IFS Cloud + on-prem AI
AI for IFS Cloud in aerospace and defense manufacturing
Short answer
IFS Cloud programmes in aerospace and defense sit on IFS's projection layer (OData-based) and Aurena UI, with heavy use of project-based manufacturing, configuration management, and document control. A private LLM grounded on those projections, run on infrastructure the programme controls, gives engineers and planners natural-language answers without exporting export-controlled or classified-adjacent data to a multi-tenant AI service.
- ERP
- IFS Cloud, IFS Applications 10
- Industries
- Aerospace, Defense, MRO
- Written for
- CIO
If your IFS Cloud instance runs a defense programme or an aerospace supply contract, the questions you get from planners and engineers are rarely one-click answers inside Aurena. "Which work orders are behind on this project because of a late-arriving part" or "what configuration changes touched this serial number" require joining project, manufacturing, and document data that IFS presents as separate lobbies and pages.
IFS.ai, IFS's own generative AI layer, is improving quickly, but for programmes carrying ITAR technical data, CUI, or a foreign military sales condition, the question is not whether the built-in assistant is capable, it is whether the underlying model call, telemetry, and any cached context stay inside your compliance boundary. Many A&D IFS shops cannot answer that question with certainty for a vendor-hosted feature, which is what pushes them toward a self-hosted option.
IFS Cloud's architecture actually helps here: everything is exposed through typed OData projections, the same interface Aurena itself uses, so a private AI layer can read the identical data model without touching the database directly or maintaining a shadow copy of business logic. That symmetry is unusual among ERPs and worth building on rather than working around.
This page describes what an on-prem or private-cloud AI layer over IFS Cloud looks like in practice for aerospace and defense programmes: what to connect to, what to keep read-only, and where a private LLM earns its keep against project managers, planners, quality engineers, and configuration management staff.
What usually gets in the way
The problems we hear most from cio teams running IFS Cloud.
Programme status buried across lobbies
Project managers piece together programme health from separate IFS lobbies for project, manufacturing, and procurement rather than one grounded answer that cites the underlying projections.
Configuration management questions take specialist time
"What changed on this serial number and when" requires someone who knows the Config Management module and its link to the engineering structure; that knowledge is concentrated in a small team.
Export-controlled and CUI data limits which AI tools are usable
Engineering documents, technical data packages, and configuration records often carry ITAR or CUI markings, which rules out sending them to a public AI API for search or summarisation.
Document-heavy compliance work is manual
AS9100 audits, first article inspection packages, and customer source inspection prep involve pulling documents and records from IFS Document Management by hand.
IFS.ai roadmap uncertainty for regulated programmes
Programme offices want AI value now but are wary of committing workflow to a vendor-hosted assistant whose data handling for classified-adjacent or export-controlled content is still maturing.
Where AI earns its place in IFS Cloud
Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.
Programme status Q&A across project and manufacturing
A natural-language layer over the Project, Manufacturing, and Procurement projections lets a programme manager ask for schedule risk in plain English and get an answer that cites the shop order, purchase requisition, or milestone behind it.
Touches: Project projection, Shop Order, Purchase Requisition, Activity, Milestone
Outcome: cuts the weekly programme status prep from a half-day of screen-hopping to a first draft in minutes, reviewed by the PM before it goes out
Configuration management change history
An agent grounded in Config Management and Engineering Structure data answers "what changed on serial number X and under what change order" with citations back to the change record, useful for customer audits and as-built reconciliation.
Touches: Config Management, Engineering Structure, Change Order, Serial Structure
Outcome: reduces the time to assemble an as-built history for a single unit from hours of manual structure comparison to a reviewed summary
AS9100 and first article documentation assembly
Retrieval over Document Management and Quality projections drafts the document index for an FAI (AS9102) package or a customer source inspection, flagging any missing or superseded documents for a human to resolve.
Touches: Document Management, Quality Inspection, FAI record, Non-Conformance
Outcome: shortens FAI package assembly by finding and cross-checking documents automatically, with the quality engineer confirming completeness
Purchase order and subcontractor follow-up drafting
For long-lead and single-source aerospace parts, an agent drafts the follow-up email or portal message for late purchase orders, with the buyer approving before it sends.
Touches: Purchase Order, Purchase Order Line, Supplier, Requisition
Outcome: keeps routine expedite messages moving without buyer time on every line, freeing buyers for the exceptions that need judgment
Non-conformance and CAPA narrative drafting
Given a non-conformance record, an agent drafts the root-cause section and CAPA narrative from linked inspection and process data, which the quality engineer edits and signs off.
Touches: Non-Conformance, Quality Inspection, CAPA, Process Instruction
Outcome: reduces the drafting time for routine NCR write-ups so quality engineers spend more of the review on the analysis, less on the prose
Inventory and kit shortage explanation
An agent explains why a kit is short for a shop order, tracing back through inventory, purchase order, and BOM data to the specific missing component and its expected arrival.
Touches: Inventory Part, Shop Order Material, Purchase Order Line, BOM
Outcome: gives production control a direct answer instead of a manual trace through inventory and procurement screens
Engineer-to-order estimate support
For proposal teams, retrieval over historical project actuals (labour hours, material cost, change order frequency by project type) supports estimating for a new engineer-to-order bid.
Touches: Project Actuals, Activity Cost, Change Order history
Outcome: gives estimators a faster, evidence-based starting point pulled from comparable past projects rather than starting from a blank sheet
Reference architecture
The layer sits beside IFS Cloud and reads through the same OData projections Aurena uses, rather than querying the database directly, so business rules and access checks are respected by construction.
- 1
IFS connector layer
Calls IFS Cloud's standard OData projections (Project, Manufacturing, Procurement, Document Management, Quality, Config Management) using a service account scoped to the same permission sets a real IFS user would have.
- 2
Data and semantic layer
Maps IFS entities and lookup values (states, codes, custom fields) to a business glossary so the model understands programme-specific terms like FAI, CAPA, or long-lead item without hardcoding logic per programme.
- 3
Model serving layer
Runs an open-weight model (Llama, Qwen, or Mistral class) on vLLM or Ollama inside your network or approved private cloud, with no calls to a public model API by default.
- 4
Retrieval and agent layer
Combines structured OData retrieval with document retrieval from IFS Document Management, and constrains any agent action, such as drafting a supplier email, to a human-approved send step.
- 5
Governance and audit layer
Logs every query, the projections it touched, and the answer given, in a store your compliance team can inspect, distinct from IFS's own audit log but cross-referenceable to it.
Integration notes for your ERP team
- Connect through IFS Cloud's standard OData projections (the same ones Aurena and IFS Connect use) rather than direct database access, so upgrades to newer IFS Cloud releases do not break the integration.
- Use a dedicated service account with role-based permission sets scoped to only the projections the AI layer needs, mirroring least-privilege practice for any other integration user.
- For document retrieval, index IFS Document Management metadata and, where permitted, document text, respecting the same access classes and revision control IFS already enforces.
- Config Management and Engineering Structure data has its own object model in IFS Cloud; map it explicitly rather than assuming it behaves like a flat BOM.
- Any write-back, such as updating a non-conformance record or creating a follow-up activity, should go through the same OData projection with a human approval step before commit, never a direct table write.
- Aurena's Lobby and Insight pages are a UI layer, not a data source; the AI layer should query the underlying projections directly rather than scraping rendered pages.
- For multi-site or multi-company IFS Cloud instances, scope retrieval per company and site to avoid an agent answering with data from a programme the requester should not see.
Deployment options
Air-gapped on-prem
Programmes with ITAR technical data, CUI, or contractual data-residency terms that rule out any external network path.
Model, retrieval layer, and IFS connector all run inside the programme's network, on GPU hardware you own or lease, with no outbound internet requirement for inference.
Private or sovereign cloud
Organisations already running IFS Cloud in a private cloud tenancy or a sovereign region who want to keep AI inference in the same boundary.
The model and retrieval layer run in a VPC you control, connecting to IFS Cloud over a private network path rather than the public internet, with the same access model as on-prem.
Hybrid
IFS Cloud SaaS tenants with a mix of general and export-controlled data who want AI on the non-sensitive majority now.
Non-sensitive projections (general procurement, non-controlled document sets) are indexed for AI use while ITAR- or CUI-tagged data stays excluded until an on-prem path is 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 technical data
Keeping model inference and any retrieval index inside the programme's own infrastructure avoids sending technical data to an external AI provider, which US export control counsel generally treats as a deemed export risk if handled incorrectly.
CMMC 2.0 / NIST SP 800-171
Deploying inside the same enclave that already handles CUI in IFS Cloud means the AI layer inherits existing access control, logging, and boundary protections instead of introducing a new external data flow to assess.
AS9100D traceability
Every AI-assisted answer, document draft, or non-conformance narrative cites the source IFS records it drew from, so a human reviewer can verify it against the same evidence an auditor would check.
UK DEFCON 658 / Cyber Essentials Plus (for UK programmes)
For UK MOD-facing programmes, running inference within the contractor's accredited boundary avoids introducing an unassessed third-party processing step for defense-related data.
Where Netray fits
ERPray
ERPray's connector architecture is a fit for IFS Cloud's OData projection model: natural-language questions and dashboards over Project, Manufacturing, and Quality data, read-only by default, with the underlying query shown.
Custom build
Config Management change-history agents, FAI package assembly, and CAPA narrative drafting are programme-specific enough that most teams start with a scoped custom build before considering a packaged product.
How an engagement runs
Phase 1 . 2-3 weeks
Discovery
- -Inventory of IFS Cloud projections in use and their data sensitivity (ITAR, CUI, unrestricted)
- -Target use cases ranked by programme value and data risk
- -GPU and hosting plan sized to expected concurrent users
Phase 2 . 6-8 weeks
Pilot
- -Working connector to 2-3 priority projections (e.g. Project, Manufacturing, Document Management)
- -One or two agreed use cases (status Q&A, config history, or FAI drafting) live for a pilot group
- -Audit log and access-control review with your security team
Phase 3 . Ongoing after pilot sign-off
Production
- -Rollout to the full programme office or engineering team
- -Expanded projection coverage (Quality, Procurement, Config Management)
- -Documented runbook for model updates and access changes
Phase 4 . Following programme cycles
Scale
- -Extension to additional programmes or IFS Cloud companies
- -Agent capabilities with approval-gated write-back (follow-up emails, draft NCRs)
- -Periodic model and retrieval accuracy review against real usage logs
Questions to ask any vendor, including us
A short list that separates real IFS Cloud AI work from a chatbot demo.
- Where does inference actually run, and can you show the network path technical data would take end to end?
- Does the AI layer read IFS Cloud through standard OData projections, or does it require a database replica that falls out of sync?
- What happens to ITAR or CUI-marked documents if they are accidentally included in a retrieval index; can they be excluded by classification tag?
- Can every AI-generated answer be traced back to the specific IFS records it used, in a form your compliance team can review?
- What is the fallback if the model gets an answer wrong, and how is that logged and corrected?
- How does the solution handle an IFS Cloud version upgrade; does the connector need to be rebuilt?
- Who has access to the query and answer logs, and how long are they retained?
- What GPU hardware and hosting does this require, and who is responsible for keeping it patched?
Frequently asked questions
Can AI work with IFS Cloud without exporting data to IFS.ai or another cloud service?
Yes. IFS Cloud's OData projection layer, the same interface Aurena uses, can be read by a self-hosted model running on your own GPUs or in your private cloud. Nothing has to leave your network, and no call needs to go to IFS.ai or any external AI provider unless you choose to use it.
Is IFS.ai an alternative to a private LLM for aerospace and defense programmes?
IFS.ai is IFS's own generative AI capability and is worth evaluating for non-sensitive workflows. For technical data under ITAR, CUI, or a classified-adjacent handling requirement, most programme security teams want inference to stay inside their own accredited boundary, which points toward a self-hosted model beside IFS Cloud rather than a vendor-hosted assistant.
How does AI handle IFS Cloud's Config Management and Engineering Structure data?
Config Management and Engineering Structure have their own object models distinct from a standard BOM. A grounded AI layer maps these explicitly, retrieving change orders, serial structures, and revision history through the correct projections so answers about what changed on a specific unit are accurate rather than inferred.
What is a realistic first use case for AI on IFS Cloud in an A&D programme?
Programme status Q&A over Project and Manufacturing data is a common starting point: it has clear value for programme managers, touches a bounded set of projections, and does not require write access, which keeps the pilot's risk surface small.
Does this require replacing or duplicating the IFS Cloud database?
No. The recommended approach reads through IFS Cloud's existing OData projections rather than replicating the database, which avoids a second copy of sensitive data to secure and keeps the AI layer consistent with IFS's own access rules.
Can the AI layer draft documents like FAI packages or CAPA narratives, or only answer questions?
Both are workable. Drafting is typically scoped as a human-in-the-loop step: the model assembles a first draft from linked IFS records, and the quality engineer or programme manager reviews and finalises it before it becomes an official document.
How long does a pilot take for AI on IFS Cloud?
A focused pilot on one or two use cases, such as programme status Q&A plus document retrieval, typically runs six to eight weeks after a two to three week discovery phase to map which projections and data sensitivities are in scope.
Related guides
AI for IFS Maintenix in aviation MRO
AI over IFS Maintenix maintenance records, AD/SB compliance, and task cards for MRO, deployed on-prem so tech records never leave your network.
On-prem AI, any ERP, A&DOn-Prem AI for ERP in Aerospace, Defense, and Electronics Manufacturing
A hub guide to on-prem AI across SAP, Infor LN, Costpoint, IFS, and Oracle EBS for aerospace, defense, and electronics manufacturers under ITAR, CMMC, and AS9100.
SAP + on-prem AI for A&DAI on SAP for Aerospace and Defense Manufacturers, Without the ITAR Exposure
AI on SAP S/4HANA or ECC for aerospace and defense manufacturers: serial genealogy, configuration control, and quoting, without technical data leaving your boundary.
EBS + ITAR-aware AIAI for Oracle EBS in Aerospace and Defense Manufacturing
AI on Oracle EBS for aerospace and defense manufacturers: project manufacturing, serial control, and ITAR segregation handled without a cloud dependency.
AS9100D + on-prem AIAI for AS9100 Quality Management on Your ERP
AI on top of your ERP quality module for AS9100D suppliers: NCR/CAPA drafting, FAI support, counterfeit parts screening, with a full audit trail.
ITAR + on-prem AIITAR-Compliant AI for ERP Technical Data
How to add generative AI to your ERP without creating a deemed export under ITAR. On-prem architecture patterns an Empowered Official can sign off on.
Plan it with numbers
AS9100 Audit Readiness Checklist
A 30-point checklist for AS9100 Rev D certification and surveillance audits, weighted toward the aerospace-specific requirements where auditors write the most findings.
Free ToolAir-Gapped AI Readiness Assessment
A 10-question assessment that scores how prepared your organization is to deploy and operate LLMs inside an air-gapped or classified enclave.
Free ToolDefense Contractor AI Readiness Assessment
A 9-question assessment measuring whether your defense manufacturing business can adopt AI productively and compliantly - across governance, data, infrastructure, and skills.
GuideITAR and CMMC Handling of AI Workloads
How ITAR and CMMC apply to AI workloads: technical data boundaries, CUI handling, assessed environments, and where on-prem AI is the only option.
GuideAI in Aerospace MRO Operations: Use Cases and ROI
AI in aerospace MRO operations: predictive maintenance, automated part records, repair quoting, and tech-log analysis. Real use cases, ROI figures, timelines.
GuideOn-Prem AI for Defense Contractors: The Complete Guide
On-prem AI for defense contractors: deploy LLMs and AI agents inside your CMMC and ITAR boundary. Architecture, hardware costs, timelines, and vendor options.
Talk it through with an engineer who knows IFS Cloud
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.