What it is
OwnPlanner is a personal AI strategist and mentor. Not another task manager with an AI button — the model is goals-first: the system helps you formulate goals, keeps your daily work connected to them, and uses tasks as the mechanism of execution and monitoring, not the starting point.
I run my own planning on it daily. This project is also my working lab for the engineering questions behind production AI agents: agentic loops, token budgeting, and MCP tool design.
Live instance: app.controlcode.space (invite-only while in active development).
Why goals-first
Most planners optimize for capturing tasks. But a task list without a strategy layer degrades into an inbox you feel guilty about. OwnPlanner inverts the hierarchy: goals carry the context (why, horizon, metric), and the AI mentor uses that context to challenge what you plan day to day.
How AI is integrated
- Agentic loop: the assistant works through model → MCP tool → model cycles against the planner’s own API — the same interface any external agent gets.
- MCP-native: the full domain (goals, tasks, notes, contexts) is exposed as Model Context Protocol tools, so Claude, or any MCP-capable agent, can operate the planner end to end.
- Cost engineering: per-user AI request limits with token accounting, daily budgets, and a cheap default model — the unglamorous work that makes an AI feature sustainable rather than a demo.
Architecture
Clean architecture on .NET 10, separated into domain, application, infrastructure, and presentation layers — with multiple presentation heads over the same core:
- Web: ASP.NET Core server with a React (Vite) frontend
- MCP stdio adapter: protocol-based access for AI agents
- Console: direct CLI access
Test coverage spans domain, application, and infrastructure layers.
Status
Active development. Engineering write-ups from this build appear in the Engineering Log.
Repository: github.com/am-space/own-planner Live instance: app.controlcode.space