Natural language operational data access
Natural language operational data access
Access ERP data with a simple chat message—no SQL knowledge or IT ticket required.
Problem
ERP systems centralize production, inventory, procurement, and finance data. Reaching that data usually requires system expertise, the right menu path, or a report request to IT. Managers and field teams wait hours or depend on technical staff to understand operational status. Decision speed drops.
Solution with Orbis Studio
Users ask in natural language: "Which line had the biggest production loss last week?" or "Which products dropped below critical stock levels?" The ERP component ingests the message and passes it to an agent.
The Agent interprets the question and issues tool calls to the ERP API. Responses are formatted for humans—tables, summaries, or comparisons—not raw data dumps. Users can ask follow-up questions in the same conversation; the agent keeps context.
Multiple ERP modules—production, procurement, inventory, maintenance—can be unified in one conversational interface. Users do not need to know which module holds the answer.
Business outcomes
- Decision makers access operational data instantly without IT requests; daily and weekly reporting shrinks.
- Non-technical users query the system in plain language; training and adoption costs fall.
- Data silos across departments merge behind one conversational layer.
- Management reporting automates; repetitive query workflows disappear.
- Data access standardizes—everyone reaches the same live source.
Technical details
Orbis Studio components
| Component | Role |
|---|---|
| ERP | Ingests user queries from the ERP channel |
| Agent | Interprets questions, calls tools, formats responses |
| Language Model | Converts ERP responses into natural language |
| Chat Output | Returns answers to users |
Flow architecture
Agent tool configuration
Expose each ERP module as a tool. Add tools as Run Flow components or custom Python components:
| Tool name | Description |
|---|---|
get_production_report | Returns production data for a line and date range |
get_inventory_status | Queries stock levels and critical thresholds |
get_purchase_orders | Lists open purchase orders |
get_maintenance_log | Fetches equipment maintenance records |
Tool descriptions guide the agent to pick the right tool for each question.
Example agent system prompt
_10You are an operational data assistant for a manufacturing plant. Identify the ERP query_10that matches the user's question and fetch answers using connected tools._10_10Always:_10- Respond in clear, concise language_10- Include specifics such as dates, line names, or product codes when available_10- State clearly when a query returns no data and suggest alternatives_10_10Call multiple tools in sequence when the question requires more than one data source.
Multi-turn query example
_10User: Which lines underperformed last week?_10_10Agent: [Calls get_production_report for lines A, B, C]_10 Line 3 fell significantly below average yield last week._10 Would you like me to check maintenance logs?_10_10User: Yes, for Line 3._10_10Agent: [Calls get_maintenance_log for Line 3]_10 Line 3 had two unplanned stops this week. Details below...
The agent keeps context across turns; when the user says "Line 3," it understands the reference from prior messages.
Integration points
- The ERP component accepts messages from SAP, Oracle, Microsoft Dynamics, or other systems. Connect through REST API or database tools.
- Start with read-only tools to avoid unintended writes; add write permissions with HITL Gate approval when needed.