I built a local AI agent because I kept forgetting what I did at work all week

Hi DEV community! 👋

Let me start with a confession: I am terrible at weekly and monthly summaries.

Every day, I do a lot of things. I jump between coding, debugging, reading documentation, answering messages, and handling sudden ad-hoc requests. But when Friday afternoon rolls around and it's time to write my weekly report, my mind goes completely blank. I literally forget what I actually accomplished.

Worse yet, this constant context-switching leaves me in a perpetual state of chaotic task planning. I often lose track of what I should be focusing on next.

I desperately needed a solution—a local work assistant that could "watch" my every move, help me untangle my daily schedule, and maybe even take some of the simple, repetitive tasks off my plate.

Because I couldn't find a privacy-first tool that did exactly this, I decided to build it myself.

Meet DoWhat — an open-source, AI-native personal work agent that lives entirely on your local machine.

💡 How DoWhat solves the chaos

Running silently in the background (built with Electron and React), DoWhat is designed to be your intelligent digital twin:

• It never forgets (Passive Tracking): It automatically takes a screenshot every 15 seconds. You don't have to manually log anything.
• It writes your summaries (AI Vision): Using any OpenAI-compatible vision model you configure, it groups your activities into 15-minute slots and generates AI summaries of what you were actually doing. When monthly review time comes, you have a perfect, automated timeline.
• It untangles your tasks (Proactive Planning): The AI automatically extracts actionable tasks from your work context and adds them to a backlog, helping you regain control over your schedule.
• It helps you do the work (Agent Execution): I recently integrated the OpenClaw agent framework. Now, you can delegate those extracted tasks to the agent for autonomous completion or connect it directly to IM apps like WeChat and Feishu.

🔒 Privacy First: Why "Local" matters

Having a tool monitor your "every move" sounds terrifying if the data goes to the cloud. That was my non-negotiable red line.

All your data stays yours. The screenshots and the SQLite database are stored purely on your local machine (userData directory). There is no telemetry, no forced cloud sync, and no user accounts. It only communicates with the specific AI API endpoint that you provide.

🛠️ The Tech Stack

To keep the UI snappy and the architecture maintainable, I went with:

• Core: Electron 39
• Frontend: React 19 + TypeScript 5 + Tailwind CSS 4
• Storage: SQLite (via better-sqlite3)

🙏 A Humble Request for Feedback (and a Star!)

I am developing this in my free time to solve my own daily struggles, and I'm sure the codebase has plenty of room for improvement. The architecture might not be perfect yet, and there are still bugs to squash.

If you struggle with chaotic task planning or forgetting your weekly achievements like I do, I would love for you to try it out.

More importantly, I would be incredibly grateful if you could check out the repo and give it a ⭐️ on GitHub! 👉 GitHub Repository: Laworigin/DoWhat

A single star is a huge source of motivation for an independent open-source creator. Also, if you have any architectural critiques or want to contribute a PR, my doors are wide open. I'm here to learn from this amazing community.

Thank you so much for reading!

— wisdomtree