Skip to content

Back to the catalog

Armur-Ai/Pentest-Swarm-AI
github.comrepo

Armur-Ai/Pentest-Swarm-AI

Pentest Swarm AI is an open-source (AGPL-3.0), self-hosted CLI tool that runs dozens of LLM agents concurrently against a target, coordinating through a shared "stigmergic" blackboard rather than a fixed recon→exploit→report pipeline, so findings propagate and wake relevant agents in real time. It supports any provider with tool-calling (Claude, OpenAI-compatible endpoints, Together AI-hosted Llama/Qwen/DeepSeek/GLM, or fully local Ollama/LM Studio), ships bundled vulnerable labs (crAPI, Juice Shop, VAmPI, DVGA) for safe testing, and positions itself against the closed SaaS bug-bounty bot XBOW. It currently automates verified attack playbooks (BOLA/IDOR, mass assignment, NoSQL injection) with evidence-backed reporting, with roadmap items for runtime object-reference mining, self-correcting attacks, and on-demand specialist sub-agents.

TL;DR Open-source, self-hosted multi-agent "swarm" pentesting tool that coordinates LLM agents via a stigmergic blackboard to autonomously find, exploit, and report real vulnerabilities.
Open original ↗
Document

What it is

@armurai/pentestswarm (pentestswarm), a CLI/TUI harness that runs a swarm of AI agents against a target concurrently. The summary covers the concept; the page adds the concrete install/run surface and the specific coordination mechanism.

How it works

Three swarm primitives distinguish it from a fixed pipeline:

  • Stigmergy — agents coordinate only by reading/writing findings on a shared blackboard, with no central planner. Each finding carries a pheromone weight that biases other agents toward it and decays over time, so stale paths die on their own.
  • Emergence — no prescribed order. A recon finding wakes the classifier; a high-severity classification wakes the exploit agent; exploit results feed back and wake the report agent.
  • Decentralization — each agent runs its own trigger predicate. You add an agent (with its predicate) and it joins the swarm without editing any orchestrator.

The pitch over a pipeline: concurrency across the whole surface, mid-run discoveries folded back in, and no bound by a slowest stage. It exploits and captures evidence rather than emitting unverified "maybes."

What to do

Install via any of:

npm install -g @armurai/pentestswarm
curl -fsSL https://raw.githubusercontent.com/Armur-Ai/Pentest-Swarm-AI/main/scripts/install.sh | sh
brew install Armur-Ai/tap/pentestswarm
go install github.com/Armur-Ai/Pentest-Swarm-AI/cmd/pentestswarm@latest
docker run --rm ghcr.io/armur-ai/pentestswarm:latest --help

Then run the interactive launcher — no flags:

pentestswarm run

The TUI lets you pick a provider and paste a key inline. Providers named on the page: Together AI (hosted Llama/Qwen/DeepSeek), Claude, OpenAI, Gemini, or local Ollama / LM Studio (no key). Note Gemini appears here but not in the existing summary. Point it at a URL or a bundled lab (crAPI, Juice Shop, VAmPI, DVGA) that spins up, gets attacked, and tears down. Docs open via pentestswarm docs. Node ≥16 for the npm path.

Caveats

  • The core "we're the only real swarm; everything else is a pipeline" claim is the authors' own framing — they explicitly invite counter-examples via issues, so treat the category claim as unverified marketing.
  • Authorized testing only: the project cites CFAA / Computer Misuse Act and requires explicit written permission from the target owner; authors disclaim liability for misuse.
  • The page states no version number, benchmark, or leaderboard result for this tool itself — the "machine speed" and breadth claims are unquantified.

Technologies named here

Techniques named here

Related in Offensive Security