TL;DR: The biggest change in Q2 was moving my scattered AI tool workflow into a local personal operating system that can preserve inputs, knowledge, skills, drafts, previews, and publishing feedback.

---

§Background

On April 1, I wrote a Q1 version of my AI tools and personal operating system. At that point, the focus was building a workflow for input, thinking, output, and productization.

Q1 was more like a tool map: flomo for quick ideas, communities, WeChat, X, and OpenClaw for input, X, WeChat, and Knowledge Planet for output, and YouMind, Codex, GLM, and Gemini for writing and development.

That structure worked at the time. The real problem was that tools could hand work to each other, but experience did not stay in the system. After writing an article, I would revise the opening, delete filler, add details, and test titles. The next run had no memory of why I made those changes.

The core Q2 upgrade was turning the tool flow into a local project.

§How the System Is Structured Now

This personal operating system now lives in the my-digital-twin project.

The first layer is sources/, which stores flomo notes, GetNote entries, interviews, WeChat articles, and X observations.

The second layer is wiki/, which stores identity, writing style, project facts, and content strategy.

The third layer is skills/, which stores reusable capabilities. A skill is an operating manual for an agent: when to use it, how to execute the task, and how to check the result.

The fourth layer is factory/, which stores briefs, drafts, titles, and release checks.

The fifth layer is Factory Workbench, built on May 17. It handles editing, previewing, copying to the WeChat backend, and recording revision feedback.

In Q1, I was organizing a tool flow. In Q2, I started turning that flow into a project.

§Tools Have New Roles

Codex is now the main execution layer. It reads the project, edits skills, generates drafts, runs checks, and operates the browser. You can think of Codex as an agent that can work directly on my local machine.

md2wechat became the WeChat publishing outlet. It now has 2,177 stars, supports 43 structured layout modules, 40+ professional themes, AI images, draft upload, and MCP Server.

flomo is still the L1 cache for fast idea capture. YouMind handles learning and remixing. OpenClaw handles discovery. Gemini Web handles temporary exploration tasks. GLM has moved into lighter planning and support work.

§The Problem I Ran Into

On April 11, I wrote a note in flomo: relying on web tools made writing efficient, but it could not localize the workflow or build a local knowledge base.

Conversations, writing preferences, revision habits, and project facts were scattered across third-party platforms. Every time I switched tools, I had to explain the same context again.

So I moved content production back into the local project. I started by drawing clear folder boundaries, then let agents read those files.

Once materials, drafts, and revision records had stable places to live, the agent knew where to read, where to edit, and where to record the result next time.

§Who This System Is For

If you only write an occasional article, ChatGPT or Gemini is enough.

A local personal operating system has maintenance cost. It is a better fit for people who write for a long time, run long-term projects, or want to preserve their own judgment and writing style.

§If You Want to Build One

First, create input folders and separate voice notes, article excerpts, and external references.

Second, write a style guide that tells AI how you speak and which expressions you dislike.

Third, turn repeated tasks into skills. Start with WeChat writing and title generation.

Fourth, keep briefs, drafts, titles, and revision records. When there is enough evidence, upgrade stable patterns into long-term rules.

§My Q2 Tool List

• ·Input and cache: flomo, GetNote, X, WeChat, communities, OpenClaw
• ·Knowledge and memory: sources/, wiki/, project facts
• ·Creation and execution: OpenAI Codex, md2wechat CLI, Factory Workbench
• ·Learning and remixing: YouMind, OpenClaw, Gemini Web
• ·Output and feedback: WeChat, X, Knowledge Planet, jieni.ai, factory/

Individual tools will change. Models will change too.

What is worth preserving is the input, judgment, style, workflow, and feedback that belong to you.