Texa Documentation

Welcome to the Texa platform. Build, deploy, and manage intelligent omnichannel AI agents that handle WhatsApp, Instagram, and Voice seamlessly.

Texa Main Dashboard — The centralized command center for monitoring all AI operations.

Traditional chatbots rely on rigid decision trees. Texa agents use a combination of Vector Search, Emotional Intelligence, and Multi-layer Memory to hold dynamic, human-like conversations. They understand intent, adjust their tone, and execute real-world actions like booking appointments or processing payments.

NoteAre you an agency? Texa supports multi-agent workspaces, allowing you to manage AI agents for dozens of clients from a single, centralized dashboard.

Quickstart Guide

Get your first AI agent live on WhatsApp and Instagram in under 10 minutes by following these three steps.

Connect your Channels

Before an agent can communicate, it needs a conduit. Navigate to the Channels tab in your console.

Click Connect WhatsApp and log in with Facebook. Select the business number you wish to automate.
Click Connect Instagram to link your business profile.

Configure the Agent Persona

Your agent needs an identity and rules. Go to AI Agents and select your active agent.

System Prompt: Define who the agent is (e.g., "You are Alex, the friendly support rep for L10 AI Labs.").
Owner Rules: Input absolute rules that override standard LLM behavior (e.g., "Never offer discounts greater than 10%").

Upload Knowledge

An agent is only as smart as the data it has access to. Go to the Knowledge Base tab.

Upload your pricing PDFs, policy documents, and product FAQs. The system automatically chunks, embeds, and indexes these documents for instant retrieval during conversations.

TipYou can also connect a Google Drive folder. Whenever you update a file in Drive, the agent's knowledge updates automatically.

Texa Architecture

Understanding how the three foundational pillars interact with the core reasoning engine.

Memory Layer

Context Provider

Texa Core

Reasoning Engine

Tool Engine

Deterministic Execution

Human Escalation

Safety Net Takeover

Unlike basic LLM wrappers, the Texa Core operates synchronously with its environment. It pulls immutable facts and recent context from the Memory Layer before generating a response. If an action is required (like processing an order), it deterministicly invokes external APIs via the Tool Engine. If it detects emotional distress or high complexity, it breaks the autonomous loop and routes the session to the Escalation net.

Memory System

The Texa memory architecture operates on three distinct layers to ensure context is maintained both short-term (during an active session) and long-term (across a customer's lifetime).

Awaiting: memory-profile.png

A view of a customer's profile, showing extracted long-term facts and distilled summary.

1. Session State (Short-term)

Tracks the active conversation. It records the current intent, emotional sentiment, and immediate goal. If the session goes idle for 24 hours, it closes automatically.

2. Working Memory (Medium-term)

Holds facts relevant to the last few days. If a customer says "I'll be out of town until Friday," this resides in working memory so the agent doesn't schedule a call on Thursday.

3. Long-term Facts (Permanent)

Extracts and stores immutable traits: Name, Email, preferred language, VIP status, and purchase history. These are injected into every prompt for this specific customer.

Omnichannel Engine

Texa unifies WhatsApp, Instagram, and Voice. Customers don't exist in silos; they are treated as a single unified identity across platforms.

Cross-Channel Transitions

If a customer starts a conversation on WhatsApp but decides to call your business phone number (powered by Voxa Voice), the voice agent is instantly injected with the WhatsApp chat history.

The voice agent might say: "Hi John, I see you were just asking about the Pro plan on WhatsApp. Do you have any specific questions about it?"

TipVoice agents can also trigger chat actions. A voice agent can use the send_whatsapp tool to send a document to the user while they are still on the call.

Smart Escalation

AI isn't perfect, and high-value customers shouldn't be left in loops. The Smart Escalation engine acts as a safety net.

Detection

The system monitors every incoming message for explicit requests ("Let me talk to a human") and implicit frustration (via the Emotional Intelligence module).

Routing

Based on the detected topic (e.g., Billing vs. Tech Support), the system pings the assigned manager via their personal WhatsApp or Email.

Invisible Relay

The manager replies directly from their phone. Texa relays the manager's message back to the client through the bot. The customer never knows a handoff occurred.

Commerce & Payments

Turn conversations into revenue. Texa integrates deeply with Razorpay to facilitate secure, owner-approved transactions entirely within chat.

The Transaction Workflow

To prevent hallucinations regarding money, Texa strictly enforces an owner-approval loop:

The agent uses the Product Catalog to help a user select items.
The agent creates a Pending Order using backend tools.
The business owner receives a notification and reviews the order in the Dashboard.
Upon Owner Approval, a Razorpay payment link is automatically generated and sent to the customer's WhatsApp.
The agent monitors the check_payment_status tool and sends a receipt once payment succeeds.

WarningThe AI agent cannot generate payment links or alter prices autonomously. All financial transactions require human approval in the dashboard to ensure 100% safety.