Attesor: AI-Powered Reverse Engineering of Rosetta 2 for Linux VM

Attesor is a GitHub repository containing a reverse-engineering project focused on Apple's Rosetta 2 binary translation system. The project aims to understand and document how Rosetta 2 enables x86_64 applications to run on ARM64 Apple Silicon hardware, with potential implications for Linux virtualization.
Project Background
The project documents Apple's architecture transitions: 1994 (Motorola 68000 to PowerPC), 2006 (PowerPC to Intel x86_64), and 2020 (Intel x86_64 to Apple Silicon ARM64). Rosetta 2 is Apple's third-generation binary translation solution, following the original Rosetta (2006-2011) that enabled PowerPC applications on Intel Macs.
Rosetta 2 Architecture
According to the source material, Rosetta 2 operates as a translation layer between x86_64 user applications and the ARM64 macOS kernel. The architecture includes:
- Translator (AOT/JIT): Handles both ahead-of-time and just-in-time translation
- Runtime Library: Provides runtime support functions
- System Call Translation: Converts x86_64 syscalls to ARM64 equivalents
Key Technologies
- Ahead-of-Time (AOT) Translation: Translates x86_64 binaries to ARM64 at install time, storing translated code in a cache
- Just-in-Time (JIT) Translation: Translates code blocks on-demand during execution, handling dynamically loaded code
- Instruction Set Translation: Maps x86_64 to ARM64 instructions, SSE/AVX to NEON vector instructions, and x86_64 flags to ARM64 condition codes
- System Call Translation: Manages different calling conventions and register state across syscall boundaries
Implementation Details
Rosetta 2 is located at /Library/Apple/usr/libexec/oah/ (where "oah" stands for "Old Architecture Hardware"), containing:
rosetta- Main translator binaryrosettad- Rosetta daemonlibrosetta.*- Runtime libraries
On Apple Silicon Macs, Rosetta 2 is not installed by default. Installation is triggered either by the first launch prompt of an Intel application or via the command line with softwareupdate --install-rosetta.
Project Structure
The repository contains multiple files including:
ExportDecomp.javaandexport_decomp.pyfor export and decompilationrosetta_decomp.candrosettad_decomp.cfor decompiled componentsrosetta_function_map.hand various refactored C filesrosetta.TODO.mddocumenting remaining work
The project represents an ongoing effort to document Rosetta 2's internals, which could inform development of similar translation layers for Linux virtualization environments.
📖 Read the full source: HN AI Agents
👀 See Also

Hyper iOS App: Voice Recorder with Real-Time Transcription and Action Extraction
Hyper is an iOS voice recorder app that transcribes conversations in real-time, provides summaries and action items, and allows mid-conversation queries via wakeword detection. It's designed for unstructured meetings like 1:1s, coffee chats, and standups.

Framework for AI Teammates in Slack, Operated Entirely from Claude Code
ginnie-agents is an open-source framework for running autonomous AI agents with Slack identities, three-tier memory, cron routines, and work hours — all configured and operated through Claude Code. Requires Claude Code Max, Docker, Node 22+, and a Slack workspace with app creation permissions.

Qwen 3.6 27B with MTP on V100 32GB: 54 t/s via llama.cpp Branch
am17an's MTP branch of llama.cc runs Qwen 3.6 27B at 54 t/s on V100 32GB via PCIe adapter, dropping to 29-30 t/s without MTP.

Running NemoClaw with Local vLLM: Setup Notes and Agent Engineering Observations
A developer documented running NVIDIA's NemoClaw sandboxed AI agent platform with a local Nemotron 9B v2 model via vLLM on WSL2. Key findings include inference routing details, parser compatibility issues, and observations about the agent engineering gap.