​

How It Works

The Zild Platform operates as an AI Agent Management Platform (AMP) — a centralized orchestration and governance layer that connects generative AI models, communication channels, workflows, and enterprise systems. Zild is designed to transform AI agents into managed, secure, and scalable enterprise assets. At a high level, the execution flow follows this structure: Channel → App → Agent → Conversation Engine → Workflow Engine → Integrations → Response Each layer has a defined responsibility, enabling modularity, observability, and enterprise-grade control.

---

​

1. Channel Layer

Channels are the external communication interfaces where users interact with AI agents. Currently supported channel types include:

• WhatsApp
• Voice (SIP / Twilio)
• Custom API endpoints

Zild applies channel abstraction, meaning agent logic is completely decoupled from the communication provider. The same agent can operate across multiple channels without duplicating business logic.

---

​

2. App Layer

An App represents a channel integration view for end users. We are commited to create a Zild App when the users needs a different inferface to interact with AI. Zild Apps:

• Zild Assist: For chat conversations
• Zild Voice: For call conversations
• Zild Insight: For QA the conversations
• Zild Coach: To train the users using voice
• Zild Docs: To process documents
• Zild Chat: Chat interface between users and AI Agents
• Zild Tasks: For background jobs
• Zild Admin: For manage all the AI usage of company

---

​

3. Agent Layer (AI Orchestration Layer)

The Agent is the managed AI entity governed by the AMP. When an event arrives, the platform:

1. Identifies the Tenant
2. Identifies the App
3. Resolves the configured Agent
4. Loads conversation state
5. Executes AI reasoning

Agents operate with:

• System instructions
• Knowledge base
• Tool definitions
• Escalation rules
• Guardrails
• Conversation memory

An Agent can:

• Generate natural language responses
• Execute structured tools
• Trigger workflows
• Call external APIs
• Route to human operators
• Update conversation state

Infrastructure services are stateless, but conversations are stateful. This separation enables horizontal scalability.

---

​

4. Conversation Engine

Every interaction belongs to a Conversation object. The Conversation Engine manages:

• Context persistence
• Message history
• Metadata
• Event logging
• Workflow triggers
• Audit records

This enables:

• Context-aware responses
• Multi-step flows
• Long-running interactions
• Operational traceability

Conversation data is fully isolated per tenant.

---

​

5. Workflow Engine (Operational Layer)

Zild separates AI reasoning from operational automation. After AI reasoning completes, structured workflows can be triggered. Workflows may be activated by:

• Agent decisions
• Conversation events
• Status changes
• External callbacks

Workflows can:

• Send webhooks
• Update CRM systems
• Create tasks
• Send contracts
• Assign internal teams
• Route leads
• Escalate conversations

Workflows are deterministic and auditable. This ensures compliance, transparency, and enterprise control.

---

​

6. Integration Layer

Zild integrates with enterprise ecosystems via:

• REST APIs
• Secure webhooks
• CRM platforms
• Telephony providers
• Messaging providers
• Internal enterprise systems

Integrations are configured per tenant and governed through:

• API authentication
• Signature verification
• Role-based permissions
• Access isolation

Zild acts as the secure intermediary between AI agents and enterprise systems.

---

​

7. Response Delivery

After AI reasoning and workflow execution:

1. The response is formatted internally
2. The App converts it into channel-specific format
3. The message or action is delivered to the end user

For Voice channels, this may include:

• Text-to-Speech generation
• Call transfers
• DTMF capture
• SIP routing

For messaging channels, it may include:

• Structured messages
• Media attachments
• Interactive elements
• Templates

All outbound communication passes through the App abstraction layer.

---

​

End-to-End Execution Example (WhatsApp SDR)

1. A user sends a WhatsApp message
2. The WhatsApp provider triggers a webhook
3. The Zild App validates and normalizes the payload
4. Tenant and Agent are resolved
5. Conversation memory is loaded
6. The AI Agent performs qualification reasoning
7. A Workflow is triggered to update the CRM
8. A webhook sends structured lead data
9. The Agent generates a contextual response
10. The message is formatted and sent back to WhatsApp
11. Conversation state and audit logs are persisted

All operations occur inside tenant isolation boundaries.

---

​

Multi-Tenant AMP Architecture

Zild is natively multi-tenant. Each tenant has:

• Isolated data storage
• Independent agents
• Independent apps
• Independent workflows
• Independent integrations
• Segregated authentication boundaries

There is no shared state between tenants. This allows Zild to support:

• Mid-market companies
• Enterprise deployments
• Multi-department structures
• White-label environments

---

​

Scalability Model

Zild is designed for horizontal scalability. Key architectural characteristics:

• Stateless application services
• Persistent conversation state
• Asynchronous event processing
• Webhook-based integrations
• Channel abstraction
• AI orchestration decoupled from transport layer

This enables:

• High message throughput
• Parallel AI execution
• Multi-channel concurrency
• Enterprise-grade reliability

---

​

Governance and Control (AMP Capabilities)

As an AI Agent Management Platform, Zild enforces:

• Role-Based Access Control (RBAC)
• Tenant-level isolation
• Secure API authentication
• Webhook signature verification
• Conversation logging
• Event auditing
• Workflow traceability
• Managed agent lifecycle

This ensures AI agents operate within enterprise compliance boundaries.

---

​

Summary

Zild separates responsibilities into structured layers:

• Channels manage communication
• Apps normalize and authenticate
• Agents perform AI reasoning
• Conversations manage state
• Workflows execute operations
• Integrations connect enterprise systems

This modular AMP architecture allows organizations to:

• Deploy AI once
• Govern centrally
• Integrate deeply
• Scale safely

Zild turns generative AI into an enterprise-managed operational layer.

---

Next, continue to:

• Core Concepts — Object model and terminology
• Platform Overview — AMP positioning and architecture
• Quickstart Guide — Deploy your first AI agent