How Search Works
Eventually the question stops being "can this help me write the unit plan" and becomes "where did I put the unit plan." Here's what you can search, what your agent can search, and how it works underneath.
What you can search
Search in the sidebar covers everything at once, in one ranked list. Most list pages also keep their own box for searching just that kind of thing.
| Where | What it searches |
|---|---|
| Conversations | Titles, the messages themselves, notes, instructions, subjects, tags |
| Documents | Titles, descriptions, body text, spreadsheet cells |
| Files | Filename, your description, the text inside parsed PDFs |
| Folders | Names, descriptions, notes, instructions |
| Memories | Titles and bodies |
| Skills | Titles, descriptions, tags, and the instructions themselves |
Worth knowing:
- Conversation search reaches into messages. If you remember what an agent told you but not which chat it was in, search the phrase itself.
- Quotes,
or, and-all work."exit ticket"for an exact phrase,rubric or checklistto widen,assessment -summativeto exclude. - Titles rank highest. A word in a title beats the same word in a body, so descriptions and notes are worth filling in — but the title is what pulls something to the top.
- Narrow it down. Filter by type, by created or updated date, by yours vs. shared with you, or to starred items only; sort by relevance, date, or title. All of it lives in the URL, so a search can be bookmarked or sent to someone.
What your agent can search
Ask "haven't I already written a rubric for something like this?" and the agent goes looking while you stay in the conversation, then answers with a link to what it found.
An agent can search:
- Your documents — and it finds more than the search box does, for the reason below
- Your conversations
- Your files and folders
- Your memories
- TeamTeacher's guides and documentation
- Curriculum knowledge bases
- The web
Words vs. concepts
Your documents get indexed two ways.
By word. Every term you wrote, reduced to its root, so "assessing" and "assessment" match each other. This finds what you literally typed. The technical name for it is a sparse vector — a very long list with a slot for every possible word, where only the ones you actually used light up.
By concept. The same text also goes through a model that turns it into a list of numbers describing what it's about — a dense vector. Two texts on the same idea land near each other even with no words in common, so "how do I grade group work fairly" can find a document you called Collaborative project rubric.
Neither works on its own. Words miss paraphrases. Concepts miss exact labels — a code like RL.6.2 doesn't mean much by itself, so it sinks underneath conceptually similar standards. So both searches run, and the results are combined into one ranking.
Search matches words. Agents can also match concepts. That's why asking often turns up something the box didn't.
Knowledge bases
Curriculum collections — Common Core, the IB programmes, NGSS, state standards, UDL — are TeamTeacher's material, not yours. Agents search them both ways too, and can narrow to a subject, grade band, or framework first, so a high-school question doesn't come back with elementary standards.
For what's in each one, see Getting the Most Out of Knowledge Bases.
What search won't do
- Of your own material, only documents and memories can be searched by concept, and only when an agent does the searching. Everything else of yours matches on words — so search the phrase you remember, not the idea.
- A word buried in a filename won't match. Searching
fractionsdoesn't findhandout-fractions.pdf, because the whole filename reads as a single word. Renaming the file, or giving it a description, makes it findable.
Results can include things shared with you — a conversation someone added you to, a skill your school shared — and those are labelled Shared with you. Nothing you write goes into the curriculum knowledge bases.