# How to Add a Private AI Layer to DELMIAworks (IQMS) On-Prem > Source: https://netray.co/erp-help/private-ai-layer-delmiaworks-iqms-on-prem-architecture > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: ERP Help ## Short answer DELMIAworks, formerly IQMS EnterpriseIQ, runs on an Oracle database that you can read through a standby or replicated copy. A private AI layer queries well-known tables such as ARINVT for items, indexes documents locally, serves an open-weight model on your own GPUs, and writes back only through the vendor's supported interfaces after a person approves. ## Kind advanced ## System delmiaworks ## Erp DELMIAworks / IQMS EnterpriseIQ ## Module AI integration / Oracle database and web services ## Query how to add AI to DELMIAworks IQMS EnterpriseIQ on-prem ## Alternate Queries - DELMIAworks AI integration without cloud - IQMS Oracle database AI assistant - private LLM for DELMIAworks ERP ## Applies To On-premises DELMIAworks / IQMS EnterpriseIQ on Oracle Database; confirm your release and Oracle version, and note that cloud-hosted DELMIAworks is managed by the vendor ## Steps ### Steps to add a private AI layer to DELMIAworks / IQMS EnterpriseIQ **Items:** - Confirm the Oracle version and edition, the DELMIAworks release, and whether you hold licences for Data Guard, Active Data Guard or GoldenGate. - Create a read-only copy: Active Data Guard standby if licensed, otherwise a periodically refreshed copy through RMAN duplicate or Data Pump exports. - Create a dedicated Oracle user with SELECT on a curated set of objects or on views in its own schema, never on the full IQMS schema. - Define views for items (ARINVT), customers (ARCUSTO), orders, work orders, purchasing and inventory movements, with documented columns and status meanings. - Index work instructions, quality records, drawings and customer documents in a local vector store, tagged by plant, customer and document class. - Serve an open-weight model with vLLM or Ollama on local GPUs behind a tool server with an audit log. - Map DELMIAworks users and groups to roles and apply row and column scope through views or the tool server. - Phase the rollout from read-only questions to drafts to approved write-backs through supported vendor interfaces. ## Sections ### How DELMIAworks data can be reached **Body:** DELMIAworks is built on an Oracle database, with application objects in an IQMS schema. Long-standing tables include ARINVT for inventory items, ARCUSTO for customers, and ORDERS, PO and WORKORDER families for sales, purchasing and production. Column names should be verified against your release, but the Oracle foundation means standard SQL clients, JDBC and ODBC all work with a suitable account. Oracle gives you strong options for isolation. A standby opened read-only (Active Data Guard) or a GoldenGate-fed copy keeps analytical load off production. Views in a dedicated schema, with SELECT granted only on them, are the cleanest way to restrict columns. For changes, use the vendor's supported interfaces and screens. DELMIAworks has substantial business logic for production, costing and plant scheduling, and direct DML against its tables is risky. **Code:** -- Read-only query against the item master (verify column names in your release) SELECT a.itemno, a.descrip, a.class, a.rev FROM iqms.arinvt a WHERE a.itemno LIKE 'AX-%' FETCH FIRST 50 ROWS ONLY; -- Dedicated user with access to curated views only CREATE USER ai_reader IDENTIFIED BY ""; GRANT CREATE SESSION TO ai_reader; GRANT SELECT ON ai_views.v_items TO ai_reader; GRANT SELECT ON ai_views.v_open_orders TO ai_reader; ### Reference architecture **Body:** Point the assistant at a read-only Oracle copy, and put a tool server in front of it. The tool server offers parameterised reads such as stock by item and warehouse, open orders by customer, work order status by press or line, and purchase order lines past due. For plastics and process manufacturers that use DELMIAworks, add reads for tooling and production monitoring data where your release exposes it. The retrieval layer indexes documents and long-text fields separately from the transactional data. Oracle 23ai includes vector search, but a separate vector store such as pgvector or Qdrant keeps the AI workload off the ERP database and works on older Oracle versions too. For air-gapped or sovereign deployments the stack, including embeddings and the audit log, stays on site. The architecture supports a customer's own controls on controlled technical data; it does not assert any certification. ### Permissions and data scope **Body:** DELMIAworks maintains users, groups and plant-level rights inside the application, whereas the assistant reads through an Oracle account. Rebuild the relevant scope in views and the tool server: identify the person, find their DELMIAworks groups and plants, and allow only the corresponding views and rows. Oracle Virtual Private Database can add row-level predicates keyed to a session context, which fits multi-plant companies. Cost, price and margin columns remain restricted to finance and management roles, and the document index is filtered on plant and customer tags before retrieval. Test with users from two plants and with a supervisor versus a controller, and confirm that results differ as the application would. ### Write-back, sizing and phased rollout **Body:** Write-back uses vendor-supported routes only, and a user approves the exact change. In early phases, have the assistant prepare the transaction and let a user enter it in DELMIAworks. A mid-size manufacturer with 100 to 400 named users typically needs one to two 48 GB class GPUs for a 14B to 32B parameter model, plus a modest CPU server for the vector store. Document volume, especially scanned quality records, sets index size. Phase one: inventory, order and production status for 4 to 6 weeks. Phase two: document Q and A and drafts. Phase three: a short, reviewed list of approved write-backs. ### Honest limits **Body:** Oracle licensing is the main trap: Active Data Guard and GoldenGate are separately licensed options, so confirm entitlements or use a refreshed copy. Real-time production monitoring data may be high volume and is better summarised than fed raw to a model. If the vendor offers an AI feature in your release, test it first; it may cover simple cases. A private layer is worth building when you need cross-system questions, document retrieval and control over where the model runs. ## Pitfalls - Granting SELECT on the whole IQMS schema instead of curated views. - Assuming Oracle replication options are licensed when they are separate options. - Feeding raw machine-monitoring streams to the model instead of summaries. - Skipping plant-level scope in a multi-plant deployment. - Writing to application tables with SQL and bypassing production and costing logic. ## Ai Angle ERPray can connect to an Oracle-based ERP through the read-only pattern described here and show the query behind each answer. Netray's custom build service handles the views, tool server and document index for DELMIAworks, with the model on your own hardware. ## Faq ### What database does DELMIAworks use? **Answer:** On-premises DELMIAworks, formerly IQMS EnterpriseIQ, runs on Oracle Database. Confirm the Oracle version and your licensed options before choosing a replication approach. ### Do I need Active Data Guard? **Answer:** Not strictly. It is a convenient way to get a read-only standby, but it is a licensed option. A periodically refreshed copy through RMAN or Data Pump works for many status and planning questions. ### Can I use Oracle 23ai vector search? **Answer:** Yes, on that version, but a separate vector store keeps the AI workload off the ERP database and also works with older Oracle releases. Choose based on your version and operations team. ### Does it work for multi-plant companies? **Answer:** Yes, if plant scope is enforced in views, Oracle VPD or the tool server, and mirrored as metadata filters on the document index. ## Search terms DELMIAworks AI, IQMS EnterpriseIQ AI, DELMIAworks Oracle database AI, ARINVT query, private LLM DELMIAworks, DELMIAworks on-prem AI assistant, air-gapped ERP AI Oracle, IQMS ERP integration AI --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json