Manufacturing SystemsERP Platform

PlanetTogether + private AI

AI for PlanetTogether APS: Explain the Schedule, Compare Scenarios, Reconcile with the ERP

Short answer

AI for PlanetTogether adds a private, grounded assistant next to the APS so planners can ask why a job moved, which scenario protects a due date, and where the ERP and the schedule disagree. It reads scheduling data and ERP data through governed connectors, shows the query behind each answer, and leaves the optimizer itself and any publish step under planner control.

ERP: PlanetTogether
Industries: Manufacturing, Process Manufacturing, Aerospace, Electronics
Written for: Production Planner

Discuss your PlanetTogether environment See use cases

PlanetTogether sits between the ERP and the shop floor. The ERP holds sales orders, production orders, routings and inventory; PlanetTogether imports that demand, builds a finite-capacity schedule with its optimization rules, and the planner publishes the result back or hands it to the floor. The planner lives inside scenarios: a live plan, a what-if for a rush order, a version with an extra shift, and a baseline for comparison.

The hard part of the job is rarely pressing Optimize. It is answering the questions that follow. Why did this job slip two days since yesterday? Which resource is the real constraint this week? Which late sales orders trace back to the same missing component? Is the ERP due date still reachable under the scenario the plant manager prefers? Those answers sit across the schedule, the imported ERP data, and the planner's memory of what rules were bent for which customer.

PlanetTogether's own features cover much of the visual and rule-driven side well, and many planners never need more. A private AI layer earns its place when the questions are explanatory and cross-system, when the plan must be reconciled against live ERP or MES data, and when the manufacturer cannot send order, customer or routing data to a public AI service because of contracts or export control.

This page describes what AI around PlanetTogether realistically does, how it connects, and the questions to put to any vendor before an AI agent gets near your scheduling data. It does not replace the optimizer. The schedule remains the output of PlanetTogether's own logic, and the AI explains, compares and flags.

What usually gets in the way

The problems we hear most from production planner teams running PlanetTogether.

Schedule changes are hard to explain

After a re-optimization, dozens of activities may have moved. Planners spend real time reconstructing why, then relay it to sales and production managers by email or chat, often from memory.

Scenario sprawl

Copies of the live scenario for rush orders, overtime and maintenance windows pile up. Comparing two scenarios on late jobs, setup time and resource load is possible but slow, so people compare fewer options than they should.

ERP and plan drift apart

Between imports, production orders are released, quantities change and routings are edited. The plan on screen and the ERP no longer agree, and finding the exact differences means exports and spreadsheets.

Tribal knowledge in the rules

Optimization rules, resource capabilities, setup matrices and custom fields carry years of decisions. When the one planner who understands them is away, nobody can say why a job was sequenced behind another.

Scheduling data is sensitive

Orders, customer names, lead times and routings reveal your business. Defense and aerospace programs often forbid copying that data into a public AI tool.

Where AI earns its place in PlanetTogether

Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.

Schedule change explanation

After an optimize run or an import, the assistant compares the previous and current scenario and explains the biggest movements: a resource went down, a material was marked late, a higher-priority job was inserted.

Touches: Jobs, Manufacturing Orders, Operations, Activities, Resources, Scenario history

Outcome: Turns a half-hour reconstruction before the daily production meeting into a short reviewable summary the planner confirms.

Scenario comparison

Ask which of three what-if scenarios has the fewest late jobs, the lowest total setup time and the least overtime, and see the numbers side by side with the jobs that differ.

Touches: Scenarios, Jobs, Resources, Capacity Intervals, Setup data

Outcome: Lets planners evaluate more alternatives per decision, with the evidence visible rather than recalled.

Late order root cause

For each late sales order, trace through the job, operations, bottleneck resource and material availability to state the likely cause in plain language.

Touches: Sales Orders, Jobs, Operations, Inventory, Purchase Orders

Outcome: Gives sales a defensible reason and a realistic date, and shows the planner which constraint is worth fixing.

ERP versus plan reconciliation

Compare released production orders and quantities in the ERP with the imported manufacturing orders in the schedule and list the differences with the likely reason.

Touches: ERP production orders, BoM, routing, Manufacturing Orders, import tables

Outcome: Catches stale or duplicated orders earlier, before the floor works to an outdated plan.

Constraint and capacity Q&A

Natural-language questions such as which resource is above 95 percent loaded next week, or what an extra shift on a given resource would pull in.

Touches: Resources, Capacity Intervals, Departments, Plants, Activities

Outcome: Moves capacity conversations from screenshots to answers tied to the scenario.

Planner knowledge capture

Index optimization rule settings, capabilities, setup matrices and planner notes so a new planner can ask why a resource is restricted or why a customer's jobs get priority.

Touches: Optimization rules, Capabilities, Custom fields, planner notes

Outcome: Shortens onboarding and reduces dependence on one experienced planner.

Shop-floor status summary

Combine schedule dispatch lists with ERP or MES completion data so a supervisor can ask what is behind on a line this shift and what is next.

Touches: Dispatch lists, Activities, ERP shop-floor transactions, MES events

Outcome: Reduces trips to the planner's desk for routine status questions.

Reference architecture

The AI layer reads from PlanetTogether data and the ERP through governed, read-only connectors. It answers questions and drafts explanations; it does not run the optimizer or publish a schedule on its own.

1
Scheduling and ERP connectors
Reads schedule data from the PlanetTogether database or its export and import interfaces, and ERP data through the ERP's API or a read replica, with a scoped service account.
2
Semantic layer
Maps jobs, manufacturing orders, operations, resources, capabilities and scenario names to the terms your planners use, including custom fields added during implementation.
3
Model serving
An open-weight model (Llama, Qwen, Mistral or gpt-oss class) served with vLLM or Ollama on customer GPUs or a private cloud, sized for planner-scale concurrency.
4
Retrieval and agents
Retrieval over live data and planner documentation grounds each answer; any suggested action, such as an adjusted priority, is a proposal a planner approves in PlanetTogether itself.
5
Governance and audit
Every question, query and answer is logged against the requesting user. Customer and program data stay inside your boundary.

Integration notes for your ERP team

PlanetTogether imports ERP data into its own model and publishes results back; the AI layer reads both sides and should be pointed at the same staging tables or interface your integration already uses, not a new side channel.
Read scheduling data from a restricted database role or exported snapshots; avoid giving the AI any account that can change the live scenario.
Scenario names and version timestamps are indexed so questions can specify which scenario is meant. Without this, answers about the plan are ambiguous.
Custom fields and user-defined data added by the implementation partner are mapped in the semantic layer, otherwise rules specific to your plant are invisible to the assistant.
ERP-side reads use the ERP's documented API or a read replica (for example SyteLine IDOs, SAP OData, or NetSuite SuiteQL), respecting the ERP's own role model.
Write-back is limited to drafts: a proposed priority change or note is queued for a planner, who applies it inside PlanetTogether and publishes as usual.
Plan for version differences: the connector is tested against your PlanetTogether release and re-tested at upgrade, as with any integration.

Deployment options

Air-gapped on-prem

Defense, aerospace and electronics plants whose customer contracts forbid external AI services.

Model, retrieval and connectors run on customer hardware beside the scheduling server with no outbound path.

Private or sovereign cloud

Manufacturers already hosting PlanetTogether and the ERP in their own cloud tenant.

Deployed in the same VPC or region so schedule and ERP data do not cross a jurisdiction boundary.

Hybrid

Multi-plant groups with a central planning team and local ERPs.

A central governance layer queries each plant's scheduling and ERP sources, leaving data in place.

Compliance and data control

How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.

ITAR / EAR data handling

Because the model and data run in an environment the customer controls, the customer can keep controlled technical data and program schedules inside access-controlled infrastructure and apply their own US-person access rules.

Customer NDAs and program flow-downs

Order, routing and delivery data never reaches a public AI API, which is the usual condition in prime contractor terms.

AS9100 and ISO 9001 records

Schedule explanations and reconciliations are logged and reproducible, supporting the control of planning records during audits.

GDPR and regional data protection

Operator and customer personal data stays on infrastructure you control, simplifying transfer assessments.

Where Netray fits

ERPray

Planners and supervisors get grounded natural-language questions and dashboards across the ERP side of the picture: orders, inventory, purchasing and shortages, alongside schedule snapshots.

DataRay

Where scheduling databases, exports, planner spreadsheets and rule documents need to be queried together on-prem, DataRay covers the mixed sources.

Custom build

Scenario comparison and change-explanation agents tied to your own rules and custom fields are usually bespoke work.

How an engagement runs

Phase 1 . 2-3 weeks

Discovery

-Map of PlanetTogether data, scenarios and ERP interfaces in use
-Ranked planner questions and pain points
-Access and security review
-Deployment recommendation

Phase 2 . 6-8 weeks

Pilot

-Read-only connectors to a non-production scenario set
-Schedule change explanation and late-order tracing live for a pilot team
-Accuracy review against planner-verified answers
-Logging validated

Phase 3 . 4-6 weeks

Production

-Production cutover with scoped service accounts
-Runbook for adding questions and fields
-Planner training on scenario-aware prompting
-Audit log review process

Phase 4 . Ongoing

Scale

-Additional plants and ERPs
-Approved draft-action workflows
-Model refresh as open-weight options improve
-Quarterly accuracy review

Questions to ask any vendor, including us

A short list that separates real PlanetTogether AI work from a chatbot demo.

Does the AI change the schedule or run the optimizer, or does it only read and propose?
How does it know which scenario a question refers to, and can it compare two scenarios reliably?
Which credentials does it use for PlanetTogether and the ERP, and are they read-only?
How are our custom fields, optimization rules and capabilities represented so answers reflect our plant?
Where does the model run, and can you show us the network path?
How are answers validated against what a planner would find manually?
What happens at our next PlanetTogether or ERP upgrade?
If we stop using you, do we keep the connectors, prompts and logs?

Frequently asked questions

Does AI replace the PlanetTogether optimizer?

No. The schedule remains the output of PlanetTogether's own optimization rules and the planner's decisions. The AI layer explains what changed, compares scenarios, traces late orders and reconciles with the ERP. Anything it proposes is a draft for a planner to apply inside PlanetTogether.

Is a private AI layer worth it if PlanetTogether already has good reporting?

For fixed reports and Gantt views, often not. The value is in ad hoc, explanatory and cross-system questions: why did this move, which constraint is real, where do the ERP and plan disagree. If your planners rarely ask those, built-in tools may be enough.

Can it handle several scenarios and plants?

Yes, provided scenario names and plant identifiers are indexed in the semantic layer so each question is scoped. Multi-plant groups typically start with one plant and extend once answers are verified.

What about data security for defense programs?

The model, retrieval and connectors run on infrastructure you control, so program schedules and customer data are not sent to a public AI service. You apply your own access controls and export-control procedures to the environment.

How does it connect to our ERP?

Through the ERP's documented API or a read replica using a scoped read-only account, so ERP role rules still apply. Typical sources include SyteLine IDOs, SAP OData services, NetSuite SuiteQL, or direct SQL views where policy allows.

Can it write back priorities or dates?

It can draft a proposed change and queue it for a planner, but it does not alter the live scenario or publish. A planner applies the change in PlanetTogether, so existing approval and audit practice stays intact.

How long does a pilot take?

Discovery is usually two to three weeks and a pilot six to eight, depending on the number of ERPs, custom fields and scenarios in scope. A single-plant, single-ERP pilot sits at the shorter end.

Related guides

Forecasting plus a reason why

AI Demand Forecasting for Manufacturers: Better Numbers and an Explanation Planners Can Trust

AI demand forecasting explains ERP forecast changes in plain language, flags exceptions for planners, and handles short-history SKUs without Excel overrides.

SAP PP/MRP + AI agents

AI Agents for SAP Production Planning, MRP, and Exception Triage

AI agents that triage SAP MD04 exceptions, explain predictive MRP shortages, and draft planned-order actions, grounded on your own SAP data, deployed on-prem.

Shop floor + on-prem AI

AI shop floor assistant for operators working inside your ERP

An on-prem AI copilot answers operator questions against your ERP work instructions, travelers, and routings in plain language, at the machine, without a screen full of menus.

RAG + SQL + permissions

A Private LLM Grounded on Your ERP Data

How a private LLM answers questions on your ERP data: RAG plus text-to-SQL, role-based permissions inherited from the ERP, and where each fits.

Ask your ERP anything

Natural Language Query for ERP Data: Ask SAP, Infor, or Oracle a Question in Plain English

See how natural language query over SAP, Infor, Oracle, and NetSuite data works: grounded text-to-SQL, role-based permissions, and a visible audit trail.

Kinaxis concurrent planning + private AI

AI for Kinaxis RapidResponse: Explaining Plans Across Every Connected ERP

Kinaxis RapidResponse plans across ERPs - pair it with private AI to explain scenarios and answer planning questions grounded in your own data on-prem.

Plan it with numbers

Free Tool

Talk it through with an engineer who knows PlanetTogether

Bring one real question your team cannot answer from the ERP today. We will map the data path, the model, and where it runs, and tell you honestly if AI is the wrong tool for it.

Start the conversation See ERPray

All AI for ERP guides