Local Agentic AI
An on-prem agentic AI platform running on NVIDIA DGX Spark architecture. Reads your Unified Namespace in real time, calls MCP tools across the MFX platform, and serves up to 8 concurrent users per box. Available via a land-and-expand trial — we install it, you keep it if it delivers.
One Command Loads the Schedule
Point SparkHub7 at any MFX workspace directory and it parses the schedule file, counts manufacturing orders and runs, and confirms what it loaded — before you ask the first question.
Schedule Gaps Surfaced Instantly
Ask for schedule improvement advice and SparkHub7 returns ranked, specific findings — material shortfalls by SKU with exact quantities, work center utilisation rates, and underutilised capacity — all derived from the loaded schedule.
Live MFX Data, One Question Away
SparkHub7 queries MFX modules via MCP — ask about the latest shipping order for any company and site and get the exact SSM record back: order number, product, quantity, customer, and delivery status.
Cross-Module Queries in Plain English
Ask about the most recent Integrated Planning run and SparkHub7 returns the plan ID with timestamp, the product being produced, and the customer order it will fulfil — spanning PPM planning and order data in one question.
Real-Time Equipment Health, On Demand
SparkHub7 subscribes to the Unified Namespace and can spawn a continuous monitoring agent — ask for process train status and get a timestamped table of every stage, asset health, and a specific advisory warning where action is needed.
What It Does
SparkHub7 is the agentic AI layer of MFX — an on-premises conversational intelligence system that connects to every MFX module via MCP, subscribes to the plant's Unified Namespace in real time, and reasons over operational data, sensor streams, and supplemental documentation in a single interface. Ask it about production state, quality results, or logistics; it answers from live data.
MFX & Third-Party Data via MCP
SparkHub7 connects to all subscribed MFX modules through MCP Servers — Order Fulfillment, Production Scheduling, Execution, Quality Control, and Logistics. The same MCP layer extends to third-party data sources, allowing SparkHub7 to reason across any system. Data can be contextualized at the source or through supplemental documentation in the RAG database.
Unified Namespace Subscriptions
SparkHub7 sees the current execution state of the enterprise and every connected site through the Unified Namespace. It supports both broad and targeted MQTT topic subscriptions — monitoring a single sensor or the entire plant floor from the same interface — and uses its built-in basic historian or external historian integration for trend analysis over any metric.
Agent Orchestration
SparkHub7 spawns AI agents from natural language prompts — either single-task agents or continuous monitoring modes — without any configuration. It also manages OpenClaw agents in a NemoClaw sandbox running on cloud or on-premises LLMs, supporting enterprise-grade agentic workflows at any scale.
Documentation & Knowledge Graph
Ingest spec sheets, SOPs, reports, and any factory documentation directly into SparkHub7's RAG and knowledge graph databases. Ingested documents guide further analysis of live operational data — the same session that queries a live production order can reference an internal product specification or a historical incident report.