NG Tara: an AI agent for business workflows — how it runs a task across several systems

A large part of a company's work lives in no single system, but between them. A customer sends a message, someone checks the ERP, looks at the warehouse, opens an order, asks the sales manager for approval, then replies to the customer — and notes a reminder to check delivery on Friday. Each of those steps is simple. What is hard is the chain: several systems, several people, deadlines, and rules nobody ever wrote down in one place.

NG Tara is the AI agent for exactly that chain. It turns a request that arrives in a conversation into a task, carries it out in the systems the company already uses, and sees it through to the end — with human confirmation where it is needed.

What NG Tara does

The scope has three parts: where the request comes from, where the task is carried out, and what the agent does with it.

• Channels. Voice, chat, e-mail, Viber, WhatsApp, Instagram and other channels that customers, partners and employees already use.
• Systems. API and MCP integrations with ERP, CRM, WMS, DMS and MES systems — including NG One, NG CRM, NG DMS and NG WMS — the systems where the work actually lives.
• Actions. Searching data, checking conditions, creating documents and orders, coordinating people and systems, tracking deadlines and requesting approval where the rules require it.

None of this replaces an ERP or a CRM. Tara is a layer that works on top of existing systems, through their APIs, not as yet another database that data gets copied into.

What one workflow looks like

The example below is an illustration with made-up data.

1. A customer writes on WhatsApp asking to move a delivery to next Wednesday and add twenty more units of the same item.
2. Tara finds the open order in the ERP and checks the delivery terms in the contract.
3. It checks in the warehouse system whether the extra quantity is in stock.
4. It prepares a change to the order. Because both quantity and date change, company rules send the change to the sales manager for approval.
5. After approval it writes the change to the ERP, and the claim that the change was written comes from the ERP, not from the agent.
6. It sends the customer a confirmation on the same channel and schedules a delivery check for the day before the new date.

Every one of these steps already exists in a company today — a person just does it, switching between applications. The value of an agent is not that it does any single step more cleverly, but that the chain does not break between two steps.

The rules it works by

An agent that carries out actions across several systems has more power than one that only answers, so the rules are stricter. NG Tara works on the same principles described in the piece on an AI strategy for the enterprise:

• It works within permissions. The agent may not see or do more than the rules allow for a given task and user.
• A person approves what is sensitive. Changes to terms, irreversible actions and anything company rules mark as risky wait for human confirmation.
• Every action leaves a trail. Who asked, what was checked, what was written and who approved.
• Autonomy per type of task. The level of independence is chosen for each type of work separately and extended only when results justify it.

The security questions this raises — permissions, instructions injected through messages, guardrails on tools — are covered in the piece on AI agent security.

How it differs from NG Sara and NG Nora

How to start

The business value shows on the first workflow: less retyping and chasing, faster answers to customers and colleagues, and no task left forgotten in someone's inbox. That is why we start with a demo on one of your workflows.

The preparation is simple: an inventory of the workflows that cross several systems, a check of which of those systems have an API, and written-down approval rules. We apply the same approach in bespoke AI automation projects, described on the services page. If you would like a demo on your company's workflows, book it here. The product overview is on the NG Tara page.