ai-hacking-lab

Learn about AI Hacking!

Screenshots:

Chatting:

Run locally

1. Install templ, redis, Ollama, and Go.

2. sudo systemctl start redis # start redis
   ollama run tinyllama # install model
   # then...
   templ generate # if you've made any modifications to .templ files
   go run main.go

Project Overview

This website is meant to serve as a simple framework for challenging students to manipulate large language models into doing what they want.

Administrators can provide prompts with CTF flags and challenge students to get the flags from the model. See ./example-prompts.json for example prompts. The URL determines prompt selection with /chat/{promptName}.

Note

Usage of ./ai-hacking-lab:
    -address string
        the address to host the server on (default ":3000")
    -maxTokens int
          the maximum number of tokens in a response. (default 100)
    -modelName string
          the name of the LLM in the Ollama library (default "tinyllama")
    -modelTemperature float
          the 'temperature' of the LLM (default 0.1)
    -redisAddress string
          the address to connect to redis on; must not include protocol (default "localhost:6379")
    -ollamaAddress string
          the address to connect to ollama on; must include protocol (default "http://localhost:11434")
    -promptPath string
          the filepath to load prompts from (default "./example-prompts.json")
  

For example,

go run main.go -promptPath='/your/path/here'

Tech Stack

Technology	Used for
Go	Programming language
templ	HTML Templating
htmx	Render chat messages
redis	Store temporary chats
Ollama	LLM access
LangChain	Integration with Ollama
slog	Structured Logging
go-chi	Router
httprate	Rate limiter
Tailwind CSS	CSS Framework
Hero Icons	Icons