BinReX — Task-Adaptive Binary Analysis via Semantic Retrieval and Verifiable Reasoning (USENIX Security '26 artifact)

The online demo is now available. Feel free to try it out. If you have any questions, please email yuzheliu25@m.fudan.edu.cn 

Demo: https://www.binaryai.cn/ng/

----

Research artifact for the USENIX Security '26 paper of the same name. Contains the BinReX agent framework (TypeScript + Python), the BinREval benchmark (72 task oracles, schema, and evaluation harness), prompts, the SHA-256 sample-retrieval workflow, and the dual-encoder training pipeline that implements paper §3 Function Hint Prediction, with a replication path for independent researchers. Malware samples are distributed by SHA-256 only and are not redistributed here; pre-trained dual-encoder weights and the 57M-function corpus are excluded under the Open Science exemption documented in the paper. Equivalent encoders can be reproduced on independent corpora via the included training scripts.