How to Use Knowledge Base in CogniAgent for Product-Aware Agents

Reading time: ~6 minutes
Level: Beginner to intermediate
What you’ll build: An agent that answers customer questions accurately – pricing, services, FAQs – by pulling from your own documents

Why “product-aware” matters

An AI agent that doesn’t know your business is a liability.

Ask a generic LLM about your pricing and it will guess. Ask it about your service area, your cancellation policy, or whether you cover a specific neighbourhood – and it will confabulate a confident, plausible, wrong answer.

The Knowledge Base in CogniAgent solves this. You upload your documents, policies, and FAQs once. Your agents pull from them at runtime, every time – answering based on what your business actually says, not on what the model assumes.

This guide walks you through how to set it up and how to wire it into your workflows.

What the Knowledge Base is (and isn’t)

The Knowledge Base is a standalone section in your CogniAgent account, separate from your Applications (workflows) and Conversational Flows. Think of it as a library your agents can consult – not a node in the workflow itself.

What you can store there:

• Product catalogues and pricing sheets
• Service coverage maps and area rules
• FAQ documents
• Brand guidelines and tone-of-voice references
• Hiring criteria and job descriptions (for recruiting agents)
• Ideal customer profiles (for lead qualification agents)

What it doesn’t do: It doesn’t replace your workflow logic. It doesn’t make decisions. It gives your agent accurate facts to draw from – the agent still decides how and when to use them.

How to upload content to the Knowledge Base

1. Open CogniAgent and navigate to the Knowledge Base section from the main menu.
2. Click Add Knowledge and enter the name.
3. Import your documents – Files, Sync from Website or Import from Google Drive.
4. In the “What to Parse” field, specify which parts of the document matter. Example: “Extract product names, prices, and service descriptions. Ignore headers, footers, and page numbers.” This keeps your agent focused on the relevant content and prevents it from quoting boilerplate back at customers.
5. Save. Your document is now available to reference from any agent in your account.

Tip: Name your documents clearly. When you’re connecting them inside an actor, you’ll select them by name — “Pricing_Sheet_2026” is easier to find than “Document_v3_final.”

Two ways to connect the Knowledge Base to your agents

Once your documents are uploaded, there are two ways to make them available inside a workflow.

Option 1: Use the Search Knowledge node

For more precise, programmatic retrieval – when you need to pull specific information at a defined point in your workflow rather than letting the actor decide when to look – use the Search Knowledge node from the Actions category.

When to use this:

• You need to retrieve a specific document section before the conversation starts (e.g., pull the relevant job description before a screening actor runs)
• You want to log or route based on what was retrieved (e.g., check if a customer’s request matches your service list, then branch accordingly)
• You’re building a workflow where multiple agents each need different knowledge — you can query once, store the result in a variable, and pass it to whichever actor needs it

How to set it up:

1. Add a Search Knowledge node from the Actions tab.
2. Configure the query – what to search for. You can use a natural language.
3. The node returns the retrieved content as output.

Option 2: Search Knowledge Base directly in Conversational Flow 

Open the Conversational Flow tab, go to flow settings, and connect your Knowledge Base in the dedicated tab. The actor doesn’t read the documents upfront – it searches mid-conversation, choosing what to look for based on the document name and your description.

What to store in the Knowledge Base (and how to structure it)

The agent is only as accurate as the document you feed it. A few principles:

Be specific, not exhaustive. A 40-page brand bible uploaded without parsing instructions will produce unfocused results. Break it into smaller, focused documents: one for pricing, one for service area, one for FAQs.

Use the “What to Parse” field. This is the most underused feature. Telling CogniAgent to focus on specific sections (and ignore others) significantly improves retrieval quality.

Store your ideal customer profile (ICP) here. If your agent is qualifying leads, upload your ICP as a Knowledge Base document. The agent can then qualify candidates against your real criteria – not just the fields in the form.

Keep it current. If your pricing changes, update the document. Agents pull from whatever is uploaded – stale documents produce stale answers.

Example: A product-aware support agent for a cleaning company

Goal: A website widget agent that can answer questions about services, pricing, and availability — and book jobs.

Knowledge Base documents:

• Services_List.pdf — what you clean, what you don’t, minimum job size
• Pricing_2026.pdf — hourly rates, flat fees for standard jobs, add-on pricing
• Service_Area.pdf — list of postcodes you cover

Common mistakes to avoid

Connecting the Knowledge Base but not specifying what to parse. The agent will read the whole document and may quote irrelevant sections. Always specify what matters.

Uploading a single giant document. Harder to retrieve from, harder to update. Smaller, focused documents work better.

Using the Knowledge Base for workflow logic. The Knowledge Base stores facts, not decisions. Use Condition and Multi-Condition nodes for branching logic – not a document that says “if the customer asks X, do Y.”

Forgetting to update it. Agents answer based on what’s uploaded. A pricing document from last year will produce last year’s answers.

Summary

The Knowledge Base in CogniAgent is what separates a generic LLM response from an agent that actually knows your business. Upload once. Connect to your actors or use the Search Knowledge node to retrieve programmatically. Keep documents focused and current.

The two connection methods – actor-level attachment for conversational retrieval, and the Search Knowledge node for precise mid-workflow queries — cover most use cases. Start with actor-level attachment for support and FAQ agents. Add Search Knowledge when you need to branch or route based on what the documents say.