I built a local-first Kanban board where AI agents do the work and you stay in control with less cognitive load.
I am building a Task Tracker because I wanted a task manager that gets simpler as AI gets more capable, not more complex.
Most task apps are either cloud-hosted and heavy, or plain text and bare. None were built for agents and my personal progress. So I built one that is.
What makes it different:
- Local-first : One SQLite file. No account. No cloud. No telemetry. Point it at Ollama or any local LLM engine and nothing leaves your machine.
- Native MCP support (18 tools) : Claude, Cursor, Windsurf can plan work, create tasks, log progress, and complete work directly on your board.
- Installable PWA: with reminders that work even when the app is closed.
- Approval-based Chat AI : Ask the assistant to move overdue tasks, create subtasks, or reprioritize. It proposes changes on an approval card. Untick what you don't want. Apply. Everything is checked against board rules and applied all-or-nothing. AI can't delete anything.
- Voice : Talk to the assistant and hear it answer. Hands-free.
- Timeline (Gantt) : Drag bars to reschedule, drag deadlines into start dates. Across individual projects or all of them at once.
- Low cognitive load by design.
GitHub: https://github.com/Parv17k/tasktracker
Built it because I wanted a task manager that gets simpler as AI gets more capable, not more complex.
Would love feedback and stars, especially from people working with agentic AI tools.