Skip to main content

Creating Your First Agent

Every agent in Agntlabs.ai is trained on your own data. This page explains how to create an agent and feed it the knowledge it needs to answer your customers’ questions.

Create an Agent

  1. Log in to your dashboard
  2. Click Create Agent
  3. Enter a name for your agent — this appears in the chat widget header
  4. Proceed to add your knowledge sources

Adding Sources

Your agent’s quality depends on the sources you provide. You can use any combination of the three source types below.

Files

Upload PDF, DOCX, or TXT files. Great for product manuals, help docs, policy documents, and FAQs.

Website URLs

Paste a URL and the agent crawls the page content. Useful for pulling in existing help center or marketing pages.

Q&A Pairs

Add specific question-answer pairs. Gives you precise control over how your agent responds to common questions.

File Uploads

Supported formats: PDF, DOCX, TXT File size limits depend on your plan:
PlanMax File Size
Free5 MB
Standard50 MB
Pro100 MB
Elite200 MB
Break large documents into smaller, focused files. An agent trained on 5 targeted documents often performs better than one trained on a single 50-page manual.

Website URLs

Paste any publicly accessible URL. The agent extracts and indexes the text content from the page.
  • One URL per entry
  • The agent reads the page content at the time of submission
  • To update, remove the old URL and add it again

Q&A Pairs

Q&A pairs let you define exact answers for specific questions. This is the most precise way to control your agent’s responses.
  • Write the question as a customer would ask it
  • Write the answer exactly as you want the agent to respond
  • Q&A pairs take priority over general document content when a match is found
Q&A pair limits vary by plan: Free gets 1,000 pairs, Standard gets 1,500, Pro gets 3,000, and Elite gets 5,000.

What Happens When You Create an Agent

When you click Create Agent, the system:
  1. Processes all your uploaded files and extracts the text content
  2. Splits the content into smaller chunks for efficient retrieval
  3. Creates vector embeddings of each chunk and stores them in your agent’s knowledge base
  4. Your agent is ready to answer questions based on this knowledge
This process usually takes a few seconds to a couple of minutes depending on the amount of content.

Tips for Good Training Data

  • Be specific — Detailed, well-written content produces better answers than vague bullet points
  • Cover common questions — Look at your support inbox and add content that addresses the most frequent topics
  • Use Q&A pairs for precision — If a specific question needs an exact answer, add it as a Q&A pair rather than relying on document retrieval
  • Keep content current — Remove outdated sources and add new ones as your product or service evolves
  • Test after adding sources — Use the test page to verify your agent handles key questions correctly