Miasma: A tool to trap AI web scrapers with poisoned data

What Miasma does
Miasma is a tool designed to trap AI web scrapers by serving them poisoned training data alongside multiple self-referential links, creating what the developers call an "endless buffet of slop for the slop machines." The tool is built to be fast with minimal memory footprint.
Installation and setup
Install with Cargo: cargo install miasma or download pre-built binaries from releases.
Start with default configuration: miasma
View all configuration options: miasma --help
How to trap scrapers
The typical setup involves:
- Embedding hidden links on your site pointing to a specific path (e.g.,
/bots) with attributes that make them invisible to human visitors but visible to scrapers:<a href="/bots" style="display: none;" aria-hidden="true" tabindex="1">Amazing high quality data here!</a>
- Configuring a reverse proxy (like Nginx) to route that path to Miasma:
location ~ ^/bots($|/.*)$ { proxy_pass http://localhost:9855; } - Running Miasma with specific parameters:
miasma --link-prefix '/bots' -p 9855 -c 50
The -c 50 flag limits max in-flight connections to 50, which results in 50-60 MB peak memory usage. Requests exceeding this limit receive a 429 response.
Configuration options
--port: Default 9999 - The port the server should bind to--host: Default localhost - The host address the server should bind to--max-in-flight: Default 500 - Maximum number of allowable in-flight requests--link-prefix: Default / - Prefix for self-directing links (should match your hosting path)--link-count: Default 5 - Number of self-directing links to include in each response page--force-gzip: Default false - Always gzip responses regardless of Accept-Encoding header--poison-source: Default https://rnsaffn.com/poison2/ - Proxy source for poisoned training data
Important considerations
The developers recommend protecting friendly bots and search engines in your robots.txt file:
User-agent: Googlebot User-agent: Bingbot User-agent: DuckDuckBot User-agent: Slurp User-agent: SomeOtherNiceBot Disallow: /bots Allow: /
Miasma is licensed under GPL-3.0 and the developers note that "primarily AI-generated contributions will be automatically rejected."
📖 Read the full source: HN AI Agents
👀 See Also

OpenClaw Local Agent Implementation with TurboQuant Caching for Mid-Range Hardware
A one-click app for OpenClaw with local models now runs on mid-range devices like MacBook Air with 16GB RAM using TurboQuant caching and context warming. The implementation patches llama.cpp for reliable tool calling and achieves 10-15 tokens per second with Gemma 4 and QWEN 3.5.

Developer shares solution for Claude AI ignoring rules beyond 50-count threshold
A developer reports Claude Code started silently dropping rules once their shared rule set exceeded approximately 50 items, particularly during frontend-heavy tasks. They built a hook that scans prompts and loads only 2-3 relevant rules based on keyword matching.

Fullerenes: Open-source persistent memory layer for coding agents cuts tokens by 64% on SWE-bench
Fullerenes uses a local SQLite knowledge graph built via Tree-sitter to give coding agents like Claude Code persistent memory, reducing token usage by 64% on SWE-bench and up to 96.6% on internal benchmarks.

8 Advanced Claude Code Tips: Cost Saving, Context Management, Custom Commands
Practical tips from heavy daily use of Claude Code, covering git workflow automation, multimodal image input, API usage tracking, context compaction, session resumption, rule management, thinking triggers, and custom commands.