Data Stays Local

How much of an AI-assisted workflow can actually run on your own machine? This walkthrough traces a local-first system layer by layer, from the moment an AI agent installs it to the one point where it could ever touch the cloud. A public starter kit on GitHub installs itself: point an AI assistant at the repo and it handles every dependency. Below that sits a local Python substrate — video and audio processing, file handling, none of it hosted. Above that, a memory layer stores everything gathered across sessions, retrievable long-term or kept as a short-term working file. A harness sits on top of all of it, deciding which tool to call and when. Every layer below that harness is a local file or a local process; the harness is the only place a cloud call is even possible, and it's a deliberate choice, never a requirement. Chapters 0:00 The Typewriter System on GitHub 0:34 Point an AI Agent at the Repo to Install It 1:08 A Local Python Substrate, Not Hosted 1:38 Local Video and Audio Processing 2:09 Everything You Give It Gets Stored 2:43 Memory Retrieval: Long-Term and Short-Term 3:16 The Harness Sits on Top of Memory 3:49 Workflows Live in Markdown Files 4:24 Privacy Is Built In, No Data Center Needed 4:57 Every Layer Below the Harness Is Local Starter kit shown at the beginning: https://github.com/duhring/typewriter