Hey HN! I built this because existing AI detectors (GPTZero, etc.) are all cloud-based, paid, and require sending your content to third parties.

HumanMark is: - 100% self-hosted (your data never leaves your server) - Zero dependencies (single Go binary) - Multi-modal (text, image, audio, video) - Free forever (MIT license)

Current accuracy is ~60-75%. I'm being upfront about that. The algorithms use statistical analysis (sentence variance, vocabulary richness, AI phrase detection) rather than ML models.

Looking for feedback and contributors to improve accuracy. Would love to hear what use cases you'd find this useful for.

GitHub: https://github.com/vinpatel/humanmark