Features
RAG knowledge base from URLs and file uploads (PDF, TXT, CSV, Markdown, DOCX)
Multi-model LLM agent supporting OpenAI, DeepSeek, Anthropic, Gemini, Jina, SiliconFlow
Embeddable multilingual chat widget with session persistence
SSE streaming responses for real-time answer generation
Role-based access control (super admin, admin, agent)
SSRF protection and widget origin whitelisting
Vector search using Qdrant or pgvector
Docker Compose deployment with FastAPI, Next.js, Redis, nginx
Admin console for managing knowledge, sessions, users, and site config
Self-hosted with no GPU required (inference via external APIs)
API for custom integration
Supports SQLite or PostgreSQL database
Session persistence across page reloads
No-code visual workflow builder
Omnichannel deployment (web, phone, mobile)
Low-latency voice (500ms)
High throughput (300k messages/min)
Integrated prototyping and NLU modeling
Entry conditions for playbooks/workflows
Required variables for workflows
Organization-level usage breakdown
Voiceflow Core optimized model (May 2026)
Transcript filtering by workflow and tool
