Training Your Agent

Open Dashboard → Agents → your agent → Training Data to manage what your agent knows. There are four tabs for adding sources, plus a shared list showing every source and its processing status.

Files

Drag and drop files, or click to browse. Supported formats: .txt, .pdf, .md, .docx, up to 20 MB per file. Uploads are disabled once you hit your plan's total training-data cap.

Website

Add a source URL and choose a crawl mode:

• Full site — follows internal links from the starting URL.
• Sitemap — reads an XML sitemap and crawls the URLs it lists.
• Individual URLs — crawls only the specific pages you list.

Advanced options let you include or exclude specific paths. Each source shows crawl status, page counts, and size in the source list; you can expand it to see individual links grouped as trained, not found, or excluded, and bulk include/exclude links from there. Sources can be recrawled or deleted from the same list.

You can also quick-import a single URL without going through the full crawl flow.

Text

Paste content directly with an optional title — useful for policies, FAQs, or any text you don't have as a file.

Q&A

Write a manual question-and-answer pair: a title, one or more question variations (one per line, so the agent recognizes different phrasings of the same question), and the answer.

Processing sources

Adding a source doesn't train the agent immediately — it has to be processed first. From the source list you can process an individual document, or click Train Agent (N) to process every unprocessed source at once. Processing chunks the content and generates embeddings for retrieval. You can also edit a document's text after upload (this regenerates its embeddings) or delete it entirely.

Crawl and processing jobs show live status while they run.

Plan limits

Uploads and crawls are disabled once an agent's training data hits its plan cap. See Plans & Limits Reference for the complete table.