Demo · Cursor (or any MCP agent) + Cosmos
Learning that remembers what you miss
Same certification topic, session after session
An agent that teaches from your graph and schedules review when you are liable to forget.
New here?
Cosmos is a personal memory graph from Polarity Lab. You connect apps you already use. Any AI agent can read it through MCP before it replies.
Alex is studying for a professional certification with a deadline in a few months.
Proves
Path, query, capture, and recall across sessions
Terms in this demo
- Path
- Where you said you are headed. Cosmos reads it before giving advice.
- Briefing
- What Cosmos surfaces at the start of a session: goals, drift, and what is due.
- Connectors
- Apps that suggest moments from your week. You keep, edit, or dismiss each one.
- Week card
- A short Sunday compile of the moments that counted this week.
- Drift
- The gap between what you declared and what your week actually showed.
- Recall
- Something you learned that is due for review before you forget.
01 · Alex declares a goal
In Cosmos, Alex sets a path: pass the certification by spring.
Alex opens Cursor with Cosmos connected. The briefing returns three things: the current study module, one review item due today, and a misconception Alex got wrong last week.
02 · The lesson uses Alex's graph
Alex asks for help on a topic in the current module.
The agent searches the graph: last week's wrong answer, Alex's preference for mechanism-first explanations, and a real example Alex captured from work.
The reply references those nodes instead of starting from zero.
03 · A miss becomes durable
Alex confuses two steps. The agent captures the correction as a memory linked to the module.
It does not disappear in chat scrollback.
04 · Review is scheduled on Alex's curve
Cosmos schedules the correction for review in two days based on Alex's forget pattern, not a generic reminder.
The review queue (Horizons) shows what is due before it fades.
05 · Without Cosmos
Same model and a detailed system prompt with the syllabus in the thread.
The one-session answer can be good.
Next week there is no link to the misconception, no scheduled review, and the next chat starts cold.
Closing
The difference is not a smarter single reply. It is a trajectory that remembers what to teach and when you are liable to forget.