An assistant with knowledge accesses stored files that have been vectorized into text chunks. It doesn't engage in complex rule-based reasoning or make autonomous decisions. Instead, it searches for relevant text passages and formulates an answer.

This is particularly suitable for structured knowledge repositories where clearly defined information needs to be looked up.

10 Typical Use Cases

1. HR Rule Lookup Assistant

Accesses employee handbooks, company agreements, and travel expense policies.

Example:

"How many days of special leave are granted for relocation?"

2. Data Privacy FAQ Assistant

Searches internal data privacy policies, TOM documents, or data processing agreements.

Example:

"Are we allowed to store customer data in Tool X?"

3. Product Documentation Assistant

Accesses service descriptions, feature lists, and release notes.

Example:

"Does our product support SSO via Azure AD?"

4. Contract Template Assistant

Searches standard contracts, terms and conditions, and framework agreements.

Example:

"What is the notice period stipulated in the standard customer contract?"

5. Price List & Quote Assistant

Accesses current price lists and discount rules.

Example:

"What is the tiered pricing for 50 or more licenses?"

6. Support Knowledge Base Assistant

Searches typical support cases and troubleshooting guides.

Example: An email with "Login not working" is analyzed, and relevant instructions are found.

7. Process Manual Assistant

Accesses internal process descriptions.

Example:

"How does invoice approval work?"

8. ISO/Certification Document Assistant

Searches the quality management manual, work instructions, and audit documents.

Example:

"Where is the access control process described?"

9. Project Archive Assistant

Accesses project logs, specifications, or requirements documents.

Example:

"What was the scope of Project Müller 2024?"

10. Training & Onboarding Assistant

Searches internal training materials and guides.

Example:

"How do I set up VPN access?"

When Knowledge Works Well

Knowledge in assistants is suitable for:

Clear fact-based questions
Text-based documents
Unambiguous rule sets
Stable content that doesn't constantly change
Lookup scenarios ("Where can I find that?")

If the work is done properly (good document quality, clear structure, sound chunking strategy), the risk of hallucinations is low.

Alternative: Using Knowledge Directly in the Prompt

Knowledge is useful when looking up specific information.

However, if a document needs to be fully understood, analyzed, or processed in its entirety (e.g., revised, summarized, evaluated, or rewritten), it is often better to enter the relevant content directly into the prompt in the assistant's form (rule knowledge) or into the chat during the conversation (case knowledge).

This requires that the content does not exceed the model's context window. With current models, this limit is typically several hundred pages of text.