Turn a topic into a real, verified course.
Name a topic. Lunaris plans it, writes the lessons, and verifies every claim against evidence. You watch each step run.
“A sorted range lets you discard half the remaining candidates with a single comparison.”
One canvas, four moves. You name it; the agent does the rest in the open, and you read the result.
Every topic is decomposed into concepts and assembled into a directed acyclic graph, ordered topologically in deterministic code. The model proposes; the graph builder proves.
A claim-level verifier checks every factual sentence against retrieved evidence. What survives carries its source, a trust tier, and a credibility score. What doesn't is cut before publish.
Merrill-structured lessons, objectives, and claims that carry their sources. This is the Reader, on the course this page has been building.
Recall how you find a name in a paper phone book. You never scan page by page; you open somewhere sensible and decide which half to keep. That instinct is the whole algorithm.
“A sorted range lets you discard half the remaining candidates with a single comparison.”
Probe the middle. If the value is too small, the answer lives in the right half; too large, the left. Either way, half the candidates are gone.
Trace a search for in [4, 8, 15, 16, 23, 42]. Which index does the second probe hit, and why?
Key takeaway
Open the studio and name a topic, or run the whole stack yourself with one command.
No keys? Lunaris falls back to a deterministic build and says so.