Kaitai is absolutely one of my favorite projects. I use it for work (parsing scientific formats, prototyping and exploring those formats, etc) as well as for fun (reverse engineering games, formats for DOSbox core dumps, etc).
I gave a guest lecture in a friend's class last week where we used Kaitai to back out the file format used in "Where in Time is Carmen Sandiego" and it was a total blast. (For me. Not sure that the class agreed? Maybe.) The Web IDE made this super easy -- https://ide.kaitai.io/ .
(On my youtube page I've got recordings of streams where I work with Kaitai to do projects like these, but somehow I am not able to work up the courage to link them here.)
RE, especially of older and more specialized software, involves dealing with odd undocumented binary formats. Which you may have to dissect carefully with a hex editor and a decompiler, so that you can get at the data inside.
Kaitai lets you prototype a parser for formats like that on the go, quick and easy.
A shot in the dark, but maybe you could give me a hint. Recently, I was interested in extracting sprites from an old game. I was able to reverse the file format of the data archive, which contained the game assets as files. However, I got stuck because the image files were obviously compressed. By chance, I found an open source reimplementation of the game and realised it was LZ77+Huffman compressed, but how would one detect the type of compression and parameters with only the file? That seems a pretty hard problem or are there good heuristics to detect that?
Some simpler cases like various RLE-type encodings can be figured out with that pattern recognizing brain - by staring at them really really hard.
For harder cases? You take the binaries that read or write your compressed files, load them in your tool (typically Ghidra nowadays), and track down the code that does it.
Then you either recognize what that code does (by staring at it really really hard), or try to re-implement it by hand while reading up on various popular compression algos in hope that doing this enlightens you.
Third option now: feed the decompiled or reimplemented code to the best LLM you have access to, and ask it. Those things are downright superhuman at pattern matching known algorithms, so use them, no such thing as "cheating" in RE.
The "hard mode" is compression implemented in hardware, with neither a software encoder or a software decoder available. In which case you better be ready for a lot of "feed data to the magic registers, see results, pray they give you a useful hint" type of blind hardware debugging. Sucks ass.
The "impossible" is when you have just the compressed binaries, with no encoder or decoder or plaintext data available to you at all. Better hope it's something common or simple enough or it's fucking hopeless. Solving that kind of thing is cryptoanalysis level of mind fuck and I am neither qualified enough nor insane enough to advise on that.
Another thing. For practical RE? ALWAYS CHECK PRIOR WORK FIRST. You finding an open source reimplementation? Good job, that's what you SHOULD be doing, no irony, that's what you should be doing ALWAYS. Always check whether someone has been there and done that! Always! Check whether someone has worked on this thing, or the older version of it, or another game in the same engine - anything at all. Can save you literal months of banging your head against the wall.
Thanks for your reply and advice! I guess what you describe as "impossible" is the case I am mostly interested in, though more for non-executable binary data. If I am not mistaken, this goes under the term "file fragment classification", but I have been wondering if practitioners might have figured out some better ways than what one can find in scholarly articles.
And yes, searching for the reimplementation beforehand would have saved me some hours :D
It's not about the data being executable. It's about having access to whatever reads or writes this data.
Whatever reads or writes this data has to be able to compress or decompress it. And with any luck, you'll be able to take the compression magic sauce from there.
I understood "binaries" in "compressed binaries" as "executables", e.g. like a packed executable, but I see that you mean indeed a binary file (and not e.g. a text file).
Wow. Looking at the schema for gif, it’s so readable, I can’t help to wonder why something like this hasn’t become the standard way to work with binary formats over the decades already!
Seems like eveything has to be JSON and text based these days, because binary is more difficult DX.
When reading articles discussing binary formats, I usually see them using box diagrams of packets, description tables or hexdumps.
This neatly describes nested structure, names and ”types” - just enough.
I wonder if there’s a hexdump like viewer in IDEs that can present binary files like this? I can also imagine a simple UI to make the files editable using this.
One of my personal favorites. I've used it for parsing SAP's RPC network protocol, reverse-engineering Garmin apps [0], and more recently in a CTF challenge that involved an unknown file format, among others. It's surprisingly quick to pick up once you get the hang of the syntax.
The serialization branch for Python [1] (I haven't tried the Java one) has generally done the job for me, though I've had to patch a few edge cases.
One feature I've often wished for is access to physical offsets within the file being parsed (e.g. being able to tell that a field foo that you just parsed starts at offset 0x100 from the beginning of the file). As far as I know, you only get relative offsets to the parent structure.
Wow this is good. My only complaint is annoyingly verbose yaml. What if I would like to use Kaitai instead of protobuffs, my .proto file is already a thousand lines, splitting each od these lines into 3-4 yaml indented lines is hurting readability
KS isn't for general data mangling, it's for "I have this format and I need a de novo parser for it that works under explicit rules" and you're willing to do the work of fully implementing it from the bytes up.
Even if you don't want to use it since it is not as efficient as a hand-written specialized parser, Kaitai Struct gives a perfect way of documenting file formats. I love the idea and every bit of the project!
My dream for a parsing library / language is that it would be able to read, manipulate, and then re-serialize the data. I'm sure there are a ton of edge cases there, but the round trip would be so useful for fuzzing and program analysis.
> Kaitai Struct is in a similar space, generating safe parsers for multiple target programming languages from one declarative specification. Again, Wuffs differs in that it is a complete (and performant) end to end implementation, not just for the structured parts of a file format. Repeating a point in the previous paragraph, the difficulty in decoding the GIF format isn't in the regularly-expressible part of the format, it's in the LZW compression. Kaitai's GIF parser returns the compressed LZW data as an opaque blob.
Taking PNG as an example, Kaitai will tell you the image's metadata (including width and height) and that the compressed pixels are in the such-and-such part of the file. But unlike Wuffs, Kaitai doesn't actually decode the compressed pixels.
---
Wuffs' generated C code also doesn't need any capabilities, including the ability to malloc or free. Its example/mzcat program (equivalent to /bin/bzcat or /bin/zcat, for decoding BZIP2 or GZIP) self-imposes a SECCOMP_MODE_STRICT sandbox, which is so restrictive (and secure!) that it prohibits any syscalls other than read, write, _exit and sigreturn.
They overlap, but none does strictly more than the other.
Kaitai is for describing, encoding and decoding file formats. Wuffs is for decoding images (which includes decoding certain file formats). Kaitai is multi-language, Wuffs compiles to C only. If you wrote a parser for PNGs, your Kaitai implementation could tell you what the resolution was, where the palette information was (if any), what the comments look like and on what byte the compressed pixel chunk started. Your Wuffs implementation would give you back the decoded pixels (OK, and the resolution).
Think of Kaitai as an IDL generator for file formats, perhaps. It lets you parse the file into some sort of language-native struct (say, a series of nested objects) but doesn't try to process it beyond the parse.
I wanted to use this a long time ago but the rust support wasn't there. I can see now that it's on the front page with apparently first class support so looks like I can give it a go again.
This seems to say flags is a sort of unsigned integer.
Is there a way to break the flags into big endiaN bits where the first two bits are either 01 or 10 but not 00 or 11 with 01 meaning DATA and 01 meaning POINTER with the next five bits as a counter of segments and the next bit is 1 if the default is BLACK and 1 if the default is WHITE ?
I had a ton of fun using Kaitai to write an unpacking script for a video game's proprietary pack file format. Super cool project.
I did NOT have fun trying to use Kaitai to pack the files back together. Not sure if this has improved at all but a year or so ago you had to build dependencies yourself and the process was so cumbersome it ended up being easier to just write imperative code to do it myself.
Kaitai is pretty nice. Hex editors with structure parsing support used to be more rare than they are now, so I've used https://ide.kaitai.io/ instead a few times.
Also, the newest Kaitai release added (long awaited) serialization support! I haven't had a chance to try it out.
I didn't check exactly what Kaitai does but, MPEG uses a custom SDL for it's binary syntax: https://mpeggroup.github.io/mpeg-sdl-editor/
Just sharing, in case someone is interested :)
Kaitai Struct is really great. I've used it several times over the years to quickly pull in a parser that I'd otherwise have to hand-roll (and almost certainly get subtly wrong).
Their reference parsers for Mach-O and DER work quite nicely in abi3audit[1].
Does it support incremental parsing? For example, when I am parsing a network protocol, can it still consume some data from the head of the buffer even if the data is incomplete? This would not only avoid multiple attempts to restart parsing from the beginning but also prevent the buffer from growing excessively.
Enjoyable tool. When I developed my text RPG game, I prepared a Kaitai specification for the save file data format so that it would be easy to create third-party software for viewing and modifying it =)
The recent release of 0.11 marks the inclusion of the long awaited serialization feature. Python and Java only for now.
I've been using it for a while for Python and although it has some rough edges, it works pretty well and I'm super excited for the project.
The most success I had so far on doing a project where I had to work with binary data parsing is Deku in rust, I would give this a try if I have the opportunity
It's not C but we have sponsored a Zig target for Kaitai. If anyone reading this knows Zig well, please comment because would love to get a code review of the generated code!
It would be premature to review now because there are some missing features and stuff that has to be cleaned up.
But I am interested in finding someone experienced in Zig to help the maintainer with a sanity check to make best practices are being followed. (Would be willing to pay for their time.)
If comptime is used, it would be minimal. This is because code-generation is being done anyway so that can be an explicit alternative to comptime. But we have considered using it in a few places to simplify the code-generation.
I discovered this project recently and used it for Himawari Standard Data format and it made it so much easier. Definitely recommend using this if you need to create binary readers for uncommon formats.
I gave a guest lecture in a friend's class last week where we used Kaitai to back out the file format used in "Where in Time is Carmen Sandiego" and it was a total blast. (For me. Not sure that the class agreed? Maybe.) The Web IDE made this super easy -- https://ide.kaitai.io/ .
(On my youtube page I've got recordings of streams where I work with Kaitai to do projects like these, but somehow I am not able to work up the courage to link them here.)
RE, especially of older and more specialized software, involves dealing with odd undocumented binary formats. Which you may have to dissect carefully with a hex editor and a decompiler, so that you can get at the data inside.
Kaitai lets you prototype a parser for formats like that on the go, quick and easy.
For harder cases? You take the binaries that read or write your compressed files, load them in your tool (typically Ghidra nowadays), and track down the code that does it.
Then you either recognize what that code does (by staring at it really really hard), or try to re-implement it by hand while reading up on various popular compression algos in hope that doing this enlightens you.
Third option now: feed the decompiled or reimplemented code to the best LLM you have access to, and ask it. Those things are downright superhuman at pattern matching known algorithms, so use them, no such thing as "cheating" in RE.
The "hard mode" is compression implemented in hardware, with neither a software encoder or a software decoder available. In which case you better be ready for a lot of "feed data to the magic registers, see results, pray they give you a useful hint" type of blind hardware debugging. Sucks ass.
The "impossible" is when you have just the compressed binaries, with no encoder or decoder or plaintext data available to you at all. Better hope it's something common or simple enough or it's fucking hopeless. Solving that kind of thing is cryptoanalysis level of mind fuck and I am neither qualified enough nor insane enough to advise on that.
Another thing. For practical RE? ALWAYS CHECK PRIOR WORK FIRST. You finding an open source reimplementation? Good job, that's what you SHOULD be doing, no irony, that's what you should be doing ALWAYS. Always check whether someone has been there and done that! Always! Check whether someone has worked on this thing, or the older version of it, or another game in the same engine - anything at all. Can save you literal months of banging your head against the wall.
And yes, searching for the reimplementation beforehand would have saved me some hours :D
Whatever reads or writes this data has to be able to compress or decompress it. And with any luck, you'll be able to take the compression magic sauce from there.
Seems like eveything has to be JSON and text based these days, because binary is more difficult DX.
When reading articles discussing binary formats, I usually see them using box diagrams of packets, description tables or hexdumps.
This neatly describes nested structure, names and ”types” - just enough.
I wonder if there’s a hexdump like viewer in IDEs that can present binary files like this? I can also imagine a simple UI to make the files editable using this.
The serialization branch for Python [1] (I haven't tried the Java one) has generally done the job for me, though I've had to patch a few edge cases.
One feature I've often wished for is access to physical offsets within the file being parsed (e.g. being able to tell that a field foo that you just parsed starts at offset 0x100 from the beginning of the file). As far as I know, you only get relative offsets to the parent structure.
0: https://github.com/anvilsecure/garmin-ciq-app-research/blob/...
1: https://doc.kaitai.io/serialization.html
KS isn't for general data mangling, it's for "I have this format and I need a de novo parser for it that works under explicit rules" and you're willing to do the work of fully implementing it from the bytes up.
> Kaitai Struct is in a similar space, generating safe parsers for multiple target programming languages from one declarative specification. Again, Wuffs differs in that it is a complete (and performant) end to end implementation, not just for the structured parts of a file format. Repeating a point in the previous paragraph, the difficulty in decoding the GIF format isn't in the regularly-expressible part of the format, it's in the LZW compression. Kaitai's GIF parser returns the compressed LZW data as an opaque blob.
Taking PNG as an example, Kaitai will tell you the image's metadata (including width and height) and that the compressed pixels are in the such-and-such part of the file. But unlike Wuffs, Kaitai doesn't actually decode the compressed pixels.
---
Wuffs' generated C code also doesn't need any capabilities, including the ability to malloc or free. Its example/mzcat program (equivalent to /bin/bzcat or /bin/zcat, for decoding BZIP2 or GZIP) self-imposes a SECCOMP_MODE_STRICT sandbox, which is so restrictive (and secure!) that it prohibits any syscalls other than read, write, _exit and sigreturn.
(I am the Wuffs author.)
Wuffs is intended for files. But, would it be a bad idea to use it to parse network data from untrusted endpoints?
There's also a "wget some JSON and pipe that to what Wuffs calls example/jsonptr" example at https://nigeltao.github.io/blog/2020/jsonptr.html#sandboxing
Kaitai is for describing, encoding and decoding file formats. Wuffs is for decoding images (which includes decoding certain file formats). Kaitai is multi-language, Wuffs compiles to C only. If you wrote a parser for PNGs, your Kaitai implementation could tell you what the resolution was, where the palette information was (if any), what the comments look like and on what byte the compressed pixel chunk started. Your Wuffs implementation would give you back the decoded pixels (OK, and the resolution).
Think of Kaitai as an IDL generator for file formats, perhaps. It lets you parse the file into some sort of language-native struct (say, a series of nested objects) but doesn't try to process it beyond the parse.
This seems to say flags is a sort of unsigned integer.
Is there a way to break the flags into big endiaN bits where the first two bits are either 01 or 10 but not 00 or 11 with 01 meaning DATA and 01 meaning POINTER with the next five bits as a counter of segments and the next bit is 1 if the default is BLACK and 1 if the default is WHITE ?
I did NOT have fun trying to use Kaitai to pack the files back together. Not sure if this has improved at all but a year or so ago you had to build dependencies yourself and the process was so cumbersome it ended up being easier to just write imperative code to do it myself.
https://github.com/dloss/binary-parsing
Personally I like GNU Poke.
For example:
[1]: https://github.com/evincarofautumn/protodata
Also, the newest Kaitai release added (long awaited) serialization support! I haven't had a chance to try it out.
https://kaitai.io/news/2025/09/07/kaitai-struct-v0.11-releas...
Their reference parsers for Mach-O and DER work quite nicely in abi3audit[1].
[1]: https://github.com/pypa/abi3audit/tree/main/abi3audit/_vendo...
I used Kaitai in a IoT project for building data ingress parsers and it was great. But not having write support was a bummer.
https://hachoir.readthedocs.io/en/latest/index.html
https://github.com/kaitai-io/kaitai_struct_compiler/commits/...
It would be premature to review now because there are some missing features and stuff that has to be cleaned up.
But I am interested in finding someone experienced in Zig to help the maintainer with a sanity check to make best practices are being followed. (Would be willing to pay for their time.)
If comptime is used, it would be minimal. This is because code-generation is being done anyway so that can be an explicit alternative to comptime. But we have considered using it in a few places to simplify the code-generation.
DFDL is heavily encroaching on Kaitai structs territory.
highly recommended if you like functional languages