GitHub radar
Run a team of AI agents that keeps working without you
sapper-v2 isn't a single AI model — it's a harness that coordinates several AI agents as a team. One commander hands out tasks, the rest work in parallel and pass their findings to each other.
It's a system for running multiple AI agents at once: one assigns tasks, the others carry them out and share results through a common notes file.
When it helps
Useful if you run your own CTF challenges or authorized pentests and want several AI agents working in parallel without constant supervision. For office work or research, the author recommends starting from the plain harness instead, without the built-in pentest playbook.
Pros
- Found the same flags in a contest while using 4.5x fewer tokens than the runner-up
- Ran unattended for 8 straight hours
- One agent crashing doesn't lose progress — the next one reads the shared notes
- Basic setup runs on a plain laptop, no GPU needed
Cons
- The full toolset needs Docker and about 1.2GB of space
- Heavy reasoning needs your own paid cloud model keys
- The built-in strategy is tuned for a CTF contest, not general tasks
How to set it up — step by step
- 1Open your AI agent (Claude Code or similar) in your project folder.
- 2Clone github.com/N1nEmAn/sapper-v2 and run install.sh — no extra dependencies needed.
- 3Send your agent the text from the block below.
- 4Give it a target you're explicitly allowed to test — your own CTF task or your own server.
- 5Check the task board with sapper status and review the agents' status.
- 6When it asks which model keys to use, connect your own paid keys.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up sapper-v2 from here: https://github.com/N1nEmAn/sapper-v2 — install it per the README (bash install.sh, no external dependencies), swap GLM-5.3 and DeepSeek for my own model or Claude Code, and run sapper setup --goal only against a target I've explicitly authorized you to test, my own CTF challenge or my own server. Show me the task board and agent status, and ask me which model keys to use.
A regular laptop — no GPU needed, the base CLI is pure Python with zero dependencies. The full Docker setup with the recon toolchain needs about 1.2GB of space and Docker itself; the heavy lifting runs on cloud models you plug in with your own paid API keys.
Open on GitHub▌ More finds