Over the last few months my way of working changed quite a lot. I write less code myself and spend more time directing coding agents. On a busy day I have up to ten Claude Code sessions open at once.
With that many sessions, it was difficult to stay focused. A session would finish and wait for my answer, and I would only notice ten minutes later. In between I kept switching to Linear to see what I was supposed to work on, to GitHub for the pull requests waiting for my review, to my own PRs to see if someone had commented, and to the calendar before the next meeting. Each switch was small, but together they broke up the day.
At the end of the day I also often had only a vague idea of what had actually been built, and which decisions had been mine and which the agent's. And because there was always another agent I could start, it was hard to tell when the day was done.
When I talk to other developers who work this way, I hear similar things:
- it's harder to feel ownership of code the agent wrote,
- it's easy to lose track of what's happening across many sessions and branches,
- it's hard to get into flow when you keep checking on the agents,
- and it's not obvious when to stop.
I don't think the answer is more agents or more speed. What I wanted was a calmer way to see where things stand. So I built a small tool for myself, and I'm now sharing it as DevTopo.
What it is
DevTopo runs on your own machine and gives you one page for your day with coding agents. It reads the session files Claude Code and Codex already write, and shows them together with your tickets from Linear or Jira, the pull requests waiting for your review, the reviews on your own PRs, and your calendar.
It is not a tool for running more agents, and it's not meant to make you produce more. It's also not something to monitor anyone with: there is no server and no account, and the data stays on your machine.
Seeing which session needs me
The Today page sorts the sessions into three lanes: running, waiting on you, and idle. With many sessions open, this is the part I use most, because I don't have to remember what is being built in which session.
The same page shows today's plan, my pull requests that got a review, and what I've finished. When a session needs me, one click opens it in the app it runs in, such as VS Code, a terminal, Orca or the Claude app.
Keeping track of what was decided
For each session there is a short summary of what it did today, one line per piece of work with its time. The summaries are written by the Claude Code or Codex you already have, under your own subscription.
If I want more, More detail splits the day into four parts: what was done, what I decided, what the agent decided on its own, and what is still open.
There is also an "Ask about today" window, where I can ask things like "in which session did we fix the login?" and get an answer that points to the session.
Planning the day
The Planner puts my Linear or Jira tickets, the reviews waiting for me and my meetings on one timeline.
It's a simple thing, but deciding in the morning what the day is for keeps me from just reacting to whichever agent finishes next.
At the end of the day there is a list of what I finished, and a History page shows any past day with its sessions and summaries. Having the open items written down makes it a bit easier for me to stop.
Privacy
Since this is about your own work, everything stays local:
- DevTopo only reads the session files the agents already write. It doesn't change your projects or your agent settings.
- It keeps numbers, times and summaries in a local database, not the transcripts.
- The summaries are written by your own Claude Code or Codex, which DevTopo runs without tools that could change files.
- The only request it makes on its own is a twice-daily check for a new version, which you can turn off.
Trying it
DevTopo is free and open source (MIT) and runs on Windows, macOS and Linux: github.com/devtopo-app/devtopo
It's still early. A few friends are testing it, and there will be rough edges. The Windows and Mac builds aren't signed yet, so the first start shows a warning; the README explains how to get past it.
I'd be glad to hear whether these problems sound familiar to you, and how you deal with them. Feedback and issues on GitHub are very welcome.



