GitHub radar
Check suspicious code and logs for vulnerabilities offline
OrcaCyber Nano 1.0 is a model built for defensive security work that runs on your own computer with no cloud connection. You get to analyze sensitive code and logs without worrying they'll leak anywhere.
It's an AI model with 27 billion parameters (its size), made by OrcaRouter on top of Qwen from Alibaba specifically for security tasks — it understands both text and images.
When it helps
Useful when you're analyzing a leaked binary, an internal log, or an unpublished vulnerability that can't go into someone else's cloud chat; for everyday coding, use an agent like Claude Code instead.
Pros
- Runs fully offline, nothing leaves your machine
- Reads images too — hand it a debugger or disassembler screenshot
- Parses decompiled code and spots injection flaws in patches
- Sorts logs, alerts, and traffic dumps by what actually matters
Cons
- Needs a GPU with at least 16GB, 24GB+ for image work
- Official benchmark table isn't fully filled in yet
- Not a replacement for your everyday coding agent
How to set it up — step by step
- 1Open the model's page on Hugging Face at the link from the facts.
- 2Send the agent the text from the block below.
- 3Tell the agent how much VRAM your GPU has.
- 4Let the agent set it up and run it through llama.cpp.
- 5Check the result — ask the model to review a code snippet or log for a vulnerability.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up the security-review model OrcaCyber Nano 1.0 locally for me: https://huggingface.co/orcarouter/OrcaCyber-Nano-1.0 — pick the build that fits my GPU, run it through llama.cpp, and show me how to ask it to review a piece of suspicious code or a log for a vulnerability.
Needs a GPU with 16 GB of video memory or more for the text-only build; for image input and the best quality, 24 GB or more.
Open on Hugging Face▌ More finds