Hi PH, I run another project called EvoRadar (evoradar.ai) — an autonomous AI engine that generates and scores startup ideas. Across multiple independent runs, my own engine kept converging on the same niche, 23 separate times: EU AI Act compliance. I built the tool my engine kept pointing at. That's ComplianceLint. Two layers: • MCP scanner in your IDE (Claude Code / Cursor / Windsurf) — reads your codebase and checks the ~10% of obligations that are code-verifiable (logging, data quality, risk classification metadata) • SaaS dashboard — Human Gates capture the 90% needing human judgment (FRIA, Art. 14 oversight assignments, Art. 26(7) worker notifications). Evidence trail. Audit-ready PDF exports (Declaration of Conformity, Technical Documentation). A few decisions worth surfacing: • Deterministic obligation engine — early versions had Claude read the legal text freely. Same code, different answers on different runs. Now every obligation is a structured entry (source_quote, detection_method, what_to_scan). AI is the eyes — the engine is the brain. • Code stays on your machine — MCP scanner runs entirely local. Only findings + evidence submissions sync to the dashboard. • BSL 1.1, source-available (converts to Apache 2.0 on 2030-04-11) — solo founder math: pure OSS doesn't fund 4+ years of obligation tracking, the 2030 sunset makes it credible. Live demo, no signup: compliancelint.dev/demo — pre-loaded with limited-risk chatbot, high-risk finance, and high-risk medical AI systems. Most interested in feedback from anyone wrestling with actual EU AI Act work — especially Art. 17 QMS for small teams, or Art. 14 oversight design patterns. Thanks for taking a look 🙏 — Juan, Solo founder, Munich

ComplianceLint

EU AI Act compliance linter that runs in your IDE

EU AI Act compliance linter that runs in your IDE