Deploying Enterprise Knowledge to Voice Agents

When enterprises adopt Voice Agents, it is essential to build a reliable knowledge base so that the agent accurately reflects the company's specific policies or product details. The knowledge base stores documents, policies, technical references, and product specifications, helping the agent generate grounded answers instead of relying on general model knowledge.

The ElevenLabs platform offers two modes for utilizing knowledge.

• Direct inclusion in context: Suitable for small-scale knowledge bases, this injects content directly into the model's context window to minimize latency.
• RAG (Retrieval-Augmented Generation): When the knowledge base is too large to fit into the context, only the sections most relevant to the user's query are retrieved.

For an effective knowledge base, the Curation stage before implementation is important. Outdated drafts or unnecessary materials should be removed, and documents should be structured by domain, such as HR policies, product documentation, and technical manuals. Focusing on quality over quantity to secure accurate and complete data is key to reducing the agent's Hallucination phenomenon.

Implementation methods vary depending on scale. The simplest form, Single-Agent knowledge base, is suitable for focused use cases such as a specific product line or HR policies, enabling the fastest value creation. However, as scale grows, performance degradation can occur, which may require a more complex architecture.