Sem
Sem's suggested questions, its questions to you, how it asks before changing your data, what it remembers, and how it handles long conversations.
Sem is the assistant in Semestra. It needs an AI provider with a model for the Sem purpose.
Conversations
Sem's page lists your conversations in a column beside the open one, pinned first and then by day, with a search box and New conversation at the top. Rename, pin or delete a conversation from its ⋯ menu. On a narrow window the list moves under the History button.
Suggested questions
Above the input of a new conversation, Sem lists a few questions you can click, drawn from your data: assessments and tasks that are due soon or overdue, classes still to come today, courses with no assessments yet, and courses with some grades in. A course a question names is sent to Sem as a reference.
With a Light tasks model set up, Semestra has it reword these questions, also taking your recent conversation titles into account, at most twice a day per program. Turn off Personalized suggestions in global settings → AI → Sem → Conversations to keep the fixed wording and make no model calls.
Questions from Sem
When Sem needs your choice to go on, for example which of two courses you mean or which evenings suit you for studying, it asks in place of the input box: a question with a few options, one of them sometimes marked as recommended. Pick an option (or press its number), type your own answer, or skip the question.
Approving changes
Sem reads freely. Before it creates or changes something, it shows the change in place of the input box, and nothing happens until you approve it. You can also edit the values first, or type what Sem should do instead. When several changes wait, move between them or approve them all. Global settings → AI → Sem → Approvals sets which changes Sem may make without asking: Ask every time, By module, or Auto-approve all. Deleting always needs your approval.
Memory
Sem remembers lasting facts about you, how you want it to work, your plans, and where to find things. Review and edit them in global settings → AI → Sem → Memory. A Temporary chat neither reads nor saves memories.
Long conversations
When the conversation nearly fills the model's context, Sem compacts it before answering: it
summarizes earlier messages and keeps the latest ones word for word. To do it yourself, type
/compact, optionally followed by what the summary must keep. While it runs you can't send a new
message.