# Iqra AI - The Dynamic AI-First Computing Engine (EN)

> Orchestrate superhuman agents that think dynamically but act systematically. Bridge the gap between the chaos of LLMs and the reliability of code.

## Links
- Website: https://iqra.bot/
- Cloud Dashboard: https://app.iqra.bot
- Documentation: https://docs.iqra.bot
- API Reference: https://docs.iqra.bot/api-reference
- Developer Docs: https://docs.iqra.bot/developers
- GitHub: https://github.com/abdofallah/IqraAI

---

## 1. Core Philosophy: The Context
The way software handles conversation has evolved. We are building the third generation of automation infrastructure.

### The Evolution of Automation
1. **2010 - 2022 - Static Rigidity**: User flows forced into rigid boxes. "Press 1 for Sales." or "Fill Form.". Complex to code every scenario, and frustrating for users.
2. **CURRENT STATE - Dynamic Chaos**: LLMs introduced intelligence, but without boundaries. They hallucinate, leak data, and can't be trusted with business logic.
3. **IQRA AI - Systematic Flow**: We harness the dynamic power of AI, but channel it through Deterministic Gates. The AI thinks, but the System executes.

---

## 2. System Architecture (System Architecture)
A unified infrastructure layer handling perception, cognition, logic, and security in real-time.

### A. Inputs (Perception)
- **Telephony**: SIP / PSTN
- **WebRTC**: Browser
- **WhatsApp**: Messenger

### B. The Core Engine (Cognition)
- **Context Layer**: RAG / Knowledge (Vector Retrieval) & Tools (Function Calling)
- **Neural Loop**: STT (Input) -> LLM (Inference) -> TTS (Output)
- **Guardrails**: VAD / Interruption & Multilingual

### C. Logic & Control (Execution)
- **Flows**: Deterministic Logic
- **Apps**: External Plugins
- **Secure**: Redaction Vault
- **Analysis**: JSON Extraction

---

## 3. The Visual IDE
One interface to orchestrate logic, configure intelligence, and debug conversations. No context switching.

### View 1: Logic Builder (Conversation)
Orchestrate the conversation flow and business rules.
- **Action Flows**: Embed deterministic logic (If/Else, Loops) to enforce strict business rules.
- **Secure Input**: Collect PCI/PII data (DTMF) in a "Clean Room" without LLM exposure.
- **FlowApps**: Native plugins for Cal.com, HubSpot, and Custom Tools.
- **Global Context**: Switch language views instantly. Validates missing translations per node.
- **State Variables**: Manage global session memory, hidden flags, and Scriban data.

### View 2: Agent Studio (Agent)
Configure the intelligence and personality of the fleet.
- **Agent Fleet**: Manage your workforce. Create distinct personas for Sales, Support, or Operations.
- **Context Injection**: Automatically inject business branding, services, and branch details.
- **System Prompt**: Define the Role, Ethics, and Capability boundaries.
- **Dynamic Greeting**: Configure the opening message ("Salam!") or wait for the user.
- **Smart Turn Detection**: Intelligently decide when the user has finished a sentence.
- **AI Verification**: Prevents false interruptions (e.g., distinguishing "Stop" from "Uh-huh").
- **The Neural Stack**: Configure Brain (LLM), Ears (STT), and Voice (TTS) with fallback chains.
- **Hyper-Parameters**: Fine-tune specific models and adjust temperature settings.
- **Search Strategy**: Decide *when* to search: Always, Keyword Match, or Smart Intent.
- **Semantic Cache**: Auto-save vectors. Reduces latency to 0ms and costs to $0 for repeats.
- **Environment Audio**: Mix in background noise for hyper-realism.

---

## 4. Platform Capabilities (Detailed Specs)

### A. Deploy Everywhere
One brain, many bodies. Orchestrate conversations across Telephony, WebRTC, and Apps.

**Deep Dive:** Your Agent is the core intelligence. The Routing Layer is the bridge that connects that intelligence to the world. Build your logic once, deploy everywhere.

**1. Telephony Bridge**
Iqra AI operates on a BYOC (Bring Your Own Carrier) architecture. We provide the engine that powers your numbers. By handling the SIP signaling directly, we offer features standard carriers can't:
- Global Region Routing: Connect a UK number to London and a US number to Virginia in one workspace.
- Smart IVR: Programmatic language selection ("Press 1 for English") pre-AI initialization.
- Native Integrations: Twilio, Telnyx, and Vonage supported out of the box.
*Documentation:*
- Read Phone Numbers Docs: https://docs.iqra.bot/build/tools/phone-numbers
- Read Inbound Route Docs: https://docs.iqra.bot/build/campaigns/inbound
- Read Outbound Campaign Docs: https://docs.iqra.bot/build/campaigns/outbound

**2. The Digital Edge**
Bypass the telephone network entirely. Deploy directly to the web or mobile apps to unlock HD Audio (44.1kHz) and remove per-minute carrier fees.
- **WebRTC (UDP)**: Ultra-low latency for browser/app voice.
- **WebSockets (TCP)**: Robust fallback for server streams.
*Documentation:*
- Read Web Campaign Docs: https://docs.iqra.bot/build/campaigns/web
- View Client SDK Guide: https://docs.iqra.bot/developers/sdks/web-widget

**3. Omni-Channel Roadmap**
- WhatsApp Voice (Coming Soon): Native inbound & outbound calls.
- Meeting Bots (Coming Soon): Joins G-Meet & Zoom as a participant.
- Social Audio (Coming Soon): Discord Stage & Telegram Voice.


### B. Deterministic Logic
Don't let AI hallucinate business rules. Embed strict If/Else logic gates and loops inside the conversation.

**Deep Dive:** Creativity is a liability when calculating a loan rate. Action Flows introduce strict, code-based reliability into the fluid world of conversational AI.

**1. The Integrity Gap**
Standard AI agents rely on the LLM to handle everything. You might prompt: "Ask for ID, check database, read balance." But LLMs are probabilistic. They skip steps, hallucinate numbers, or get distracted.
> The Risk: For enterprise operations (Banking, Healthcare), 99% accuracy is failure. You need 100% adherence to business rules.

**2. Embedded Workflows**
An Action Flow is a "Mini-Program" embedded directly into your conversation script. When the AI detects a critical intent, it stops "thinking" and hands control to the Logic Engine.
- **AI Muted**: The LLM stops generating tokens. Zero hallucination risk.
- **System Execution**: The engine runs strict If/Else logic, loops, and API calls.
- **Return to Chat**: The result (not the raw data) is handed back to the AI to speak naturally.

**3. The Logic Toolkit**
- Condition Nodes: Branching based on variables.
- Run Code: Sandboxed JS environment.
- Loops: Iterate through API lists.
*Documentation:* View Documentation: https://docs.iqra.bot/build/script/action-flows


### C. Native Multilingual
Parallel logic contexts for English, Arabic, Chinese and more.

**Deep Dive:** Translation layers miss the soul of a language. Iqra AI uses a Parallel Context Architecture to ensure your agent speaks with correct grammar, tone, and cultural nuance.

**1. The "Translation Layer" Trap**
Most platforms use a cheap pipeline: Translate User Audio -> Process in English -> Translate AI Response -> Speak.
> This creates Double Latency and loses meaning. An English AI doesn't know that "Assalamu Alaykum" is formal while "Ahlan" is casual. It treats language as math, stripping away culture.

**2. Parallel Contexts**
Iqra AI allows you to define distinct "Brains" for each language within a single agent. The flow logic (e.g., "Book Appointment") is shared, but the Content Layer is unique.
- **Distinct Prompts**: Tell the English AI to be "Professional" and the Arabic AI to be "Warm & Hospitable."
- **Native Stack**: Use Deepgram for English (Fast) but switch to Azure Speech for Arabic (Best Dialects) automatically.

**3. Live Switching**
Users switch languages mid-sentence. Our engine detects the intent, unloads the English stack, and instantly loads the Arabic configuration (LLM + STT + TTS) without dropping the call.
*Documentation:* Read Guide: https://docs.iqra.bot/build/agent/multi-language


### D. Secure Sessions
Collect sensitive data in a "Clean Room" isolated from the LLM.

**Deep Dive:** Standard AI agents are "always listening." This is a security nightmare. Secure Sessions collect sensitive data in an air-gapped environment, ensuring zero exposure to the LLM.

**1. The "Always On" Risk**
When a user speaks a Credit Card number to a standard AI, the data travels through multiple servers in plain text: STT Provider -> Cloud LLM (OpenAI) -> Logs.
> Compliance Violation: Even with system prompts ("Ignore numbers"), the raw audio has already been processed by third parties. This violates PCI-DSS and GDPR.

**2. Zero-Trust Collection**
When the script reaches a sensitive step (e.g., "Collect Payment"), the system triggers a Secure Session.
- **AI Muted**: The connection to the LLM is physically severed. The AI stops "hearing."
- **System Takeover**: The deterministic engine collects input via DTMF (Keypad) or Voice.
- **Instant Redaction**: As data arrives, it is masked (****) in memory immediately.

**3. The Blind Handoff**
Q: How does the AI process payment if it can't see the card number?
A: Reference Variables. The AI sees a variable name var.cc_number and its status Collected. It cannot read the value. When the AI calls the Stripe Tool, the System passes the decrypted value in the background. The AI gets a Success/Fail result, never touching raw data.
*Documentation:* Read Security Docs: https://docs.iqra.bot/build/script/secure-sessions


### E. Smart Turn Taking
Conversations, not walkie-talkies. Distinguish between a pause, a backchannel ("uh-huh"), and a true interruption.

**Deep Dive:** Standard Voice AI relies on silence to know when to speak, leading to awkward interruptions. Iqra AI uses Semantic Intelligence to understand context, pauses, and backchannels.

**1. The "Silence" Trap**
Legacy systems use VAD (Voice Activity Detection). It follows a dumb rule: "If silence > 500ms, start speaking."
> The Problem: Humans pause to think."My email is... john... dot... doe..."Legacy bots interrupt after "john" because of the pause. Result: Frustrated users.

**2. Context Aware Endpointing**
Iqra AI offers multiple strategies to detect the end of a turn, balancing speed vs. context.
- **VAD (Fastest)**: Good for simple "Yes/No" questions.
- **Grammar Check**: Ensures the user finished a complete sentence structure.
- **LLM Decision (Smartest)**: Analyzes meaning. Understands that "I'm looking for a..." is incomplete, even if there is a 3-second pause.

**3. Smart Interruptions**
- **Backchanneling ("Uh-huh")**: When the user says "Right", "Okay", or "Mhmm", the system detects agreement. It lowers the volume briefly but keeps speaking. It feels natural.
- **Barge-In ("Stop!")**: When the user says "Wait", "No", or "Hold on", the system detects a Semantic Interrupt. It halts audio immediately and listens.
*Documentation:* Read Guide: https://docs.iqra.bot/build/agent/interruption


### F. FlowApps
Native integrations for your stack.

**Deep Dive:** Don't waste time writing raw HTTP requests or debugging JSON payloads. FlowApps are managed, intelligent integrations that give your agent "Hands" instantly.

**1. Stop Reinventing the Wheel**
While our Custom Tool engine is powerful enough to connect to anything, building a robust integration from scratch is expensive.
- Reading API Docs & Mapping JSON manually.
- Writing JavaScript to parse complex responses.
- Broken connectors when APIs change.
> The Solution: FlowApps are native plugins maintained by the platform. You drag them onto the canvas, authenticate via Integrations, and they just work.

**2. Smart Context UI**
Standard Webhooks are "dumb"—they are just text boxes. FlowApps are "smart." They change the Script Builder interface based on the tool you select.
- **Dynamic Fetchers**: Instead of pasting a calendar_id, a FlowApp connects to your account and renders a Dropdown of your actual calendars.
- **Deep Validation**: The node knows exactly what data is required (e.g., Email, Date) and validates your script configuration before you deploy.
*Documentation:* User Guide: Using FlowApps: https://docs.iqra.bot/build/script/flowapps

**3. Open Plugin Architecture**
The community builds faster than any single company.
FlowApps are built on a "Code-First, Schema-Backed" architecture. You can write your own plugins in .NET/C#, define the UI schema, and contribute them back to our Open Source core.
*Documentation:* Developer Docs: https://docs.iqra.bot/developers/flowapp


### G. Global Edge (Latency)
Multi-Region Edge Architecture with Semantic Caching

**Deep Dive:** Latency kills conversation. A delay of 2 seconds breaks immersion. Iqra AI combats physics through Multi-Region Architecture, Co-Location, and Semantic Caching.

**1. The Physics of Latency**
If your user is in London, your server in New York, and your LLM in California, audio packets cross the Atlantic Ocean four times per turn. This adds ~400ms of unavoidable lag.
> The Solution: Iqra AI allows you to spin up processing nodes in specific geographies (e.g., EU-Central). We ingest the call in London, process in Frankfurt, and keep data on the continent.

**2. Service Co-Location**
Low latency requires aligning the entire stack. Don't mix a US Voice Provider with a European LLM.
Deploying for the Middle East? Use Azure OpenAI (UAE North) + Azure Speech (UAE North). Keep the data path local.
*Documentation:* Read Optimization Guide: https://docs.iqra.bot/build/agent/latency

**3. Semantic Caching**
The fastest request is the one you don't make.
If a user asks a common question ("What are your hours?"), we serve the pre-generated audio from our Edge Cache instantly. This skips the costly STT -> LLM -> TTS roundtrip entirely.

---

## 5. Ecosystem (Integrations)
We are vendor-agnostic. Connect your preferred LLMs, Voice providers, and business tools in seconds.

### Supported Providers (Native)
- **LLM Models**: Anthropic (Claude), Groq, Google Gemini, Azure OpenAI, OpenAI.
- **Voice Synthesis (TTS)**: ElevenLabs, Deepgram Aura, Resemble AI, Hamsa, Neuphonic.
- **Transcribers (STT)**: Deepgram Nova 2, Azure Speech Services.
- **Telephony**: Twilio, Telnyx, Vonage.
- **FlowApps**: Cal.com, HubSpot (and Custom).

*Link:* View Library: https://docs.iqra.bot/integrations

---

## 6. Pricing & Plans

| Feature | Free | Build | Scale | Enterprise |
| :--- | :--- | :--- | :--- | :--- |
| **Price (Monthly)** | $0 | $50 | $500 | Custom |
| **Core Platform** | | | | |
| Visual Builder | ✅ | ✅ | ✅ | ✅ |
| FlowApps & Action Flows | ✅ | ✅ | ✅ | ✅ |
| Whitelabeling | ❌ | ✅ | ✅ | ✅ |
| **Usage & Limits** | | | | |
| Per Minute Rate | $0.05/min | 500 min Included; then $0.045/min | 5,000 min Included; then $0.04/min | Custom |
| Concurrency | 5 Included; Add-on ($10/mo) | 10 Included; Add-on ($10/mo) | 55 Included; Add-on ($10/mo) | Unlimited Concurrency |
| Voicemail Detection | $0.008/per | 150 Included; then $0.008/per | 1,500 Included; then $0.008/per | Custom |
| TTS Cache Storage | 128MB Included; Add-on ($1/128MB/mo) | 512MB Included; Add-on ($1/128MB/mo) | 2GB Included; Add-on ($1/128MB/mo) | Custom |
| TTS Cache Retrieval; Concurrency | Free Currently | Free Currently | Free Currently | Free Currently |
| Knowledge Base Storage | 50K Vectors; Add-on ($5/10K/mo) | 150K Vectors Included; Add-on ($5/10K/mo) | 500K Vectors Included; Add-on ($5/10K/mo) | Custom |
| Media Storage | 1GB Included; Add-on ($0.25/GB) | 3GB Included; Add-on ($0.25/GB) | 7GB Included; Add-on ($0.25/GB) | Custom |
| Media Retrieval Bandwidth | 2GB Included; Add-on ($0.6/GB) | 6GB Included; Add-on ($0.6/GB) | 15GB Included; Add-on ($0.6/GB) | Custom |
| **Support & Deployment** | | | | |
| Discord Community | ✅ | ✅ | ✅ | ✅ |
| Email / Ticket Support | ❌ | ✅ | ✅ | ✅ |
| Dedicated Slack | ❌ | ❌ | ✅ | ✅ |
| Deployment Model | Multi-Region Cloud | Multi-Region Cloud | Multi-Region Cloud | On-Premise / Private Cloud |
| **Security & Compliance** | | | | |
| SOC 2 / ISO 27001 | ✅ | ✅ | ✅ | ✅ |
| HIPAA | ❌ | ❌ | ❌ | ✅ |

### FAQ
- **Q:** Do you charge for Telephony (Twilio) minutes?
  - **A:** No. You bring your own carrier (BYOC). You pay Twilio directly for telephony, and you pay us for the AI compute time.
- **Q:** What counts as a "minute"?
  - **A:** Billing is calculated based on the total duration the agent is active on a call or web session, rounded to the nearest second.
- **Q:** Is Whitelabeling included in all plans?
  - **A:** Whitelabeling is included for free starting from the Build plan. The Free tier does not include whitelabeling capabilities.
- **Q:** Can I switch plans later?
  - **A:** Absolutely. You can start on Free and upgrade to Build or Scale as your needs grow. Upgrades are prorated.

---

## 7. Open Source & Deployment
We believe critical infrastructure should be transparent. Audit our security, host on your own servers, or contribute to the core.

- **Data Sovereignty**: Keep data on-premise.
- **No Lock-in**: Own the binary.
- **Self-Hosting Guide**: https://docs.iqra.bot/developers/self-hosting

---

## 8. Contact & Company
- **Book a Demo**: https://cal.com/badal/iqra-ai
- **Contact Sales**: contact@iqra.bot
- **Join Community**: https://discord.gg/UkKHtmqmMH
- **Address**: Abdullah bin Amir Trading SPC, Badal Technologies™, PO Box 112, Muscat, Sultanate of Oman (OPERATED BY BADAL TECHNOLOGIES)