Point11's agent platform deploys conversational AI agents that engage website visitors, qualify leads, answer product questions, and book meetings. Agents operate 24/7 across chat and voice channels with sub-200ms response times.
Architecture
The agent platform consists of four layers:
1. Input Layer
Handles incoming signals from chat widgets, voice calls, or API integrations. For voice, this means real-time audio streaming via WebSocket connections. For chat, it handles text input with typing indicators and message queuing.
2. Understanding Layer
Converts input into intent and context:
- Speech-to-Text (STT): for voice: Streaming ASR processes speech as the user talks, rather than waiting for end-of-utterance. This cuts transcription delay to 100-200ms.
- Natural Language Understanding (NLU): for both channels: Identifies user intent, extracts entities (product names, dates, quantities), and determines conversation state.
3. Intelligence Layer
The core reasoning engine:
- Retrieval-Augmented Generation (RAG): The agent retrieves relevant information from your knowledge base before generating a response. This grounds responses in your actual product data, policies, and documentation rather than the LLM's training data.
- Conversation Management: Maintains multi-turn context, tracks conversation goals, and manages state transitions (greeting, qualification, recommendation, booking).
- Tool Use: The agent can invoke external tools — check inventory, look up pricing, query CRM records, or create calendar bookings — during the conversation.
4. Output Layer
Delivers the response:
- Text-to-Speech (TTS): for voice: Generates natural speech from the LLM's text response. ElevenLabs Flash v2.5 achieves approximately 75ms inference latency.
- Message Formatting: for chat: Structures responses with rich elements (product cards, buttons, links, images).
Key Capabilities
- Proactive Engagement: Agents initiate conversations based on visitor behavior (time on page, scroll depth, exit intent).
- Product Knowledge: Grounded in your actual product catalog, pricing, and documentation via RAG.
- Lead Qualification: Asks qualifying questions, scores leads against your ICP criteria, and routes qualified leads to the right sales rep.
- Real-Time Routing: Hands off to human agents when needed, with full conversation context preserved.
- Multilingual Support: Automatic language detection allows the agent to respond in the visitor's language.
Deployment Options
- Chat Widget: Embedded on your website with customizable branding.
- Voice (Inbound): Connected to your phone system for inbound sales and support calls.
- Voice (Outbound): Automated outbound sequences for appointment confirmation, follow-up, and outreach.
- API: Headless deployment for custom integrations.
Sources
- ElevenLabs Conversational AI Overview: https://elevenlabs.io/docs/conversational-ai/overview
- AWS — What is RAG: https://aws.amazon.com/what-is/retrieval-augmented-generation/
- AssemblyAI — Voice AI Stack 2026: https://www.assemblyai.com/blog/the-voice-ai-stack-for-building-agents