Drosera

A carnivorous honeypot for autonomous AI agents

Design Tools

Developer Tools

Artificial Intelligence

GitHub

Visit WebsiteSee on Product HuntGithub ⧉PyPI ⧉

Hey Product Hunt 👋 I built Drosera because [one or two sentences: what you saw that made you want this — e.g. agent traffic in your own logs you couldn't distinguish from scrapers]. Here's the idea. Autonomous AI agents have a weakness no scraper ever had: they read instructions and act on them. So Drosera hides a short notice in every page — invisible to people — asking automated clients to register with a one-time, HMAC-signed ticket. • A person never sees it. • A crawler parses links and moves on. It can't comply. • An LLM agent reads the sentence, understands it, and complies. That last case is a proof, not a heuristic. Nothing else returns a signed token it could only get by reading a paragraph of English. What you get: 🪤 Lure — bait across HTML, a visible /llms.txt pointer, robots.txt and llms.txt 🔎 Detect — 30 signals, scored on three separate axes: automation, LLM agency, hostility 🕸️ Trap — an endless, noindex'd maze that burns an agent's time and context, or a clean response that ends its task 🐤 Canary — inert fake credentials; if one turns up elsewhere, you have proof of exfiltration The part I care most about: Drosera detects and delays. It does not attack, hijack, or steer someone else's agent. Every lure passes a guardrail in code that rejects anything resembling prompt injection — including your own custom templates. It's Python, zero runtime dependencies, Apache 2.0. Drop in as ASGI/WSGI middleware, run standalone, or score an nginx log you already have without deploying anything. Try it in the browser (no install): https://droseraproject.org Code: https://github.com/rod-trent/Dro... I'd especially love false-positive reports and notes on how different agent frameworks behave against it. Ask me anything.