Intention
The task in one line. Planned duration. A resistance check-in (1–5). The thing you would say is the actual outcome if it goes well. Captured before the session starts; revisited at the end.
Not Pomodoro. A focused-work substrate that captures intention before the session starts, detects stalls in real time, surfaces a facilitator nudge when one is needed, and runs structured reflection at the end. Built for the people who do the hardest work alone.
Pomodoro counts down. Habit trackers tally. Both are fine for the easy half of the problem — starting the session — and useless for the hard half: the moment, twenty minutes in, when you are stuck, and the choice between pushing through and quietly drifting onto your phone is the whole game.
Dojo is the substrate we built around that moment. The session starts with intention — a one-line task and a check-in on resistance. Stuckness signals are detected as you work. When the model is confident a stall is real, a facilitator prompt appears (a peer, a coach, or a tuned LLM, depending on your install). The session ends with a structured reflection that is short enough to actually answer.
The task in one line. Planned duration. A resistance check-in (1–5). The thing you would say is the actual outcome if it goes well. Captured before the session starts; revisited at the end.
Interaction cadence, idle gaps, application switches, repeated keystroke patterns. Heuristics that surface candidate stalls — never definitive, always evidence.
When a stall is high-confidence, a single prompt appears. From a peer, a coach, or a configured LLM. Short, declinable, and never the same as the last one.
Pause-aware. The clock stops. Pauses are logged. The reflection at the end can ask why the longest one happened.
Four taps, fifteen seconds: how easy was the outcome, what was the actual result, did the intention hold, what surprised you. The data compounds; the friction does not.
Helm is our two-sided LLM pattern: a human in the loop who can nudge, redirect, or override a model mid-stream. Dojo takes the same primitive and turns it inward. The person being steered is you, working alone. The steerer is a facilitator — sometimes a real human (a coach, a peer); sometimes a tuned LLM running against a brief you wrote when you were not stuck.
The pairing is the differentiator. A timer cannot help you when you stall. A facilitator can. Dojo makes the facilitator cheap enough to be present every session, without making the session feel supervised.
Not surveillance. The signals stay in your runtime. We do not export keystroke streams, screenshots, or application logs anywhere you have not explicitly approved. The point is to help you, not to monitor you.
Not a productivity score. No daily ranking, no leaderboard, no streaks-as-shame. Reflection data is yours; we do not aggregate it into a metric you have to defend.
Not validated as a clinical instrument. The stuckness heuristics are heuristics. They surface candidate stalls; they do not diagnose. Anyone selling clinical-grade stall detection is overreaching. We are not them.
The hard part is not the start. It is the moment, twenty minutes in, when nobody is watching.