We all have a graveyard of unfinished repos. For a long time, I thought it was a discipline issue. But after abandoning my 15th project, I realized it was an architecture problem in my own brain: Context Decay.
When you take a 3-day break from a solo project, the mental energy required to remember where you left off (was I fixing the Auth hook or doing a DB migration?) and what to do next is so high that you just play video games or commit waisted hours on something related instead like unnecessary polishing or curiosity.
Jira and Linear are great for teams, but for solo devs, they are just glorified spreadsheets. If you forget to update them, they become useless.
The Solution: A "Self-Driving Jira"
I decided to build an AI Project Director that proactively watches my workspace and tells me what to do. I call it Thrust.
How I built it (The Architecture):
Instead of a bulky desktop app, I built it as a lightweight Node.js daemon paired with a web dashboard.
-
The Local Agent: A Node script (
thrust-agent) runs locally, watching my.gitdiffs and VS Code state. - The Gateway: It connects via WebSockets to a local gateway.
- The Brain: I pass the local context + progress timeline + my Project Requirements Document (PRD) to Claude 4.6 Sonnet.
- The Output: When I open my laptop, I get a "Morning Briefing" on the web dashboard: "Yesterday you finished the API. Based on the PRD, you need to build the Login UI today. Here are the files you touched last."
By utilizing MCP (Model Context Protocol) and local project change tracking, the AI actually sees my progress, eliminating the need for me to ever drag a ticket to "Done" or even create one in the first place. No write abilities so you can rest assured that your project files are SAFE, we never store your projects source code in any of our systems.
I'm currently running an Alpha test. If you suffer from Context Decay and want a proactive project lead to keep you on track, I'd love for you to test the application.
Top comments (0)