Uploading Materials¶
Add documents to your project from the Materials page. Everything you add is chunked, embedded, and indexed automatically — see Concepts → Documents for what happens under the hood.
Supported formats¶
| Category | Formats |
|---|---|
| Documents | PDF, DOCX, PPTX, XLSX, CSV, TXT, HTML |
| Code | Python, JSON, and other text-based source files |
| Media | MP4 (and other video), PNG, JPG/JPEG, SRT |
The per-file size limit is 500 MB.
Videos are transcribed
Videos are transcribed automatically with Whisper. Expect roughly 5–10 minutes of processing for an hour-long lecture; the transcript becomes searchable, citable content.
Upload methods¶
Drag and drop¶
Drag files onto the Materials page (or click Upload). Files are uploaded directly from your browser to object storage via presigned URLs, then queued for ingest.
Web crawl¶
Enter a starting URL and let the crawler discover and ingest linked pages. See the dedicated Web Crawling guide.
Canvas¶
Import files, pages, modules, syllabus, assignments, and discussions from a Canvas course. See Canvas Integration.
PubMed¶
Import articles by PubMed ID or search query — handy for literature-review projects.
GitHub¶
Provide a public repository URL to ingest its contents — great for project-onboarding assistants.
Organizing with document groups¶
Group related documents (e.g. "Lectures", "Homework solutions", "Research papers"). Users can toggle groups on and off in chat settings to scope retrieval, and the Retrieval API accepts a doc_groups filter.
Monitoring ingest¶
Each document on the Materials page shows its ingest status. Large files and crawls process asynchronously; failures are retried automatically with exponential backoff. Duplicate documents are detected by content and skipped — see duplicate handling.
Deleting documents¶
Deleting a document removes it from future retrieval. Citations that already appeared in past conversations remain visible.