I wanted to use Codex on my laptop to run kernel optimization experiments on my GPU box, with models stored in S3. That meant having Codex figure out file transfers and remote execution just to get an experiment running. Lots of opportunities to get something wrong before touching the actual kernel. Then what happens in the next session? Either Codex figures it out again, or the previous session leaves behind scripts and instructions. That works, but now there&#x27;s a small infrastructure project to maintain along with the actual experiments.<p>I decided to try out a more streamlined approach and started building Ridge. It gives agents common operations for accessing data and running commands across different environments, through MCP, a CLI, or Python. Local, SSH, Docker, and S3 are the

  • 情报分类：项目价值、技术价值
  • 命中依据：跨机编码代理协作工具，实用项目
  • 来源：Hacker News 新项目
  • 原作者：vasinov
  • 发布时间：2026/9/11 01:27:36