`aidp-fusion-bundle` — Fusion ERP/HCM/SCM → AIDP, batteries included

aidp-fusion-bundle — Fusion ERP/HCM/SCM → AIDP, batteries included

Productizes the official Oracle blog Bring Fusion Data into Oracle AI Data Platform Workbench Using BICC plus the ateam companion How to Extract Fusion Data using Oracle AI Data Platform. One install, three commands, populated lakehouse + installed OAC dashboards.

When to use

• User wants Fusion data in AIDP and asks "where do I start"
• User has BICC privileges and wants curated bronze/silver/gold layers without writing the pipeline
• User is preparing a CFO/analytics demo and needs OAC dashboards on Fusion data
• User wants to use OAC MCP (Preview) to chat with Fusion data via Claude/Cline/Copilot

When NOT to use

• For a single one-off PVO read → use aidp-fusion-bicc (sibling plugin, smaller scope).
• For Fusion REST queries with <50k rows → aidp-fusion-rest.
• For EPM Cloud Planning data slices → aidp-epm-cloud.
• For Essbase MDX → aidp-essbase.

Positioning

This bundle is additive to and complementary with Oracle's managed Fusion data offerings. It productizes Option 1 of pdf1's three-option architecture (BICC into AIDP for "Custom AI and ML, raw data access, data engineering"). Never positioned as a replacement for FDI, OAC, OTBI, BIP, or Data Transforms.

What you get

Mirrors pdf1 §"What Can You Do Once the Data is in Oracle AI Data Platform":

1. Custom ML/AI training on operational ERP/HCM/SCM data (PySpark + Python in AIDP notebooks)
2. Cross-source enrichment — join Fusion data with non-Fusion sources via the aidataplatform connector family
3. Medallion architecture — bronze (raw audit) → silver (typed + dim-joined) → gold (business marts) in Delta
4. GenAI agent grounding — ai_generate("which suppliers had >$1M Q1 spend?") against gold marts via OCI Generative AI
5. BI & reporting via JDBC — OAC, Tableau, Power BI consume the gold layer
6. Delta Sharing (v3 roadmap) — share curated datasets with other teams or external partners

Quickstart

1. Install the CLI on your laptop:

   pip install -e /path/to/oracle-ai-data-platform-fusion-bundle
   

2. Scaffold a bundle in your repo:

   aidp-fusion-bundle init
   

   Edits bundle.yaml and aidp.config.yaml to match your environment (Fusion pod URL, AIDP workspace, OAC URL, OCI Vault refs for credentials).

3. Probe prerequisites against your pod:

   aidp-fusion-bundle bootstrap --check-iam
   

   Confirms BICC role, BICC External Storage profile (set in BICC console), AIDP catalog, IAM policies, Vault access.

4. Run the orchestrator:

   aidp-fusion-bundle run --mode seed     # first-time full extract
   aidp-fusion-bundle run --mode incremental  # daily delta
   

5. Install OAC dashboards (one-time per OAC instance):

   # 5a. Upload bundle-vN.bar to your OCI Object Storage bucket. Use a folder
   #     prefix in the object name; --bar-uri later passes the Oracle-documented
   #     `file:///<folder>/<name>.bar` shape.
   oci os object put --bucket-name aidp-fusion-bundle-bar \
                     --file ./bundle-v0.1.0a0.bar \
                     --name aidp-fusion-bundle/bundle-v0.1.0a0.bar
   
   # 5b. (One-time, in OAC UI) Create the AIDP connection. Run the bundle
   #     in --print-only mode to write the 6-key JSON, then upload it via
   #     OAC UI: Data → Connections → Create → "Oracle AI Data Platform".
   aidp-fusion-bundle dashboard install --target oac --oac-url ... --print-only
   
   # 5c. Run the REST install (snapshot register + restore + poll). The bundle
   #     uses GET /catalog?type=connections&search=<name> to find the existing
   #     connection and skip the POST. Subsequent installs re-use it.
   aidp-fusion-bundle dashboard install --target oac \
     --oac-url https://oac.example.com \
     --bar-bucket aidp-fusion-bundle-bar \
     --bar-uri 'file:///aidp-fusion-bundle/bundle-v0.1.0a0.bar'
   

   Uses ONLY Oracle-documented public REST endpoints: GET /catalog?type=connections&search=<name> (precheck), POST /catalog/connections (skipped when precheck finds the existing connection — the realistic flow), POST /snapshots, POST /system/actions/restoreSnapshot, GET /workRequests/{id}. See docs/oac_rest_api_setup.md for the one-time IDCS confidential-app + Object Storage Resource Principal setup.

6. End users chat with the data via OAC MCP. Print the MCP config snippet:

   aidp-fusion-bundle dashboard mcp-config
   

   Paste into claude_desktop_config.json (or Claude Code / Cline / Copilot equivalent), restart the AI client. Then ask "what's our AR aging?" and watch MCP call discoverData → describeData → executeLogicalSQL against fusion_catalog.gold.ar_aging.

Key gotchas (live-validated where ✅)

• BICC role required — Fusion user must hold BIA_ADMINISTRATOR_DUTY or ORA_ASM_APPLICATION_IMPLEMENTATION_ADMIN_ABSTRACT. Without it, /biacm/api/v[12]/* endpoints 302-redirect to IDCS. Bootstrap probes for this. (✅ Casey.Brown demo pod: BIAdmin granted; works.)
• BICC External Storage profile — must be configured once in the BICC console (admin task: BICC Console → Configure External Storage → OCI Object Storage Connection tab → bucket name + namespace + region + OCI username + auth token → Test Connection → Save). The fusion.external.storage Spark option references this BICC profile name. There is no parallel AIDP-side registration. Bundle does not provision the BICC profile; bootstrap verifies it exists.
• First extract is slow — BICC builds a full snapshot on first call; subsequent runs use fusion.initial.extract-date for incremental.
• 499 row/page hard cap on Fusion REST (per MOS Doc ID 2429019.1) — bundle's REST fallback enforces this; anything >5k rows must use BICC.
• OAC MCP is read-only — it cannot create workbooks or register data sources. Bundle uses OAC REST API for write operations; MCP is for end-user chat consumption only.
• POST /catalog/connections REST validator does not bless AIDP idljdbc — Oracle's validator falls through to generic Oracle DB schemas requiring serviceName/password/connectionString. The realistic flow is therefore: customer creates the connection via OAC UI once (using the 6-key JSON written by --print-only), then dashboard install re-uses it via the precheck on subsequent runs. (✅ Live-validated TC10h-4, 2026-05-03 against disposable OAC1.)
• Snapshot BAR URI shape is file:///<folder>/<name>.bar — NOT oci://..., NOT bare object name, NOT the OCI Object Storage HTTPS URL. None of the seven URI variants tried during TC10h were correct. Verified live TC10h-3.
• OAC catalog browse needs search=* — GET /catalog?type=connections (no search) returns a single-element TypeInfo header ([{"type":"connections"}]), NOT the actual list. Bundle's list_connections defaults search="*" so the precheck works. (Caught + fixed during TC10h-3 live validation.)
• Use ExtractPVOs for bulk, NOT OTBI reporting PVOs — pdf1 Pro Tip; bundle's catalog refuses OTBI PVOs with a clear warning.

References

• Plan: C:\Users\anuma\.claude\plans\oracle-ai-data-platform-fusion-bundle.md
• Sibling plugin (single-PVO connector): oracle-ai-data-platform-workbench-spark-connectors
• Official Oracle BICC blog: https://blogs.oracle.com/ai-data-platform/bring-fusion-data-into-oracle-ai-data-platform-workbench-using-bicc
• Ateam blog (saas-batch path): https://www.ateam-oracle.com/how-to-extract-fusion-data-using-oracle-ai-data-platform
• Official sample notebook: oracle-aidp-samples/data-engineering/ingestion/Read_Only_Ingestion_Connectors.ipynb
• OAC MCP Preview docs: https://docs.oracle.com/en/cloud/paas/analytics-cloud/acsdv/access-oracle-analytics-cloud-mcp-server-preview.html