Terminal task manager with intelligent schedule optimization
At work, people constantly ask "when can you have this done?" — and I wanted to answer that in one second, with a real number, not a gut feeling. Maintaining the schedule by hand is the part nobody does, so the answer is always wrong. Once estimates and deadlines exist, a schedule should fall out automatically. LLMs can do it, but they're slow and non-deterministic — the same input gives a different plan every time. Taskdog is the deterministic, offline, sub-second version of that.
1. Launch the TUI: taskdog tui. 2. Add a few tasks right there — title, estimate, deadline, dependencies — all from the keyboard, no CLI flags to memorize. 3. Open the command palette with Ctrl+P, pick optimize, and watch the schedule lay itself out in the Gantt view. Everything — adding, editing, optimizing, starting/stopping time tracking, reviewing the Gantt — happens inside the TUI, driven from the command palette. The "when can you have it done?" answer is one Ctrl+P away
Scheduling is a pure function, not an LLM call. Given the same tasks, estimates, dependencies, and deadlines, taskdog optimize returns the same plan every time — deterministic, fast (sub-second on hundreds of tasks), and works offline. Architecturally it's split into 5 packages (core / client / server / ui / mcp) so the same scheduler powers the CLI, the Textual TUI, a FastAPI REST server, and an MCP server that lets Claude/agents query and mutate the schedule without owning the algorithm. Keyboard-only, SQLite-local, single-user by design — no dragging, no micromanagement.
Strengthening the retrospective side. Right now Taskdog is great at answering "when will it be done?" up front — the next focus is closing the loop after the fact: surfacing estimate-vs-actual deltas, where time actually went, which task types you consistently under- or over-estimate, and feeding that back into future scheduling so the plans get more honest over time.
By Alvin 
Discussions