Five Practical Open‑Source Projects: FPGA Inference, Agent Alignment, and Multi‑Server SSH Management

FPGA Transformer Inference: gateGPT

gateGPT implements the forward pass of a Transformer model on FPGA hardware using hand‑written Verilog, reproducing the microGPT design by Andrej Karpathy. It runs on a Virtex‑5 FPGA and processes 56 k tokens per second, offering a concrete RTL reference for AI acceleration without relying on high‑level synthesis.

The repository (https://github.com/fguzman82/gateGPT) currently has over 445 stars and is written in Verilog. The project demonstrates how each attention head and feed‑forward layer maps to hardware logic and can be customized into an ASIC for dedicated AI inference.

Agent Personality Alignment Framework: COMPASS Skills

COMPASS Skills addresses the limitation that most AI agents have a single, fixed personality. It introduces a modular set of "Skills" that define prompt strategies, tool‑calling preferences, and output styles for different task contexts, effectively acting as a personality configurator.

Developers can load an "efficiency mode" for coding, a "rigorous mode" for research, or a "creative mode" for writing, with each Skill independently developed, tested, and deployed. The project (https://github.com/dongshuyan/compass-skills) is in early alpha, has 337+ stars, and is implemented in Python.

One‑Command Multi‑Server Execution: boo

boo is a lightweight Zig program built on libssh that lets users run a single command across many hosts simultaneously. Its usage is straightforward: boo "uptime" The command executes uptime on every configured machine, aggregates the results, and displays them, offering a simpler alternative to Ansible or SaltStack for ad‑hoc batch operations. The repository (https://github.com/coder/boo) has 655+ stars and is written in Zig.

AUR Supply‑Chain Malware Detection: AUR Malware Check

Following a large‑scale supply‑chain attack that injected malicious npm install commands into over 1 600 Arch User Repository (AUR) packages, AUR Malware Check aggregates community detection resources into a single script that scans for the atomic‑lockfile attack pattern.

The lightweight, dependency‑free Shell script (also available as a Python 3.14+ module) quickly reports whether a system is compromised. The project (https://github.com/lenucksi/aur-malware-check) has 1 475+ stars and is written in Shell.

Real‑Time Phishing Domain Blacklist API: Destroylist

Destroylist provides a continuously updated blacklist of over 160 000 active phishing domains, backed by a community of 880 000 contributors. It offers a free API supporting multiple data formats, making it easy to integrate into security monitoring, web protection, or threat‑intelligence pipelines.

The repository (https://github.com/phishdestroy/destroylist) has 913+ stars and is primarily HTML‑based.

GitHub: https://github.com/fguzman82/gateGPT | ⭐ 445+ GitHub: https://github.com/dongshuyan/compass-skills | ⭐ 337+ GitHub: https://github.com/coder/boo | ⭐ 655+ GitHub: https://github.com/lenucksi/aur-malware-check | ⭐ 1,475+ GitHub: https://github.com/phishdestroy/destroylist | ⭐ 913+