Show HN: Hai – A REPL for hackers using LLMs(github.com) hai HN! We can't go a week without a new CLI so today's submission is hai (Hacker AI). There are a lot of tools in the space so I'll focus on what's different: hai is an experiment in personal computing rather than in agentic AI coding. We each rely on a patchwork of web/mobile apps to manage our personal and work lives. They all share the same flaw: these apps aren't tailored to us, and we're powerless because we can't change the app or extract our data. Woe is the hacker ethos. Well, it's 2025 and we have the cloud for our data and AI to transform it as we please. Why not relitigate the 2010s and its legacy of apps and walled gardens? hai's approach is as follows: - Powerful REPL commands: Load files, images, URLs; use tools (e.g. !sh - shell, !py - python); ask the LLM to use the REPL; prompt the human; send email; cache results; and more. - Tasks: Versioned, shareable configs that describe a programmable workflow (i.e. prompt-on-steroids) as a series of REPL commands. Tasks can be published and reused by anyone. - Assets: S3-like blob store for you and your LLM to use: public/private storage, change listeners, JSON metadata separate from data, local syncing, and temporary local copies to bypass chat context. Live tutorial in a browser terminal: http://ec2-98-84-215-186.compute-1.amazonaws.com:1337 Watch a walkthrough: https://www.youtube.com/watch?v=F6qAy8PF2WU (Too slow? Skip the first 3:25 to jump to tools/tasks/assets) (Shorter video on managing a calendar: https://www.youtube.com/watch?v=vfAnEs_Fpx8) Install: |