ElevenAgents Details Voice Agent Capabilities: Memory, RAG, and Compliance

ElevenAgents defines voice agents as real-time systems that go beyond traditional IVR by integrating live speech recognition, reasoning, and context management. The platform emphasizes six core capabilities: memory, learning, data handling, difficult call resolution, personalization, and measurable performance.

Memory and Context Management

The platform does not have built-in cross-call memory by default. Instead, it relies on three components: dynamic variables for injecting CRM data, start-call webhooks for retrieving user history, and post-call webhooks for recording outcomes. Within a single call, the system automatically maintains full conversation context. For multi-channel interactions, it retrieves background knowledge from CRM or databases to personalize subsequent conversations.

Knowledge Base and RAG

Agents learn via Retrieval-Augmented Generation (RAG) rather than traditional model training. Supported file formats include PDF, DOCX, TXT, MD, HTML, and EPUB, with a maximum size of 20MB per file. The system automatically selects the processing path:

• Full context: If the extracted text is under 250,000 characters, the entire document is included in the prompt.
• RAG: If the text exceeds 250,000 characters, the system indexes it and retrieves only relevant passages.

Compliance and Data Privacy

ElevenAgents offers granular privacy settings, including retention periods, audio storage toggles, and conversation history masking. Default retention for transcripts and recordings is 2 years, but this can be customized from 0 days to unlimited. For HIPAA compliance, a minimum 6-year retention is recommended, and Zero Retention Mode must be enabled for PHI processing, which immediately deletes request/response data and requires storing records via webhooks on the customer's own infrastructure.

Performance and Customer Satisfaction

Key metrics include CSAT, containment rate (resolution without human intervention), and first-call resolution. The platform warns that while high containment rates are desirable, a drop in CSAT alongside high containment may indicate issues are being resolved too quickly without adequate quality. Case studies, such as Admiral, prioritize first-contact resolution over containment, viewing human handoffs as a positive outcome rather than a failure.