Shadow IT has always been a headache for security teams, but now we have a new problem: Shadow AI. Developers and data scientists are spinning up local AI servers all over corporate networks, often without telling anyone.
When you are doing a security assessment and you find an open port like 11434 or 8000, you need to know what is actually running there. Is it a harmless test environment, or an exposed AI server leaking company data?
That is exactly what Julius was built to figure out.
What is Julius and what does it do?
Julius is an open-source fingerprinting tool made specifically for security professionals.
There are already tools out there that try to guess which language model generated a specific piece of text. Julius does not care about the text. It cares about the server infrastructure. It looks at an endpoint and tells you if it is running Ollama, vLLM, LiteLLM, Hugging Face, AWS Bedrock, or something else.
It works by sending targeted HTTP requests to the target and looking at how the server responds. Every AI backend has its own quirks, default ports, and API signatures. Julius knows what to look for and gives you an answer in seconds.
What makes Julius useful?
- It knows 60+ AI services: It doesn’t just look for the popular ones. It recognizes Ollama, vLLM, LocalAI, NVIDIA NIM, SGLang, and dozens of others.
It finds the models: Once it identifies the server software, it can also pull a list of the specific AI models loaded on that machine. - Specificity scoring: It gives a score from 1 to 100. This tells you if it found an exact match (like LiteLLM) or just a generic wrapper that happens to be compatible with OpenAI’s API.
- Fast and offline: It is written in Go, runs as a single file, and does not need an internet connection to work. You can scan multiple targets at the same time.
- Easy to update: If a new AI server drops tomorrow, you do not have to wait for the developers to update the tool. You can just write a quick YAML file to teach Julius how to spot it.
Who is this for?
- Penetration Testers: If you are mapping out a client’s network and need to identify unknown AI infrastructure.
Red Teamers: If you are looking for exposed AI endpoints to exploit or use as a pivot point. - Security Engineers: If you are trying to hunt down unauthorized “Shadow AI” servers running on your own corporate network.
The Takeaway
AI infrastructure is growing a lot faster than security teams can track it. You can’t secure what you can’t see, and guessing what is running on an open port is a waste of time.
Julius takes the guesswork out of it. You can grab the code on GitHub or install it straight from your terminal with a single Go command. Next time you see port 11434 open, you will know exactly what to do.
