About DecompilerAI
Origins and goals
DecompilerAI started as a research project focused on neural decompilation. The central question was how close we can get to readable, compilable and functionally correct C code from raw binaries. Current state-of-the-art research enables the development of neural decompilers, which replace components of a traditional decompiler with neural networks, or more concretely, by AI models like LSTMs [1, 2], RNNs [3], GNNs [4], Transformers [5, 6, 7], and even LLMs [8], achieving remarkable results.
The online service at decompiler.ai is an experimental spin-off of this research area. It exists to see whether these techniques are practically useful in everyday work for people who handle binaries. If there is sustained demand from practitioners and researchers, we plan to invest more time into scaling and hardening the platform for broader, industrial use.
Who this platform is for
DecompilerAI is aimed at people who analyze or study binaries. This includes security and malware analysts who need a faster path from bytes to understandable C, engineers in industry who work with in-house or third-party components without source code, and universities or research groups who want to teach reverse engineering or investigate program analysis techniques.
Current status and future directions
DecompilerAI is under active development and intentionally starting with a small and focused feature set. The platform currently targets non-obfuscated ELF binaries for x86 and x86_64, with GCC and Clang as the main compilers. We aim to generate decompilations that are readable, compilable and functionally correct.
DecompilerAI steadily tries to improve the decompiler service by employing continous evaluations using the DecEvalSE framework, and other techniques from recent research to improve readability and functional correctness.
Some technical directions we are exploring include support for additional architectures and binary formats, improving decompilation quality, extending from C towards C++ where it is useful, collecting more metrics about the binary and the decompilation.
Another area of interest for the future is security-focused tooling on top of the decompiler. This includes features that help evaluate potential security risks in binaries, scan for known vulnerability patterns and produce structured reports that are useful for security reviews and audits.
All of these plans are goals rather than guarantees. The roadmap may change depending on feedback, technical constraints and available resources.
Trust, privacy and responsibility
Uploaded binaries and generated outputs are treated as confidential and are used only to provide the service to you. We do not use your uploaded programs or chat logs to train our own models. We may collect metadata about uploaded files, such as architecture or file size, to operate and improve the platform. Details are described in the Privacy Policy.
Some internal processing steps may use external AI providers, such as OpenAI, to improve the readability and structure of decompiled code. When external providers are used, we aim to limit what is sent and how long it is retained, and we rely on appropriate contractual and technical safeguards.
You remain responsible for ensuring that you have the legal right and necessary permissions to analyze, reverse engineer or decompile any binary you upload. DecompilerAI does not verify your rights to any specific file and does not provide legal advice. Reverse engineering is regulated differently in different jurisdictions, and you must comply with the laws and agreements that apply to your situation.
Decompilation is inherently approximate. Outputs can be incorrect, misleading, non-compilable or not fully equivalent to the original binary. You must independently review and test any output before relying on it in production, security work or other critical contexts. Information about the limits of the service, refund rules and liability restrictions can be found in the Terms of Service.
Contact and community
If you have questions, want to discuss research or teaching use, or have suggestions and bug reports, you can reach out via the contact page on this site or through the social channels linked in the footer.
DecompilerAI is present on Twitter, YouTube and Discord. You are welcome to get in touch there if you want to talk about reverse engineering, decompilation benchmarks or potential collaborations.
