at two licences, not one.

The documented default is a warning, not a failure

This is the single most consequential sentence in the build documentation, and it is easy to miss.

If the model SDK is not present, configuration emits a warning and builds the diagnostic backend instead. That backend is intended for startup and platform diagnostics, and it does not provide model rendering.

So a first-time builder with no SDK sees a clean configure, a successful build, a test run that passes, and an application that draws nothing, having been told about the problem once in a warning they may have scrolled past.

The documentation knows this is a trap and provides the fix. One configuration option, set to OFF by default, makes configuration fail when a usable SDK is unavailable. Turning it ON converts a silent downgrade into an error, which is the behaviour you want in any build that is supposed to produce a working application.

The same pattern explains the Windows prerequisite note, where one compiler can build the diagnostic backend but is not supported for the full SDK runtime.

Five dependencies arrive over the network during configuration

The build has two modes and the default one is the networked one.

By default, five dependencies are downloaded at configure time: SDL3, yyjson, stb, miniaudio, and Nuklear. The documentation states the consequence plainly, that the first configuration needs network access.

The option controlling it is the first row of the table. Its default is ON, and its description says to set it to OFF only when all five are already available to the toolchain. So the offline path is explicitly the exception rather than the default.

The offline requirements are spelled out afterwards, and they are more specific than the option name suggests. Turning fetching off is not enough on its own: you must provide package configurations for two of them, one of which needs a static variant as well, and point the configuration at include directories for the other three.

For a hermetic or air-gapped build, that is three extra settings that the options table never mentions.

The offline example uses three options the options table does not list

There is a table of four options, and then an example that needs seven.

The table covers fetching dependencies, the path to the model SDK, whether a missing SDK should fail the configuration, and whether native compiler warnings should be treated as errors. Those are the four switches a reader would expect to find, with defaults and a description each.

The offline example then adds three more:

code
cmake -S . -B build -G Ninja \
  -DCMAKE_BUILD_TYPE=Release \
  -DBONGO_CAT_FETCH_DEPS=OFF \
  -DBONGO_CAT_STB_INCLUDE_DIR=/path/to/stb \
  -DBONGO_CAT_NUKLEAR_INCLUDE_DIR=/path/to/nuklear \
  -DBONGO_CAT_MINIAUDIO_INCLUDE_DIR=/path/to/miniaudio

Three include-directory options, one per header-only dependency, and not one of them appears in the table that is supposed to be the reference. The text introduces them as include directories to supply when the libraries are not discoverable, which is honest phrasing, but a reader working from the table would not know they exist.

This is the shape of the documentation's one real gap: the prose is careful, the table is the summary, and the summary is missing entries.

Two licence files, and the models are under a different one from the code

The licensing section splits the project in two, which is why there are two licence files in the root.

The source code and the native runtime are licensed AGPL-3.0-only.

The default built-in model mode, named standard, remains MIT licensed, and the bundled model assets in three specific asset directories, named standard, keyboard, and gamepad, are covered by a separate MIT notice. The readme says the MIT licence applies to the model assets rather than the code.

So the permissive terms cover the artwork you get out of the box, and the copyleft terms cover everything you compile. Both notices are in the repository, which is the right way to do this, and the distinction is easy to miss if you only read the licence badge.

Governance files sit alongside them: a security policy, a code of conduct, and a privacy document. A privacy document is an unusual thing to find in a desktop input utility, and its presence says something about how the project sees its own responsibilities.

There is also a project status heading in the readme with nothing written under it.

Tests are on by default, and the test command changes per generator

Two decisions here are unusual enough to be worth naming.

The first is that test targets are enabled by default, so a plain configure gives you a test suite rather than requiring an option to turn one on. The documentation then tells you to run them after building.

The second is that the same test invocation is documented twice, because the command genuinely differs by generator:

code
ctest --test-dir build --output-on-failure

with a configuration flag added for multi-configuration generators such as Visual Studio, where the build produced several configurations and the test runner needs to be told which one to look in.

That flag is not cosmetic. A test run against the wrong configuration reports missing binaries, which reads like a broken build rather than a wrong argument.

The same split shows up in the build instructions, where a single-configuration generator is recommended on Linux and macOS and the Visual Studio generator is used on Windows, and the resulting executable lands in three different paths, one per platform.

One configure command, three shells, three continuation characters

The build documentation repeats the same logical command several times, and the differences between the copies are the interesting part.

On Linux and macOS it is a backslash continuation inside a shell block, with the generator named as Ninja and the build type set to Release. On Windows it is a backtick continuation inside a PowerShell block, with the Visual Studio 17 2022 generator, an architecture argument, and a configuration-aware build command.

So the continuation character is correct for each shell rather than copy-pasted, and the multi-configuration generator also changes the build command, because that is how those generators work.

The dependency-fetching switch appears in both, and the two optional-runtime switches appear in their own block with the same Linux shape, which keeps the Windows and model-SDK instructions consistent with each other.

The one rough edge in the presentation is at the top of the document, where the demonstration video's source attribute is split across two lines in the middle of the address, so the URL as written is broken until the two halves are rejoined.

Editorial conclusion

Build it if you want to read the source or patch the input handling, and treat the option table as incomplete, since three include-directory options appear only in the offline example. Before you conclude the build is broken, check whether a Cubism SDK is present, because the documented default is a warning plus a diagnostic backend rather than a failure, and one flag turns that into a hard error. If you are evaluating the licence, note it is split: the code and native runtime are AGPL-3.0-only, and the three bundled model asset sets sit under a separate MIT notice.

Frequently asked questions

What is a Bongo Cat?

In this repository it is a desktop application written in C, built with CMake on SDL3 and desktop OpenGL. The code and native runtime are licensed AGPL-3.0-only, while the bundled model assets for the standard, keyboard, and gamepad modes are covered by a separate MIT notice, so the project ships under two terms.

How do I build BongoCat?

You need CMake 3.24 or newer, a C11 compiler, a C++17 compiler, and desktop OpenGL development files. Five dependencies, SDL3, yyjson, stb, miniaudio and Nuklear, are downloaded during configuration by default, so the first configure needs network access. Configure with a single-configuration generator such as Ninja on Linux and macOS, or the Visual Studio 17 2022 generator on Windows, then run the tests with ctest.

Why does my BongoCat build not render the model?

If the Cubism SDK is not present, configuration emits a warning and builds what the readme calls the diagnostic backend, which is for startup and platform diagnostics and does not provide model rendering. Setting the requirement option to ON makes configuration fail instead, so a missing SDK stops the build rather than producing an application that draws nothing.

Does BongoCat have an OBS plugin or work on macOS?

The repository describes a standalone desktop application distributed through a store listing and through GitHub Releases, built for Windows, macOS and Linux. It documents macOS prerequisites and an architecture selection setting for the build, and it contains no streaming plugin and no integration for one.

Is Bongo Cat safe?

The repository ships a security policy, a privacy document, and two separate licence files. What it states about itself is narrow: the code and native runtime are AGPL-3.0-only, and the three bundled model asset sets are MIT licensed. Nothing here claims anything about the wider Bongo Cat meme or the many unrelated tools that share the name.

Official sources

  1. License: AGPL-3.0
  2. Project website
  3. README
  4. Releases
  5. vladelaina/BongoCat on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/vladelaina-bongocat.svg)](https://hysenlabs.com/projects/vladelaina-bongocat)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/vladelaina-bongocat.svg)](https://hysenlabs.com/projects/vladelaina-bongocat)

Was this useful?