Knowledge base
A knowledge base is a curated collection of documents and structured information that an AI phone agent can search during a call to find relevant, business-specific facts.
How a knowledge base works
A knowledge base sits outside the agent's main instructions. Instead of placing every policy, product detail, and procedure in one long prompt, a team uploads or connects the material the agent may need. When a caller asks a question, a search process retrieves the most relevant passages and gives them to the agent as context for its answer.
This pattern is useful for information that is too detailed or changes too often to keep in the main prompt. Examples include service descriptions, office policies, troubleshooting steps, eligibility rules, and answers to frequently asked questions. The prompt still defines the agent's role and behavior; the knowledge base supplies reference material for particular questions.
Retrieval quality depends on the source material. Clear headings, focused documents, consistent terminology, and removal of outdated or conflicting versions make relevant passages easier to find. Access should also be scoped carefully so an agent can search only the information appropriate for its role.
Why it matters for AI phone calls
Callers expect a direct answer without waiting for an employee to search several systems. A knowledge base can make detailed information available during the conversation while keeping the core prompt manageable. It also gives operators a specific body of content to review when an answer is incomplete or unsupported.
A knowledge base does not guarantee a correct answer. The requested fact may be absent, stale, poorly written, or missed by retrieval. A well-designed agent should recognize uncertainty, ask a clarifying question when appropriate, and transfer or create a follow-up path when the source material does not support a reliable response.
In practice on ThunderPhone
ThunderPhone provides an organization document library for text, Markdown, CSV, PDF, and DOCX uploads, with processing status and content search. Documents are selected per agent, and a built-in knowledge search tool can retrieve material during a call. Uploads do not apply to every agent automatically. Current documented limits are 5 MB per text file and 50 MB per PDF or DOCX file.