LLM Fuzz CI проти Open-Source DaiDocs now beats Graphify by 218x low tokens, 83% on LME

A GitHub Action that lets a coding agent write adversarial inputs for your tests, and reports only the ones that fail. Hi HN!Most of the explanation is in the blog post, and the code lives there: https://github.com/tiime-software/llm-fuzz-ci/Happy to receive feedback, and to know how other people use agents to improve code security!

Open plain-text file format for AI memory. Your assistant's long-term memory as .dai files on your disk: readable by Claude, GPT, Gemini, Cursor, local models and grep (all LLM models work). MCP server + hooks for Claude Code, Claude Desktop, Cursor, Windsurf, Codex. 83% LongMemEval-S (GPT-4o), 92% (Claude Fable 5), 10x fewer tokens. - Kerneta/daidocs We have been working on making an Open-Source standard documents for AI LLM's and Agents. Just like an excel file, .dai (Document AI) is a new standard to keep memory local and use with any AI agents and LLM's. This new standard comes with lots of features such as more than 10x tokens savings (tested on LongMemEval-S) when comapred the .dai to raw files and scored 83% with gpt-4-0 and 92% with Fable 5! (This shows how well it remembers a very long conversation), as well as an increased context window (fitting more than 1m tokens chat in a model which has a max of 1m context window)...