Conversational
AI Platform
A full multi-tenant SaaS platform for deploying AI voice and chat agents. Twilio telephony piped into OpenAI's Realtime API, embeddable widget with voice and chat channels, knowledge base RAG, Zapier/MCP tool execution, and 5-tier billing — all production and handling real conversations.
Hi there How can we help?
Talk to us now
Live web call with AI agent
Send a message
AI replies in seconds
AI is speaking...
The Platform
Three products, one platform — everything needed to deploy conversational AI agents.
Agent Management Dashboard
- Create & configure agents with custom prompts
- Select from 13 voices and 45+ languages
- Attach knowledge bases and API tools
- Publish to phone, widget, or WhatsApp
- View call logs, transcripts, and recordings
- Analytics dashboard with success rates
Embeddable Widget
- Voice calls via custom WebSocket audio pipeline
- Chat via REST + SSE token streaming
- Three.js animated blob visualizer
- Deploy via script tag or iframe
- Full branding customization (colors, text, icons)
- Client-side VAD & interruption handling
Backend Engine
- NestJS 11 with 38 feature modules
- OpenAI Realtime API for voice (audio→audio)
- Twilio WebSocket media stream bridge
- Knowledge base RAG with pgvector
- Zapier/MCP + custom API tool execution
- Multi-tenant billing with Stripe
Core Capabilities
What each AI agent can do out of the box.
Real-Time Voice Agents
Inbound and outbound calls over Twilio with WebSocket media streaming piped directly into OpenAI's Realtime API. Bi-directional audio at 24kHz with client-side VAD, interruption handling, and sub-second response latency.
Web Chat with SSE Streaming
Embeddable chat widget powered by Server-Sent Events for real-time token streaming. Context-aware responses with full conversation history, inactivity timeout handling, and automatic session management.
Knowledge Base RAG
Upload documents, websites, and FAQs to build a searchable knowledge base. LangChain + OpenAI embeddings (text-embedding-3-small, 1536 dims) with pgvector for semantic search — agents pull relevant context on every turn.
Custom API Actions
Define REST endpoint templates that agents invoke as tools during conversation. Supports variable substitution, multiple auth types (API Key, Bearer, Basic, Custom Header), and request/response templating for any external system.
Zapier & MCP Integration
Connect to thousands of apps through Zapier via Model Context Protocol. Dynamic tool discovery from MCP servers, rate-limited execution (5 calls/session), and full audit trail of every action taken.
45+ Languages & 13 Voices
Support for 45+ languages including European, Asian, Middle Eastern, and Indian languages. 13 distinct voice options (alloy, ash, ballad, coral, sage, and more) with configurable temperature and reasoning parameters.
Architecture
The production stack behind every conversation.
| Layer | Technology | Details |
|---|---|---|
| Telephony | Twilio Voice + WebSocket | Inbound/outbound calls, WebSocket media streams at /ws/twilio, call recording, per-workspace credential isolation |
| Widget Transport | Socket.io + SSE | Browser WebSocket gateway at /ws/widget for voice, REST + SSE streaming for chat, domain whitelist CORS |
| Voice AI Engine | OpenAI Realtime API | End-to-end voice model — audio in, audio out. No separate STT/TTS pipeline. 24kHz PCM, real-time transcription |
| Tool Execution | MCP + Custom REST | Knowledge base search (RAG), Zapier/MCP actions, custom API endpoints, current datetime — all invocable by the LLM |
| Knowledge Base | LangChain + pgvector | Document embedding (text-embedding-3-small), semantic vector search, S3-backed file storage |
| Orchestration | NestJS 11 + TypeScript | 38 feature modules, WebSocket gateways, circuit breaker, rate limiting, OpenAI key pool load balancing |
| Multi-Tenancy | Org → Workspace → Agent | RBAC with 4 org roles + 3 workspace access levels, per-workspace Twilio credentials and billing isolation |
| Billing | Stripe | 5-tier SaaS plans ($49–$1199/mo), usage-based overage billing, metered voice minutes, add-on management |
| Session Analytics | Post-Call Analysis | AI-generated summaries, category classification (Lead, Support, Inquiry), transcript storage, recording links |
| Data Layer | PostgreSQL + Prisma | Full schema — agents, sessions, call records, knowledge bases, vectors, billing events, RBAC permissions |
| Infrastructure | AWS EC2 + PM2 | Production on EC2, PM2 process management, S3 for storage, Vercel for frontend/widget deployment |
How a Voice Call Works
End-to-end flow from caller to AI response.
Call Received
Twilio webhook triggers agent lookup, WebSocket media stream established
Realtime Session
OpenAI Realtime API session created with agent config, system prompt, and tools
Audio Streamed
Bi-directional 24kHz PCM audio flows through Twilio ↔ OpenAI bridge in real-time
Tools Invoked
LLM calls knowledge base, Zapier actions, or custom APIs mid-conversation
Session Analyzed
Transcript stored, AI generates summary, call categorized (Lead, Support, Inquiry)
Agent Tool System
Agents don't just talk — they act. Every agent can invoke tools mid-conversation.
search_knowledge_base
Queries the attached knowledge base with semantic vector search. Returns relevant document chunks from uploaded files, websites, and FAQs. Powered by LangChain + OpenAI embeddings + pgvector.
Zapier MCP Actions
Executes any action from the customer's connected Zapier instance via Model Context Protocol. Dynamic tool discovery, rate-limited to 5 calls per session, with full execution audit trail.
Custom API Actions
User-defined REST endpoint templates with variable substitution. Supports None, API Key, Bearer Token, Basic Auth, and Custom Header authentication. Agents call these to interact with any external system.
get_current_datetime
Provides the current date and time in the agent's configured timezone. Essential context for scheduling-related conversations, appointment inquiries, and time-sensitive operations.
Deployment Channels
One agent, multiple channels — publish once, deploy everywhere.
Phone Calls
Assign a Twilio number, agent handles inbound/outbound calls
Website Widget
Embed via script tag or iframe — voice and chat in one widget
Connect WhatsApp number for messaging channel deployment
In-App Copilot
Embed as an AI assistant inside your own application
SaaS Billing
5-tier pricing with Stripe-powered subscriptions, usage tracking, and overage billing.
| Plan | Price | Voice Min | Chat Sessions | Agents | Concurrency |
|---|---|---|---|---|---|
| Starter | $49/mo | — | 500 | 1 | — |
| Growth | $99/mo | 100 | 500 | 1 | 2 |
| Pro | $199/mo | 200 | 1,000 | 2 | 2 |
| Scale | $499/mo | 800 | 2,000 | 5 | 5 |
| Business | $1,199/mo | 2,500 | 5,000 | 15 | 15 |
Tech Stack
Want this for your business?
We deploy conversational AI agents tailored to your workflows, systems, and industry — or license the full Loquent platform. Let's talk.