What it is
Chair is an accountability agent. Its whole argument is one line on the site: deadlines you set yourself are the easiest ones to move.
It messages you on WhatsApp every morning and asks what you're committing to. That evening it asks whether it happened. Every promise goes into a permanent ledger, so it can tell you that this is the third partial week in a row on the same task.
A few things make it more than a reminder app:
- It tracks patterns, not just tasks. It notices which work you keep sliding and where you get stuck repeatedly.
- Trust adjusts. Miss often and it watches more closely; deliver and it backs off.
- You can talk instead of type — brief it with a voice note rather than writing.
It isn't open yet. The site is collecting a waitlist.
Why I built it
I struggle to keep up with my extracurriculars. That's the honest version — this isn't a tool I designed for someone else and then went looking for users for. I kept committing to things and then quietly not doing them, and nothing around me was set up to notice.
A reminder app doesn't fix that. You can dismiss a notification without consequence and without anyone ever knowing you dismissed it. What changes behaviour is something that asks, keeps your answer, and can point at the pattern later.
How it works
The idea underneath it is simple: you are far more likely to do something if you know you'll be asked about it.
So an AI agent does the asking. It checks in every morning about what you're taking on, checks back in the evening about whether it happened, and writes both down. None of it is passive — you can't let it slide past you the way you let a notification slide, because it's a conversation and it's waiting for an answer.
[Needs Soham's input] The stack is still missing here — what runs the agent, which API carries the WhatsApp messages, which model decides what to say, and where the ledger lives. That also fills the
techlist in this file's frontmatter, currently empty.
What I learned
How to work with AI agents, which turns out to be a different skill from prompting a model.
A chat window forgets you the moment you close it. An agent has to hold state across days, decide when to speak without being asked, and stay useful without becoming something you mute. Getting that balance wrong in either direction makes the whole thing worthless — too quiet and it may as well not exist, too loud and you turn it off in a week.