How Agents Use Knowledge Bases
- An agent can consume a knowledge base in various ways:
Context retrieval: when a user’s question is asked, the agent searches the knowledge base for relevant excerpts. - Integration with the model: the retrieved content is injected into the AI model’s prompt.
- Evidence-based response: the model generates the output by combining natural language with data from the base.
- Multimodal support: depending on the configuration (multimodal: true), the base can contain not only text but also images, audio, and video.
Types of Knowledge Bases:
- file: documents such as PDFs, spreadsheets, or presentations.
- url: links to internal or external websites.
- text: direct blocks of structured knowledge.
Best Practices
- Use tags to organize and allow the agent to perform quick searches (e.g.: [“finance”, “contracts”]).
- Security: restrict access by sectors to prevent leakage of sensitive information.
- Selective enabling: control the enabled status to activate/deactivate bases depending on the environment or project.
