I’ve been working on Noviz AI, a developer-focused platform for connecting AI agents with enterprise ERP systems.
The idea behind Noviz is simple:
Instead of building a separate AI integration from scratch for every ERP, provide a reusable AI-agent and ERP connector architecture that developers can extend and customize.
ERPNext is the first reference implementation.
What is Noviz AI?
Noviz is an ERP AI agent architecture and complete ERP connector source-code platform, built initially for ERPNext.
It connects an AI agent with ERP business data, APIs, workflows and operations, allowing users to interact with their ERP through natural language while the agent works with the underlying business system.
The architecture is designed to be extended beyond ERPNext to other ERP platforms.
The long-term vision is to provide a common architecture for building AI agents that can operate across different enterprise systems.
Why Noviz?
ERP systems contain the core operational information of a business:
- Customers
- Sales orders
- Invoices
- Products
- Inventory
- Suppliers
- Employees
- Accounting
- CRM
- Projects
- Workflows
- Reports
But accessing this information normally requires users to navigate through ERP interfaces, reports, filters and multiple screens.
An AI agent can provide a different interface.
Instead of:
Open ERP → Find module → Select filters → Find records → Read report
A user can ask:
"Which customers have overdue invoices?"
or:
"Show me the sales orders that are pending delivery."
or:
"Which products are currently low in stock?"
The agent can understand the request, retrieve the appropriate business context and interact with the ERP through its available tools and APIs.
Noviz architecture
The core Noviz architecture can be viewed as:
User
│
▼
AI Agent / LLM
│
▼
Noviz Agent Layer
│
┌───────────┼───────────┐
▼ ▼ ▼
Context Tools Workflows
│ │ │
└───────────┼───────────┘
▼
Noviz ERP Connector
│
▼
ERPNext
│
▼
Enterprise Data
The important part is that the LLM is not directly connected to the database and allowed to do whatever it wants.
Noviz provides an intermediate architecture where the agent can work through defined context, tools, APIs and business operations.
ERPNext is the reference implementation
I chose ERPNext as the initial ERP implementation because it provides a strong open-source ERP environment with APIs, business modules and structured enterprise data.
Noviz can use ERPNext as the underlying business system while providing the AI-agent layer above it.
This creates a practical reference implementation that demonstrates how an AI agent can work with a real ERP rather than just answering questions about static documents.
You can explore the Noviz implementation and live demonstration at:
From AI chatbot to ERP agent
There is an important difference between an AI chatbot and an ERP AI agent.
A chatbot might answer:
"What is the status of my orders?"
An ERP agent needs to actually understand the ERP context, retrieve the relevant records and potentially perform actions.
The Noviz architecture is therefore built around:
Intent → Context → Tools → ERP → Result
For more complex workflows, this can become:
Understand → Plan → Retrieve → Execute → Verify → Report
That is where I see the real opportunity for enterprise AI agents.
Designed to extend beyond ERPNext
Although ERPNext is the current implementation, Noviz is not intended to be limited to ERPNext.
The connector architecture is designed so that the ERP-specific integration can be replaced or extended.
Conceptually:
Noviz AI Agent
│
┌─────────┴─────────┐
│ │
ERP Connector Layer Business Context
│
┌────────┼────────┬─────────┐
▼ ▼ ▼ ▼
ERPNext SAP Odoo Custom ERP
This means developers and ERP companies can potentially adapt the Noviz architecture to their own ERP environment.
Complete source code
One of the main goals of Noviz is to make the architecture available as a complete source-code product.
Rather than providing only an API or hosted AI assistant, Noviz is intended for developers, ERP companies, system integrators and organizations that want to own and customize the underlying implementation.
The source-code architecture can be extended for:
- Custom ERP integrations
- Additional AI models
- New ERP tools
- Business workflows
- Custom context providers
- Enterprise authentication
- Permission-aware operations
- Industry-specific agents
- Custom automation
The ERPNext implementation serves as the starting point and reference architecture.
Who is Noviz for?
I see Noviz being useful for several groups:
ERP developers
Developers who want to add AI capabilities to an existing ERP without designing the entire agent architecture from zero.
ERP implementation companies
Partners and consultants who want to build AI capabilities into ERP deployments for their customers.
Software companies
Companies that want to create their own ERP-specific AI products using a reusable architecture.
Enterprises
Organizations that want greater control over their AI-agent infrastructure and source code.
The bigger idea
I don't see enterprise AI as simply putting a chatbot on top of an ERP.
The bigger opportunity is an AI intelligence layer for enterprise software.
That layer should be able to understand business context, interact with enterprise systems, use tools, follow permissions, execute workflows and provide traceability.
Noviz is my attempt to build that architecture.
ERPNext is the first implementation. The broader goal is an extensible ERP AI-agent platform that can be adapted to different enterprise systems.
I'm interested in hearing from developers working on AI agents, ERP integrations, RAG, tool calling, enterprise automation and agent architectures.
I'd especially like to know:
What would you add to an architecture like Noviz before deploying it in a real enterprise environment?
Top comments (0)