Bridging the gap between raw technical capability and digital sovereignty. We engineer private, high-performance AI solutions for digital craftsmen and selective enterprises.
Creepybits is an independent AI research and development lab founded by Zanno. We work on local-first AI systems, open inference benchmarks, and ethical identity architectures.
Operating at the intersection of aesthetic vision and hardcore engineering, our brand positions itself as a specialized laboratory focusing on inference performance measurement, local-first systems, and privacy-centric innovation away from centralized control.
Inspect SDSIE Paper & Telemetry
Our approach is evidence-based and scientifically driven, tailored for a selective market that values functional utilitarian design paired with futuristic experimentation.
Relentless exploration into undocumented architectures, sub-byte quantization, and experimental methodologies.
Developing identity systems and tools that respect boundaries without compromising raw capability.
Prioritizing local-first processing, cognitive persistence, and absolute data privacy.
Rigorous measurement, reproducible benchmarks, and carefully designed system architectures.
Differentiated innovation designed for specific, high-intent technical requirements in local and datacenter environments.
Specialized consulting at the intersection of AI research and development to address technical bottlenecks, memory-wall constraints, and architectural projects.
An open, energy-measured comparison of speculative decoding in vLLM and SGLang: 60–79% higher throughput and about 53–55% lower energy per token at batch size 1 on an RTX 5090. Paper, code and data are public.
Read the paper and data →
Development of custom ComfyUI chaining architectures designed to maximize throughput for automated, repetitive production pipelines.
Implementation of Character DNA methodology to achieve visual consistency for brands without using real human likeness or traditional LoRA training.
High-fidelity concept transformation and "Real-to-AI" translation using state-of-the-art techniques to maintain structural integrity during asset modification.
Designing and deploying privacy-focused, local-first AI systems that prioritize data sovereignty and cognitive persistence using custom memory architectures.
Engage with Creepybits for local AI systems, retrieval and workflow automation, or explore our curated tools for digital craftsmen.
