Agent is a patient coaching assistant for people who are brand new to CTF and would otherwise stare at a blank terminal wondering where to start. It's not a bot that solves the challenge and hands you the answer. It runs commands on your assigned practice box over SSH, reads the output, explains what's happening, and nudges you toward the next step. When it finds a flag or password it tells you how it got there instead of just blurting the raw flag — because the point is for you to learn.
The "brain" is a small local model served by Ollama. Nothing you type and nothing from the challenge boxes is sent to any cloud service.
You never paste a secret token or create an account anywhere. If your laptop can run it, you can use it — free, forever.
Tuned for ordinary laptop CPUs. Not instant, but it works on the kind of machine most people bring to the event.
One command installs the AI runtime, the agent, and its dependencies, then opens the app in your browser.
No cryptic command line to memorize — a clean chat window with dropdowns for picking your model and connecting to your challenge box.
The chat shows the commands it runs, the output it reads, and its reasoning. Nothing is hidden — that's part of the learning.
The interface
A single, calm screen: live machine telemetry up top, your model and practice box on the left, and a chat that narrates every step the agent takes — think, run a command, read the output, explain. Try it below — pick a suggested question and hit send.
Live, interactive demo running on simulated data so it works right here in the browser. The shipping app wires the same interface to the on-device model and your practice box.
Copy the line for your OS, paste it into your terminal, press Enter. It installs everything and launches the app in your browser.
First run downloads an AI model (a few GB) and may take several minutes. That happens once — after that, start-up is quick.
$ curl -fsSL https://raw.githubusercontent.com/Judgernaut777/CEI-Labs-Agent/main/bootstrap.sh | sh
> irm https://raw.githubusercontent.com/Judgernaut777/CEI-Labs-Agent/main/bootstrap.ps1 | iex
# after the first install, just run: $ ctf-agent
Pick a model. The app checks your free memory and pre-selects the best model that comfortably fits your laptop. If it isn't installed yet, click Install model and watch the progress bar.
Connect to your challenge box. Your CTF instance launcher gives you a host, port, username, and password. Enter them in the SSH form and click Test.
Chat. Ask for help with the level you're stuck on — "Help me figure out level 1 of Bandit." Watch it run commands, read results, and talk you through what it finds.
Bigger, smarter models need more memory. The picker labels every option so you can choose at a glance — when in doubt, take the recommended pick and start playing.
The best all-round pick for a typical laptop. Almost certainly the one you want.
Whether you have enough free memory to run it smoothly. Models that don't fit are greyed out.
Featherweight for older laptops, up through Default and Heavyweight to Max for powerful machines. Higher tiers reason better but need more memory.
Each model offers presets like Standard and Extended that trade memory for more working memory in long sessions.
Its one hands-on tool is running commands over SSH against the challenge box you connect it to. It cannot run commands on your own laptop — local shell access is off by design.
Notes live in a single folder (~/.cei-labs-agent/notes/) and it cannot read or write anywhere else on your computer.
CTF output is sometimes written to trick automated tools. The agent treats challenge output with suspicion and won't blindly obey instructions hidden in it.
"Explain, don't just hand over the flag" is about helping you learn. Anyone can still SSH in and solve by hand — the agent is just a good default for people who'd otherwise be lost.
| Requirement | Detail |
|---|---|
| Laptop | macOS, Linux, or Windows |
| Memory | ~8 GB RAM or more for a smooth experience (a lighter model is picked if you have less) |
| Disk | A few GB free for the AI model |
| Challenge access | SSH details from the CTF event's challenge page |
| Internet | Normal connection for one-time setup; the agent runs offline afterward |
| Not needed | Graphics card · API key · cloud account · prior command-line experience |
Setup snags — disk space, antivirus / SmartScreen warnings, "command not found", Ollama not starting — are covered in the repo's TROUBLESHOOTING.md ↗.
A modular, multi-tenant cyber range on bare-metal K3s.
MODULE 02 — COURSE OF FIREWargames59 challenges, three tracks, one fail-closed deploy.
MODULE 03 — TERRAINNetBuilt to survive 80 operators scanning it on purpose.
MODULE 04 — THE OPERATIONEventA full-scale, self-hosted CTF experience — 80 operators, 59 challenges.