DISCOVER THE FUTURE OF AI AGENTSarrow_forward

SmartCall-Agent

calendar_todayAdded Feb 25, 2026
categoryAgent & Tooling
codeOpen Source
PythonWorkflow Automation大语言模型FastAPIMultimodalRAGAI AgentsAgent FrameworkWeb ApplicationAgent & ToolingModel & Inference FrameworkAutomation, Workflow & RPAKnowledge Management, Retrieval & RAGEnterprise Applications & Office

A modular voice AI platform based on LiveKit and OpenAI Realtime API, integrating RAG knowledge retrieval, JWT authentication, and MongoDB persistence for real-time outbound calling and domain-specific conversations.

Overview#

SmartCall-Agent is a modular voice AI platform built with Python (FastAPI). It integrates the LiveKit Agents framework with the OpenAI Realtime API to initiate real phone calls via SIP Trunk and conduct low-latency voice conversations.

Core Features#

RAG (Retrieval-Augmented Generation)#

  • Uses Pinecone vector database for document embeddings storage
  • Generates embeddings via OpenAI text-embedding-3-small
  • Supports automatic extraction and chunking of PDF, TXT, DOCX, and web URLs (1000 chars/chunk, 200 chars overlap)
  • Namespace isolation (agent_{agent_id}) for multi-agent knowledge separation

Outbound Calling#

  • Initiate outbound calls via LiveKit SIP Trunk
  • Real-time audio stream processing
  • Voicemail detection and human transfer support

Voice AI#

  • Integrates OpenAI Realtime API for real-time voice conversations
  • TTS/ASR support via OpenAI voice models
  • Configurable voice, tone, temperature parameters

User & Permission Management#

  • JWT authentication (HS256 algorithm)
  • bcrypt password hashing
  • User registration/login/logout APIs

Data Persistence#

  • MongoDB for Agent, User, CallLog, Document metadata storage
  • Pinecone for vector embedding storage

Observability#

  • Structured logging
  • Health check endpoint /health
  • Analytics dashboard endpoint /api/analytics/dashboard

Architecture Components#

ComponentFileResponsibility
FastAPI Backendmain.pyRESTful API, user auth, Agent CRUD, document vectorization, call initiation
LiveKit Agent Workerworker.pyIndependent process running KnowledgeAgent, handling real-time voice streams and tool calls
MongoDB-Structured data storage (User, Agent, CallLog)
Pinecone-Vector data storage (Document Chunks)

Key API Endpoints#

PathMethodFunction
/healthGETHealth check
/api/registerPOSTUser registration
/api/loginPOSTUser login (returns JWT)
/api/agentPOSTCreate Agent (with RAG docs)
/api/agent/{agent_id}GET/PATCH/DELETEGet/Update/Delete Agent
/api/agentsGETList all Agents
/api/agent/{agent_id}/queryGETQuery Agent knowledge base
/api/callPOSTInitiate outbound call
/api/callsGETGet call records
/api/analytics/dashboardGETDashboard metrics

Deployment Requirements#

  • Python 3.10+
  • OpenAI API Key
  • Pinecone API Key
  • LiveKit Cloud account with SIP Trunk
  • MongoDB instance

Use Cases#

  • Automated customer support callbacks
  • Telemarketing campaign execution
  • Appointment reminders and notifications
  • Voice Q&A systems based on private knowledge bases

Related Projects

View All arrow_forward

STAY UPDATED

Get the latest AI tools and trends delivered straight to your inbox. No spam, just intelligence.

rocket_launch