Create & Train a Vector
Everything happens on one page: name the vector, add content from any mix of sources, then train.
Create the vector
Sidebar → Vectors → Create Vector. Give it a clear name (at least 3 characters), like Product Help Center.
The vector is only really saved when you press Train Vector at the end — you can safely experiment before that.
Add text content
In 📝 Text Content, paste any free text — product descriptions, policies, notes. One text source per vector.
Add Q&A pairs
In ❓ Question & Answer Pairs, type a Question and its Answer, then click Add Q&A for more pairs. Q&A pairs are great for FAQs because the question wording helps matching.
Q&A pairs are typed by hand here. If you have FAQs in a spreadsheet, upload the file in the File Upload section instead — it will be ingested as a document.
Crawl a website
In 🌐 Website Crawler:
- Enter the Website URL.
- Set Max Depth (1–5, default 2) — how many link-levels deep to follow.
- Click Crawl.
The crawler follows links within the same website, up to 200 pages or 2 minutes, whichever comes first. Found pages appear in the Discovered Links list with a status badge (Pending → Processing → Completed, or Failed). Remove any link you don't want with its delete button.
You can also Add Individual Links manually — including direct links to PDF files.
Upload files
In 📁 File Upload, add files: PDF, DOCX, XLSX/XLS, CSV, TXT, MD. The text is extracted right in your browser and only the text is stored — the original file is not kept. Previously uploaded files are listed below, where you can delete them (applied on the next train).
(Optional) Tune chunking
In ⚙️ Advanced Parameters you can change the Chunking Method, Chunk Size (default 700), and Overlap Size (default 105). The defaults work well for most content — see Advanced Settings before changing them.
Train
Click Train Vector (or Retrain Vector on an existing vector). A progress bar tracks the work; large websites and files can take a few minutes.
Retraining rebuilds everything. Each train deletes the vector's existing chunks and re-ingests all current sources. That's how edits and deletions take effect — but it means a retrain of a big vector takes as long as the first train.
After training
- Each source shows a status. A Failed source (unreachable URL, empty file) doesn't break the rest — fix or remove it and retrain.
- To rename the vector, edit the name inline. To delete it, use the ⋮ → Delete menu on the Vectors page (this permanently removes all its content).
- Your vector is now ready to use — see Use Vectors in Bots & Workflows.