Google released the OKF specification, standardizing Andrej Karpathy's "LLM-Wiki" model.

According to Beating's monitoring, the biggest obstacle for companies implementing large-scale models is the lack of internal background knowledge (such as table structures, workflows, or API specifications). Because knowledge is scattered across different systems, code comments, or even in employees' minds, developers have to repeatedly build retrieval pipelines for different AI tools. To address this, Google released the Open Knowledge Format (OKF) v0.1 specification, formally standardizing the "LLM-Wiki" model proposed by Andrej Karpathy. The specification requires no new software or development kits, advocating that all internal knowledge be written as intuitive Markdown text files and hosted in the code repository along with the source code. Technically, OKF knowledge packages consist of a Markdown directory tree. Each file uses text tags at the top to declare its type, description, and date, and files are linked together via hyperlinks; it also includes index.md for quick previewing the architecture and log.md for recording update history. Andrej Karpathy points out that humans dislike updating wikis because manually maintaining the directory and cross-file links is extremely tedious. The AI will not forget to update cross-references and can process and edit up to 15 files at a time, making it ideal as a "document administrator" to take over daily maintenance. Furthermore, the specification employs a tolerant parsing model. Even if the AI omits fields, mistypes data, or leaves dead links while autonomously generating files, the system will not crash or report errors, ensuring extremely high fault tolerance.