Uploading Content

The most direct way to train your AI agent is to paste or upload text content into the knowledge base. This is done in Step 4 (Knowledge Base) of the agent setup wizard, or from the knowledge management section in your My Agents dashboard.

How to Add Content

1. Navigate to the Knowledge Base tab in the setup wizard (or edit your agent from My Agents)
2. Select the Upload / Paste method
3. Enter a source name -- this labels the content for easy management later (e.g., "Company FAQ", "Refund Policy")
4. Paste your text content into the editor, or upload a text file
5. Click Save -- the system will automatically chunk, embed, and store your content

What Happens Behind the Scenes

When you save content, the system processes it in three steps:

1. Chunking -- Your text is split into segments of approximately 2,000 characters each. The system splits on paragraph boundaries (double newlines) to preserve natural content groupings. Each chunk overlaps with the next by about 200 characters so that information spanning a paragraph break is not lost.

2. Embedding -- Each chunk is converted into a 1,536-dimension vector using OpenAI's text-embedding-3-small model. This vector captures the semantic meaning of the text.

3. Storage -- The chunk text and its embedding vector are stored in the knowledge_chunks table in Supabase, associated with your agent instance.

Supported Content Types

The knowledge base works with any plain text. The most effective content includes:

• FAQs -- Question-and-answer pairs about your business
• Product documentation -- Feature descriptions, specifications, how-to guides
• Support policies -- Refund policies, warranty terms, SLA details
• Pricing information -- Plan descriptions, feature comparisons, add-on costs
• Company information -- Business hours, contact details, location, team bios
• Troubleshooting guides -- Step-by-step procedures for common issues

Source Naming

Every piece of content you upload gets a source name. Choose descriptive names so you can easily identify and manage content later:

You can add multiple sources to the same agent. For example, a Customer Support agent might have separate sources for "FAQ", "Pricing", "Returns Policy", and "Product Manual".

Best Practices

Structure content as Q&A pairs. The RAG pipeline performs best when your content contains explicit questions followed by clear answers. For example:

Q: What are your business hours?
A: We are open Monday through Friday, 9:00 AM to 6:00 PM IST.
Weekend support is available via email at support@example.com.

Q: What is your return policy?
A: You can return any unused product within 30 days of purchase
for a full refund. Used products can be returned within 7 days
for store credit. Contact us to initiate a return.

Include pricing details. Callers frequently ask about pricing. Include specific numbers, plan names, and what is included in each plan.

Keep answers concise. Aim for 2-4 sentences per answer. The agent reads these chunks aloud during a call -- overly long passages lead to unnatural responses.

Update regularly. Remove outdated content and add new sources when your policies, pricing, or product details change. See Managing Sources for instructions.

Avoid duplicate content. If the same information appears in multiple sources, the agent may retrieve redundant chunks, wasting context window space. Keep each source focused on a distinct topic.

Content Limits

There is no hard limit on the number of chunks or sources per agent. However, keep in mind:

• Each chunk is approximately 2,000 characters (~500 tokens)
• During a call, the agent retrieves the top 5 matching chunks
• Very large knowledge bases (1,000+ chunks) still work well because the vector search efficiently finds the most relevant content

A Product by BRTNeura Technology LLP