Sometimes, the AI needs more than the customer’s media to make the right decision. It may also need to follow your internal rules, product documentation, support procedures, or company policies.
With Knowledge, you can attach reference documents directly to an AI step instead of adding all that information to the prompt.
The AI can then use these documents as additional context when analyzing the customer’s media and generating its result.
What is Knowledge?
Knowledge allows you to attach business-specific documents to an AI analysis step.
You can upload files such as:
PDF documents
Word documents (.docx)
Text files (.txt)
PowerPoint presentations (.ppt, .pptx)
Other supported document formats
The AI can refer to the information in these files while analyzing the customer’s media.
This is useful when the correct result depends on rules or information that cannot be identified from the image, video, or document alone.
How does it work?
When Knowledge is added to an AI step, the AI uses three main sources of information:
1. The customer’s media
The photo, video, document, or other media submitted by the customer.
2. Your prompt
The instructions explaining what the AI should analyze and what result it should return.
3. Your Knowledge files
The reference documents containing the rules or information the AI should consider.
In your prompt, you can ask the AI to base its decision on the information available in the Knowledge files.
Example
Imagine that you want the AI to check whether a damaged product may be covered by your warranty.
You already have a document containing your warranty rules. You can:
Add the warranty document to Knowledge.
Ask the customer to provide photos of the damaged product.
Ask the AI to compare the visible damage with the rules in the document.
For example:
Analyze the customer’s media and determine whether the damage appears to be covered by our warranty. Use the rules available in the Knowledge files and briefly explain your decision.
The AI can then use both the customer’s photos and your warranty rules to generate its answer.
When should you use Knowledge?
Knowledge is useful whenever the AI needs information specific to your business.
Common examples include:
Warranty rules
Determine whether an issue is covered, excluded, or requires further verification.
Returns and refunds
Check whether a customer request meets your return conditions.
Product specifications
Compare the product shown by the customer with your technical documentation.
Damage assessment
Apply your internal criteria to damaged, defective, or worn products.
Operational procedures
Identify the next action based on a support or troubleshooting process.
Eligibility criteria
Check whether a case meets your predefined requirements.
How to add Knowledge to a step
Open the AI step where you want to use your business rules.
Go to the Knowledge section.
Add the relevant reference file or files.
Update your prompt to explain how the AI should use the documents.
Test the step with several realistic examples before using it in production.
Make sure the uploaded documents contain the information the AI needs to make its decision.
Keep the prompt focused
You do not need to copy your full policy or documentation into the prompt.
Use the Knowledge files for the detailed information, and use the prompt to explain:
what the AI should analyze;
which decision it should make;
how it should use the Knowledge files;
what result it should return.
Knowledge + Prompt + Media
A simple way to understand the setup is:
Media → What can the AI see?
Knowledge → What are your company’s rules?
Prompt → How should the AI use both?
Combining these three elements allows you to create AI analysis steps that follow your business rules while keeping your prompts clear and easy to maintain.
