research group
better ways for people and agents to work together.
Every tool became multiplayer. Then AI put each person back in a private room.
A private chat window.
Insights are born and die in one person's session. Every handoff to a teammate loses context and repeats the work.
Agents on laptops.
An agent with one person's access, acting alone, is an infrastructure problem. Nobody else can reach it, and it inherits everything that person can see.
One provider, forever.
The tool you work in decides which model you use. The best model changes every few weeks, and most tasks do not need it.
Agents belong in the workstream, with the whole team in the room.
One session a whole team can see, correct and teach, in the places work already happens. What one person teaches becomes a rule for everyone. Permissions follow the people in the room. The context belongs to the company, and the model is a supplier, chosen per task.
any of them, on your keys. each task goes to the model that does it well, which is rarely the most expensive one: frontier where it matters, small and open where it does not. the context stays yours when you switch.
Two products, one bet.
What a company keeps for its agents. Skills, knowledge and metrics, with a scope on every object and a version on every change. Any agent reads it.
skills · knowledge · metrics · any agent
maiaYour team's AI coworker, in Slack. One per channel. Anyone in the thread can ask it, correct it or teach it; it asks before it acts.
one per channel · any model · your keys · approvals in the thread