Academy
1. What Is a Knowledge Base and Why It Matters
Understand what a knowledge base is and how it enables the AI to ground its answers in your own documents across the platform.
A knowledge base is a private, indexed collection of your own documents. When you upload files to a knowledge base, the platform splits them into smaller passages (called "chunks"), indexes them using an embedding model, and makes them searchable by the AI. From that point on, the AI can retrieve the most relevant passages from your documents and use them to ground its answers — meaning its responses are based on your actual content rather than general training data. This grounding capability works everywhere on the platform: in ChatPro conversations, OUPI Studio agents, Deep Research, and workflow steps.
Why does grounding matter? Without a knowledge base, the AI can only rely on its general training. It doesn't know your internal procedures, your product specs, or your company policies. By creating a knowledge base, you give the AI a curated, trustworthy source of truth. When it answers, it cites the retrieved passages so you can verify where the information comes from. This dramatically reduces the risk of inaccurate or fabricated answers and makes the AI genuinely useful for domain-specific work.
Behind the scenes, your documents go through a process called chunking. Each document is split into pieces of a configurable size, with a small overlap between consecutive chunks so that no idea is accidentally cut in half. The default chunk size and overlap work well for most documents. You only need to adjust them in special cases — for example, increasing chunk size for long structured reports, or reducing it for FAQ-style content with many short entries. Once chunked, each piece is indexed and becomes instantly searchable.
You can populate a knowledge base in two ways. The most common is to drop files directly into the upload zone on the knowledge base page — processing runs in the background and a progress bar tracks each document. Alternatively, if your content lives in a connected service, you can use MCP import mode at creation time: the platform pulls content from that integration and keeps the knowledge base automatically synced as the source changes. Both methods result in the same searchable, indexed collection.
Keep one knowledge base per topic or project. Retrieval is much sharper when the AI searches a focused collection rather than a catch-all base containing everything. For example, create separate bases for "HR Policies", "Product Documentation", and "Sales Playbook".
Use the built-in search field on each knowledge base page to test retrieval quality before using the base in a conversation or agent. Type real questions your users would ask — not just keywords — and inspect the returned passages. If results are off, add more relevant documents or revisit your chunking settings.
Go to the Knowledge Bases section and click "New Knowledge Base". Give it a name and description, keep the default chunk settings, and upload one or two documents you work with daily. Once processing completes, use the search field on the base page to test a question against your content.
A knowledge base is your private, indexed document collection that lets the AI ground its answers in your own content. Documents are chunked, indexed, and made searchable across ChatPro, OUPI Studio, Deep Research, and workflows. You can upload files directly or sync from a connected service. Keep bases focused by topic, test retrieval with real questions, and share bases with your team so everyone benefits from the same trusted knowledge.