tensr is agent-native, and open by construction.
Your training data is a SQLite file on your device and your routines are plain
workout.md. Nothing here is locked behind an export tax. These docs
exist so you — and the tools and agents you control — can read, write, and build
on your own data.
The routine format: human-readable, machine-readable, printable, and useful by itself.
Data model →The SQLite tables your export contains, and how to query your own history.
Link format →How a shared session or routine rides entirely in a URL fragment — no server.
Agent skills →Teach your own model (Qwen, Kimi, MiniMax, Claude…) to coach you from your data.
Why open
Holding data is the cheap part; storage is pennies. The value is in acting on it — and in the agent era, an assistant is only as good as the data it can reach. So tensr keeps the canonical copy on your device, in formats any tool can read, and treats agent access as the default posture rather than a paid tier.
If you build something that reads tensr data — a coach, a visualizer, an importer — we want to hear about it. That's the point of publishing these.
The one principle
Your file, your history, your agent's access to all of it — free and total. tensr's server is a spotter, not a coach: it assists on request, holds no canonical copy, and you can walk away with everything.