Show HN: Amanuensis – a local-first AI persona that won't fabricate facts (github.com)

Amanuensis has introduced a groundbreaking local-first AI persona that generates content under direct human oversight. The AI, exemplified by the pilot persona AlexaPavlova, operates exclusively on a user's local GPU, eliminating reliance on cloud-based large language model (LLM) services. This setup allows the AI to draft posts, which must be approved by a human before being published, ensuring factual accuracy and compliance with established topics. The significance of this project lies in its demonstrable commitment to truthfulness in AI output, addressing a common critique of generative AI—its propensity for making false claims. Key technical intricacies include a robust framework that integrates a local text generator, deterministic cleaning processes, and a regex-based validation system to prevent the generation of fabricated technical details. Each post undergoes careful review via a Telegram integration, where users can approve or regenerate content prior to publication on platforms like Mastodon and Bluesky. This innovative model showcases a potential shift towards ethical AI usage, allowing individuals to maintain control over the narrative without sacrificing the benefits of automated content generation. With the code released under the MIT license, developers are encouraged to build upon this foundation, fostering further exploration into responsible AI development.