A new open-source project called Markdown Gatekeeper is attempting to solve one of the most persistent problems with AI agents: the tendency to hallucinate or cite outdated sources when answering questions about specific topics. The tool, posted on Hacker News as a "Show HN" project by developer nanlogic, enforces that AI agents must use exactly one designated Markdown document per topic—eliminating the chaos of conflicting information.

How It Works

The concept is straightforward but clever: instead of letting AI agents pull from whatever sources they find or hallucinate, Markdown Gatekeeper requires operators to define a single canonical source for each domain. When an agent needs information about Topic X, it can ONLY access the designated markdown file for that topic. No web scraping, no training data leakage, no outdated docs—just one trusted reference.

Why This Matters

Hallucination remains AI's dirty secret. Even with RAG (Retrieval-Augmented Generation) pipelines and grounding techniques, agents often drift toward confidently wrong answers. The root cause? Too many potential sources, none of them guaranteed to be current. By enforcing a one-source-per-topic rule at the architecture level, Markdown Gatekeeper removes that variable entirely.

Real-World Applications

The use cases are immediately obvious to anyone running AI systems in production: internal knowledge bases, technical documentation, policy documents, and any domain where accuracy is non-negotiable. Instead of trusting a model's training data or hoping it picks the right retrieval result, operators get deterministic behavior from known-good sources.

Early Community Reception

The project landed on Hacker News with modest fanfare—4 points at time of writing—and zero comments suggests it's still flying under the radar. But for developers building serious agentic systems, this approach to information hygiene might be exactly what they've been looking for.

Key Takeaways

  • Enforces single-source-of-truth per topic in Markdown format
  • Eliminates hallucination by removing access to unverified sources
  • Open-source and self-hostable via GitHub
  • Still early-stage with limited community feedback

The Bottom Line

Markdown Gatekeeper won't solve AI's fundamental limitations, but it smartly addresses a specific failure mode that's been plaguing production deployments. If you're serious about agent reliability, this kind of source discipline isn't optional—it's essential.