A zero-copy C/Python DFA engine that scrubs logs at 571 MB/s

Guardog, a zero-copy C/Python DFA engine, scrubs logs at 571 MB/s, operating 14.4x faster than standard Python regex by bypassing the GIL and using in-place memory mutation. The engine is available for acquisition and targets hyperscale text sanitization for server logs, LLM prompts, and API ingestion.

Technical Evaluation & Asset Manifest NOTICE OF ASSET SALE The intellectual property, complete source code, and exclusive commercial rights to the Guardog Enterprise Engine are currently available for outright acquisition. This document serves as the technical evaluation manifest for prospective buyers. SeePart 5: Acquisition & IP Transferfor terms. Guardog is a hyperscale, multi-threaded Deterministic Finite Automaton DFA text sanitization engine. Built as a native C-extension for Python, it is designed to intercept, scan, and scrub high-volume payloads server logs, LLM prompt streams, API ingestion for sensitive data in real-time. Clocking at 571+ MB/s , Guardog operates at 14.4x the speed of standard sequential Python re implementations by bypassing the Global Interpreter Lock GIL and utilizing a Zero-Copy memory architecture. It is designed to immediately reduce CPU compute overhead and API latency in high-throughput environments. Standard regex pipelines fail at hyperscale due to three bottlenecks: CPU thread locking, memory duplication, and backtracking latency. Guardog eliminates all three. Standard Python string manipulation forces the server to create complete copies of the payload in RAM. If a server ingests a 50MB log, standard engines spike memory usage by an additional 50MB to process it. The Guardog Advantage: Guardog utilizes Python's Read-Write Buffer Protocol w . It passes a direct pointer to the bytearray into the C-engine. The C-engine reads and mutates the data in-place . RAM overhead during scanning is strictly 0 bytes , preventing Out-Of-Memory OOM crashes in containerized environments. Python’s GIL prevents true multi-core processing. The Guardog Advantage: Guardog drops into raw C, releases the GIL, and deploys OpenMP multi-threading. It slices the payload across available CPU cores. C-threads map secrets into private memory structs Map phase without ever halting to acquire a lock. The GIL is only reacquired once at the very end to dump the structs into a Python list Reduce phase . Standard regex uses Non-Deterministic Finite Automata NFA , which suffers from "catastrophic backtracking." The Guardog Advantage: Guardog pre-compiles all rules into an Aho-Corasick-style DFA matrix. The engine processes exactly one byte per clock cycle, regardless of how many rules are loaded. Whether scanning for 3 secrets or 3,000 secrets, the execution latency remains completely flat. To maintain O 1 latency and Zero-Copy memory efficiency, specific computer science trade-offs were made. Engineering teams must evaluate the following structural profile: Cross-Boundary Secret Detection: OpenMP thread chunks feature a MAX LOOKAHEAD overlap window. If a 20-character API key is perfectly sliced in half by two different CPU cores, the engine still successfully detects and redact it. Overlapping Secret Resolution Two-Pass Mutation : Guardog maps all coordinates in Pass 1, and mutates in Pass 2, guaranteeing 100% detection of adjacent or overlapping secrets prevents the "Amnesia Overwrite" bug . Non-Destructive Pipeline Integrity: Guardog operates on the raw byte layer. It does not force destructive Unicode normalization. Legitimate JSON structures, foreign languages, and mathematical symbols are passed through flawlessly. Cryptographic Tamper Resistance: The compiled .matrix file is locked with a SHA-256 signature to prevent silent fail-open states if the matrix is corrupted. No PCRE Backreferences: Because it is a pure DFA, it has no memory of previously matched groups. No Unbounded Wildcards To prevent L3 Cache misses, the compiler enforces a hard limit of . : 15,000 states. Guardog is designed for structural tokens Keys, SSNs, Credit Cards, JWTs , not free-form linguistic parsing. No Runtime Hot-Swapping: To update or add new rules, DevSecOps must compile a new matrix file, and the application must be restarted to load the new binary signature into RAM. Because Guardog releases the GIL, it is perfectly safe to run inside asynchronous web frameworks like FastAPI without blocking the main event loop. python from fastapi import FastAPI, Request import guardog app = FastAPI title="Secure Ingestion API" Initialize the DFA matrix into memory once at startup detector = guardog.GuardogSession matrix path="guardog.matrix", meta path="guardog meta.json" @app.middleware "http" async def secure payload firewall request: Request, call next : body bytes = await request.body if body bytes: payload text = body bytes.decode 'utf-8', errors='ignore' Scrub payload at 570+ MB/s sanitized = detector.sanitize payload text if sanitized "matches" : print f" SECURITY ALERT Intercepted secrets: {sanitized 'matches' }" return await call next request The asset includes a master orchestrator script that fully automates the cleaning, compiling, verification, and performance evaluation workflows in a single command. Ensure the host environment has Python 3.8+ 64-bit and a C-compiler MSVC with OpenMP for Windows, or GCC/Clang with libgomp for Linux/macOS . To perform a clean end-to-end installation validation, execute: Bash python run pipeline test.py This command will auto-compile the rule matrix, build the C-extension, run the structural boundaries test suite, and output the final MB/s benchmark to standard output. Modifying the Security Rules To add or modify security tokens: Open matrix compiler.py. Locate the DEFAULT RULES dictionary and add your structural strings: Python DEFAULT RULES = { "AWS KEY": "AKIAIOSFODNN7EXAMPLE" , "INTERNAL API": "cyburn prod 77x9a" Custom rules here } Run python matrix compiler.py to regenerate the matrix signatures. This software asset is offered by CyBurn Digital under an Exclusive IP Buyout / Asset Transfer Agreement. Upon execution of the sale, the purchasing entity will receive: 100% Intellectual Property Ownership: Full rights to the C-source code engine.c , compilation architecture, test matrices, and integration blueprints. Unrestricted Usage: The buyer is authorized to integrate, modify, and distribute this compiled binary within their proprietary infrastructure or client deliverables. Royalty-Free: No ongoing licensing fees or volume-based throughput charges. Air-Gapped Security Guarantee: This engine contains zero external network calls, telemetry, or "phone-home" logic. To acquire this asset or request an engineering code audit, please contact: Vindana Sandun Director, CyBurn Digital mailtovindana@gmail.com mailto:mailtovindana@gmail.com +94 76 388 5727 WhatsApp