The Texa
Intelligence Platform
A comprehensive guide to the omnichannel AI command center. Manage agents, monitor analytics, and orchestrate intelligent conversations across WhatsApp, Instagram, and Voice from one place.
How Texa Works
Every message flows through a 5-step intelligence pipeline. Context is loaded in parallel for zero added latency, and the agent learns from every interaction.
Message Arrives
Channel LayerA customer sends a WhatsApp message, Instagram DM, or calls your phone number. The channel-specific responder receives it, normalizes the format, and passes it into the unified pipeline.
Context Loading
AggregationIn a single parallel fetch (zero added latency), the system loads everything the agent needs to give an intelligent response.
Intelligent Processing
Universal BrainThe orchestrator builds a rich system prompt with all context injected, then runs a multi-round LLM loop with tool calling. The agent can search the KB, escalate, send cross-channel messages, book appointments, and more. Anti-hallucination guardrails verify every response before delivery.
Response Delivered
ResponderThe final response is sent back through the original channel. For WhatsApp and Instagram, it's a text message. For Voice, it streams through real-time text-to-speech with sub-second latency.
Background Learning
Post-ProcessingAfter the response is sent, background tasks run in parallel. Memory is updated, session state is extracted, and the self-learning engine captures improvement observations. All zero-latency, zero-cost to the customer.
Analytics Dashboard
Transform raw data into actionable business intelligence. The Analytics Dashboard gives you a real-time pulse on your entire communication infrastructure across WhatsApp, Instagram, and Voice. Identify trends, optimize response times, and understand customer behavior at a glance.
Business Impact
Make data-driven decisions with instant visibility into message volume, channel performance, and customer engagement across your omnichannel pipeline.
Live Metrics Feed
Monitor active chats, total contacts, and message throughput across all channels in real-time.
Trend Visualization
Interactive 7-day charts reveal peak engagement hours and traffic patterns per channel.
AI-Generated Alerts
Automatic detection of urgent conversations (High/Medium/Low) based on sentiment analysis and emotional intelligence.
User Insights Stream
A dynamic feed of new facts and preferences learned from ongoing conversations via the memory system.
Cross-Channel Analytics
Unified view of customer journeys that span WhatsApp, Instagram, and Voice channels.
AI Agent Configuration
Your AI agents are the face of your brand across every channel. The Agent Configuration engine lets you craft sophisticated personas with deep contextual understanding. Define their voice, boundaries, knowledge base, and cross-channel capabilities to deliver a seamless, on-brand experience on WhatsApp, Instagram, and Voice.
Business Impact
Deploy intelligent agents that perfectly represent your brand 24/7. One agent, all channels, consistent experience.
Persona Design
Granular control over tone, style, and operational rules with up to 100k character prompts.
Knowledge Base Integration
Upload PDFs, CSVs, TXT files, or connect Google Drive for document-based answers with vector search.
Multi-Agent Management
Create specialized agents for Sales, Support, or HR and switch them instantly.
Feature Toggles
Enable or disable capabilities per agent: memory, self-learning, cross-channel, decision engine, emotional intelligence, and more.
Owner Rules & Guardrails
Set business rules that override AI behavior. Your brand guidelines are always respected.
Cross-Channel Intelligence
Seamlessly connect customer conversations across WhatsApp, Instagram, and Voice. Texa's Cross-Channel Identity system automatically detects when the same person contacts you on different channels and maintains a unified profile. When a WhatsApp customer calls, the voice agent already knows their history.
Business Impact
Never ask a customer to repeat themselves. Their context follows them across every channel automatically.
Unified Identity Resolution
Automatic matching of phone numbers and identifiers across WhatsApp, Instagram, and Voice channels.
Channel Transition Tracking
Full audit trail of every channel switch. Know when a customer moved from WhatsApp to a call and why.
Cross-Channel Actions
AI agents can initiate WhatsApp messages during a voice call, or trigger an outbound call from a chat conversation.
Context Handoff
When a customer transitions channels, their recent chat history, topic, intent, and sentiment are automatically carried over.
Smart Channel Routing
The Decision Engine recommends the optimal channel for each interaction (e.g., suggest a call for complex issues, WhatsApp for document delivery).
Session State Architecture
Enterprise-grade conversation session management modeled after WhatsApp's own 24-hour window architecture. Sessions track the active conversation context (goal, topic, intent, sentiment) and share it across channels in real-time. When a customer switches from WhatsApp to a Voice call, the session transitions seamlessly.
Business Impact
Conversations feel continuous and natural, even when customers switch channels. No context is ever lost.
24-Hour Session Window
Sessions auto-idle after 30 minutes and auto-close after 24 hours of inactivity, matching WhatsApp's native model.
Session vs. Memory Separation
Sessions are ephemeral (current conversation context), while Memory is permanent (customer profile, preferences). Two-layer architecture ensures both short-term and long-term intelligence.
Live State Tracking
Every session tracks the current topic, customer intent, sentiment, and conversation goal. Updated in real-time with zero LLM cost.
Channel Transition Logging
Full history of channel switches (e.g., whatsapp to voice to whatsapp) with timestamps and reasons.
Context Snapshot
When a session transitions channels, the previous channel's conversation summary is injected into the new channel's prompt.
Intelligent Memory System
A three-layer memory architecture that gives your agent permanent knowledge about every customer. Short-term memory tracks the current conversation, working memory stores recent facts and preferences, and long-term memory maintains a distilled customer profile that persists forever.
Business Impact
Your agent remembers every customer's name, preferences, past issues, and communication style. This creates genuinely personalized interactions at scale.
Working Memory
Recent facts and context from the last few conversations, automatically updated after each interaction.
Long-Term Facts
Permanent storage of key customer information like name, preferences, past purchases, and important dates. Extracted automatically.
Distilled Summary
An AI-compressed customer profile that captures their personality, communication style, and key history in a compact format.
Memory-Powered Responses
The LLM prompt automatically includes relevant memory context, so the agent references past conversations naturally.
Customer Context Profile
A structured overlay that tracks topic, sentiment, and engagement patterns for each customer.
Voice AI (Voxa Integration)
Full voice calling capabilities powered by the Vobiz telephony platform. Your AI agent can receive inbound calls and make outbound calls, with real-time speech-to-text, streaming LLM responses, and natural text-to-speech. During voice calls, the agent can execute actions like sending WhatsApp messages, booking appointments, or scheduling follow-up calls.
Business Impact
Transform your phone support with AI agents that speak naturally, understand context from chat channels, and execute real actions during calls.
Real-Time Streaming
Speech-to-text, LLM, text-to-speech pipeline with sub-second latency for natural conversation.
Proper Tool Calling
OpenAI function calling integration. The agent uses structured tools (send_whatsapp, book_appointment, schedule_call, escalate, end_call) instead of fragile text parsing.
Interrupt Handling
Customers can interrupt the AI mid-speech. The system detects voice activity, cancels TTS immediately, and processes the new input.
Cross-Channel Context
Inbound and outbound calls automatically load the customer's session state, chat history, and memory from WhatsApp/Instagram.
Outbound Calls with Context
When the AI initiates a call from a chat conversation, the full chat history is injected so the call continues seamlessly.
Call Transfer
Smart escalation to human agents via SIP transfer with configurable transfer numbers.
Smart Escalation System
A multi-channel escalation engine that routes complex queries to the right human at the right time. The AI automatically detects when it can't help (confusion, anger, explicit requests, or business-critical topics) and gracefully hands off to your team without disrupting the customer experience.
Business Impact
Never lose a lead or frustrate a customer due to AI limitations. Complex cases reach the right expert instantly.
Smart Triggers
AI automatically detects confusion, anger, or explicit requests for human help using emotional intelligence.
Multi-Channel Routing
Escalations work across WhatsApp (forwarded to manager's WhatsApp), Instagram (DM relay), and Voice (call transfer).
Intelligent Manager Matching
Routing rules match escalations to the best-suited manager based on situation type, expertise, and availability.
Invisible Relay
Managers reply to the bot, which relays the message to the client. This maintains a single point of contact.
Escalation Knowledge Capture
When managers resolve escalations, the AI learns from their responses and can handle similar situations autonomously next time.
Resolution Tracking
Full audit trail of every escalated ticket from open to resolved, with classification (general knowledge vs. customer-specific).
Self-Learning Engine
Your AI agent continuously improves from every interaction. The Self-Learning Engine captures successful patterns, manager responses, and conversation outcomes, then compiles them into learned strategies that make the agent smarter over time. No manual training required.
Business Impact
Your agent gets measurably better over time. What required escalation last month gets handled autonomously today.
Counterfactual Learning
After every response, the system evaluates what could have been better and stores improvement observations.
Escalation Knowledge Mining
When human managers resolve escalated queries, their responses are automatically extracted, classified, and stored as reusable knowledge.
Strategy Compilation
Learned observations are compiled into actionable strategies that are injected into the agent's prompt.
Metacognition Loop
Each response is scored (1-10) on quality, with specific improvement suggestions stored for future reference.
Mistake Journal
Past errors are tracked with severity and corrections, ensuring the agent doesn't repeat the same mistakes.
Anti-Hallucination Guardrails
Enterprise-grade safety nets that prevent your AI from making false claims. The multi-layer guardrail system catches hallucinated responses before they reach customers. If the AI claims 'I've sent your WhatsApp' but no tool was actually called, the system detects it and replaces the response automatically.
Business Impact
Trust your AI to never lie to customers. False claims about actions, calls, or messages are caught and corrected automatically.
Action Claim Detection
Pattern matching detects when the LLM claims an action (call, WhatsApp, appointment, DM) that wasn't actually executed.
Automatic Response Replacement
Hallucinated claims are replaced with safe, honest responses that offer to perform the action properly.
Tool Verification
Cross-references the LLM's text output against the list of tools actually called during the conversation.
Confidence Scoring
Each response receives a confidence score (0-100%) based on knowledge base coverage and response quality.
Uncertainty Detection
Responses that seem speculative or contain hedge words are flagged for review.
Emotional Intelligence
Your AI agent doesn't just understand what customers say, it understands how they feel. The Emotional Intelligence module detects frustration, excitement, confusion, or urgency in every message and adapts the response tone accordingly. Angry customers get empathy first, then solutions.
Business Impact
Transform robotic interactions into human-like conversations. Customers feel heard and understood, driving higher satisfaction.
Real-Time Emotion Detection
Classifies customer sentiment in every message (frustrated, happy, confused, urgent, neutral) using lightweight pattern analysis.
Tone-Adaptive Responses
The system prompt dynamically adjusts based on detected emotion. Frustrated customers get empathy first; excited ones get matching enthusiasm.
Sentiment Tracking
Session-level sentiment tracking shows how customer mood evolves throughout the conversation.
Emotional Escalation Triggers
Extreme negative sentiment can auto-trigger escalation to human agents before the situation worsens.
Knowledge Base with Vector Search
Your AI agent's knowledge center. Upload documents (PDF, CSV, TXT, or Google Drive URLs) and they're automatically chunked, embedded, and indexed for semantic search. When a customer asks a question, the system finds the most relevant passages across all documents and provides accurate, sourced answers.
Business Impact
Give your agent expert-level knowledge about your products, pricing, and policies. Updated instantly when you upload new documents.
Vector Embeddings
Documents are split into chunks and embedded using OpenAI's text-embedding model for semantic similarity search.
Pre-Flight KB Search
Before the LLM responds, a semantic search runs against your knowledge base to pre-load relevant context. This reduces hallucination and improves accuracy.
Sendable Documents
Mark specific documents as 'sendable' and the AI can proactively share them with customers during conversations.
Escalation Knowledge Search
When handling questions the agent has seen before (via escalation), it searches past manager responses first for instant, verified answers.
Token-Optimized Retrieval
Smart chunking and relevance scoring ensures only the most useful passages are sent to the LLM, minimizing token usage.
Decision Engine
An advanced reasoning layer that activates for complex queries. When the standard response pipeline isn't enough, the Decision Engine engages deep reasoning, analyzing the customer's situation, evaluating multiple response strategies, and selecting the optimal approach before generating a response.
Business Impact
Handle complex, multi-faceted customer queries with strategic thinking, not just pattern matching.
Deep Reasoning Activation
Automatically detects when a query requires strategic thinking (pricing negotiations, complaints, multi-step processes).
Channel Selection Intelligence
Recommends the optimal channel for each interaction based on context. Voice for urgent issues, WhatsApp for document delivery.
Contact Priority Scoring
Dynamic scoring of customer importance based on engagement patterns, purchase history, and conversation sentiment.
Strategy Evaluation
Evaluates multiple response approaches and selects the one with the highest confidence score.
Proactive Scheduling & Tasks
Your AI agent doesn't just respond, it proactively manages tasks, appointments, and follow-ups. Schedule outbound calls, set reminders, book appointments, and create follow-up tasks directly from conversations. The Heartbeat system ensures scheduled tasks are executed automatically.
Business Impact
Turn your AI from reactive to proactive. Follow-ups, reminders, and scheduled calls happen automatically without human intervention.
Appointment Booking
The AI can book, confirm, and manage appointments with customers directly during conversations.
Call Scheduling
Schedule follow-up calls for specific times. The system automatically makes the outbound call when the time arrives.
Reminder System
Create time-based reminders for both agents and customers, ensuring nothing falls through the cracks.
Heartbeat Engine
Autonomous execution system that processes scheduled tasks, sends proactive messages, and manages the agent's daily plan.
Today's Plan
Daily task pipeline that generates and executes the agent's scheduled activities (reminders, follow-ups, check-ins).
Live Chat Control
Maintain absolute control over your AI conversations. The Live Chat interface lets you monitor AI interactions across WhatsApp and Instagram in real-time, intervene manually whenever necessary, and switch between autopilot and manual mode seamlessly.
Business Impact
Complete visibility and one-click takeover. Seamlessly switch between AI autopilot and manual control.
Real-Time Monitoring
Watch messages stream in with zero latency across all channels.
Manual Takeover
Instantly pause the AI and type a response yourself, without the customer noticing.
Rich Context Sidebar
View the client's full profile, memory, tags, session state, and learned preferences while you chat.
Whisper Mode
Inject instructions to the AI mid-conversation without the customer seeing (e.g., 'Offer a 10% discount').
Searchable History
Rapid-access archives of every conversation across all channels.
LLM Intelligence Layer
Model-agnostic architecture that lets you choose the AI engine powering your agents. Swap between providers and models instantly, configure API keys securely, and optimize the cost-performance balance for your specific use case.
Business Impact
Future-proof your AI investment. Always use the best model for the job, and switch providers in seconds.
Provider Agnostic
Native support for OpenAI (GPT-4o, GPT-4o-mini), Anthropic (Claude 3.5), Google (Gemini), and Groq.
Secure Key Vault
Enterprise-grade encryption for your API keys with per-service configuration (LLM, STT, TTS).
Smart Token Management
Compression service, pre-flight KB optimization, and context window management to minimize costs.
Thinking Mode
Optional reasoning traces (think-then-respond) for complex queries. See the agent's reasoning chain.
Team Collaboration
Built for scale. The Team Collaboration suite empowers agencies and large organizations to work together securely, with granular permissions and audit logs to keep everyone accountable.
Business Impact
Scale your operations securely with robust role-based access control and team management.
Role-Based Access
Pre-defined roles for Owners, Admins, and Viewers with granular permissions.
Secure Invites
Email-based invitation system for safe onboarding.
Activity Logs
Track who changed what configuration and when.
Multi-Agent Workspace
Each team member can manage different agents independently.
Setup & Configuration Guides
1Getting Started
Before you can use the console, connect your communication channels. This is a one-time setup per channel.
Connect WhatsApp
Log in to the console and click 'Connect with Facebook'. Select your business portfolio and the WhatsApp phone number to automate. Ensure your number is verified in Meta Business Manager.
Connect Instagram
Link your Instagram Business account through the same Facebook authentication flow. Select the Instagram account you want your AI agent to manage.
Configure Voice (Optional)
Go to Settings > Phone and enter your Vobiz credentials (Auth ID, Auth Token, From Number). Point your Vobiz incoming call webhook to your console's /api/vobiz/incoming endpoint.
Access Dashboard
Once connected, you'll see your unified dashboard with metrics across all channels.
2Configuring Your Agent
Your AI agent needs to know who it is, what it knows, and what it can do. Configure this in the 'AI Agents' tab.
Define Persona
In the 'Prompt' field, describe your agent's role, tone, and rules (e.g., 'You are a helpful sales assistant for L10 AI Labs. Always be professional.'). Be specific about limitations and escalation criteria.
Upload Knowledge
Upload PDF, TXT, or CSV files to the Knowledge Base. These are automatically chunked, embedded, and indexed for semantic search. The agent will reference these documents when answering queries.
Enable Features
Toggle capabilities: Memory (remembers customers), Self-Learning (improves over time), Cross-Channel (links identities across channels), Decision Engine (strategic reasoning), and more.
Set Owner Rules
Add business rules that override AI behavior, such as pricing policies, refund rules, or items you never want shared. These are injected as high-priority instructions.
Test & Activate
Use the built-in testing tool to chat with your agent in safe mode (no real side effects). Once satisfied, activate it to handle real customer messages.
3Setting Up Cross-Channel
Cross-channel intelligence lets customers move between WhatsApp, Instagram, and Voice without losing context.
Enable Cross-Channel
In your agent's capabilities, toggle on the 'Cross-Channel' feature. This enables unified identity resolution and cross-channel actions.
Connect All Channels
For full omnichannel support, connect WhatsApp, Instagram, and Voice to the same agent. The system automatically links customer identities across channels.
Test a Transition
Send a WhatsApp message, then call the same number. The voice agent should reference your WhatsApp conversation. Check logs for 'Session channel transition logged'.
Monitor Transitions
The session_state table in your database tracks every channel transition with timestamps and reasons.
4Setting Up Escalation
Configure your escalation managers so complex queries reach the right person on the right channel.
Add Escalation Managers
Go to the Escalation tab and add managers with their WhatsApp numbers. Assign categories (billing, technical, general) for smart routing.
Configure Routing Rules
Set up routing rules that match escalation reasons to specific managers based on expertise and availability.
Test the Flow
Send a message like 'I need to speak to a human' or express frustration. The AI should detect the need and route to your configured manager.
Enable Knowledge Capture
When managers resolve escalations, their responses are automatically classified and stored. Toggle on 'Escalation Knowledge' to let the agent learn from these resolutions.
5Voice Calling Setup
Set up AI-powered phone calls with the Vobiz integration. Your agent can receive and make calls with real-time speech processing.
Get Vobiz Credentials
Sign up for a Vobiz account and obtain your Auth ID, Auth Token, and a phone number. Enter these in Settings > Phone in the console.
Configure Incoming Calls
In the Vobiz dashboard, set the 'Answer URL' of your phone number to: https://yourdomain.com/api/vobiz/incoming. This routes calls to your AI agent.
Configure STT & TTS
Go to Settings > Speech and configure your Speech-to-Text (ElevenLabs, AssemblyAI) and Text-to-Speech (Inworld, ElevenLabs) providers with API keys.
Test an Inbound Call
Call your Vobiz number. The AI should greet you and handle the conversation. Check console logs for 'VobizStream' entries.
Make Outbound Calls
From a WhatsApp/Instagram conversation, the AI can initiate outbound calls to customers. The chat context is automatically passed to the voice agent.