OAC AI Ecosystem¶

Last updated: 2026-04-27 Tags: AI ecosystem, AI Assistant, AI Agents, MCP Server, generative AI, OCI Generative AI, conversational analytics, augmented analytics

📖 Full Oracle Documentation: Oracle Analytics Blog — AI Ecosystem · Visualizing Data — AI Assistant

Summary¶

OAC's AI ecosystem is a layered stack: augmented analytics built into the product (Explain, Auto Insights, NLQ), generative AI via OCI Generative AI integration (AI Assistant in homepage, workbook, mobile), AI Agents for domain-specialized conversational experiences, and the MCP Server for external AI client integration. Together they shift OAC from a "platform you navigate" to "analytics you talk to."

Source: Oracle Analytics Blog, The Oracle Analytics Cloud AI Ecosystem + product documentation

The Stack — Five Layers of AI in OAC¶

┌─────────────────────────────────────────────────┐
│  LAYER 5: External AI Integration               │
│  → MCP Server: Claude / Cline / Copilot / n8n   │
├─────────────────────────────────────────────────┤
│  LAYER 4: AI Agents (Domain Specialists)        │
│  → Bound to dataset + RAG knowledge docs + SI   │
├─────────────────────────────────────────────────┤
│  LAYER 3: AI Assistant (Generative)             │
│  → Homepage / Workbook / Mobile conversational  │
├─────────────────────────────────────────────────┤
│  LAYER 2: NLQ + Augmented Analytics             │
│  → Ask, Explain, Auto Insights                  │
├─────────────────────────────────────────────────┤
│  LAYER 1: Built-in ML & OCI AI Services         │
│  → OML, Vision, Language, Forecasting           │
└─────────────────────────────────────────────────┘

Layer 1: Built-in ML & OCI AI Services¶

The foundation — predictive and descriptive analytics.

Oracle Machine Learning (OML)¶

Train models in Data Flows
Regression, classification, clustering
Score data with Apply Model step

OCI AI Services Integration¶

Service	Capability
OCI AI Vision	Image classification, object detection
OCI AI Language	Sentiment, entities, key phrases
OCI AI Forecasting	Time-series prediction
OCI AI Document Understanding	OCR, table extraction
OCI Anomaly Detection	Multivariate outlier detection

See Machine Learning & AI Features.

Layer 2: NLQ + Augmented Analytics¶

Quick-win AI features built into Workbooks.

Ask (Natural Language Query)¶

Type a question in the Workbook → get a visualization.

"Top 10 products by revenue last quarter"

Explain¶

Right-click a measure → Explain.

Basic facts, key drivers, segments, anomalies
One-click adoption of insights into canvas

Auto Insights¶

Proactive insights surfaced when you open a Workbook:

Significant changes
Correlations
Outliers

See Workbooks & Visualizations and Machine Learning & AI Features.

Layer 3: AI Assistant (Three Modes)¶

Conversational AI built into OAC's UI, powered by OCI Generative AI.

Homepage AI Assistant¶

Start your day with a question, not a dashboard
"Build me an executive profitability dashboard" → generates a workbook
Search catalog + answer with public/internal knowledge

Workbook AI Assistant¶

In-canvas conversational partner
Iteratively refine visualizations
"Add a trend line" / "Compare to last year" / "Filter to enterprise customers"
Smart follow-up question handling

Mobile AI Assistant¶

Voice input on phone/tablet
Podcast-style audio narration of insights
Hands-free analytics for executives

Capabilities Across Modes¶

Context-aware dialogue (remembers previous turns)
Smart semantic search (catalog + public knowledge)
Role-based access enforcement
User feedback loop for continuous improvement

Layer 4: AI Agents (Domain Specialists)¶

A step beyond assistants — dataset-specific, RAG-enabled, domain-tuned.

Difference from AI Assistant: - Bound to one dataset (governed scope) - Customized with Supplemental Instructions (R.T.C.C.O.E framework) - Enriched with Knowledge Documents (PDFs/TXT) via RAG - Acts as a domain expert (not a generalist)

Use cases: - Sales Operations Analyst Agent - HR Compensation Analyst Agent - Finance FP&A Agent - Marketing Performance Agent

See OAC AI Agents for full framework.

Layer 5: External AI via MCP Server¶

Open standard integration for any MCP-compliant AI client.

Three operations: Discover → Describe → Execute

Clients: Claude Desktop, Claude Code, Cline, GitHub Copilot, n8n, LangChain, plus any MCP-compliant client.

Use cases: - Slack bot answering "What's our quarterly bookings?" - Custom app with LangChain RAG over OAC data - IDE coding assistants aware of your real schema - n8n automation with OAC as a knowledge source

See OAC MCP Server.

Architectural Principle: One Brain, Many Surfaces¶

All five layers share the same governed foundation:

The Semantic Model (RPD) provides metric consistency
Row-level security applies across all interfaces
Roles control what each user/agent can see
The catalog is the single source of truth

                  ┌─ Homepage AI Assistant
                  ├─ Workbook AI Assistant
   GOVERNED       ├─ Mobile AI Assistant
   FOUNDATION ────┼─ Domain AI Agents
                  ├─ MCP Server (external)
                  └─ Built-in NLQ / Explain / Auto Insights

This means: a metric defined once in the Semantic Model → consistent answers across every AI interface.

Choosing the Right AI Surface¶

User Need	Recommended Surface
Casual Q&A while in OAC	AI Assistant (Workbook)
Build a new dashboard from scratch	AI Assistant (Homepage)
Mobile / voice / on-the-go	AI Assistant (Mobile)
Domain expert experience for a specific team	AI Agent
Bring OAC data into a custom app or workflow	MCP Server
Quick insight on a metric	Explain (right-click)
Open Workbook insights	Auto Insights
Predict future values	OML / OCI AI Forecasting
Apply a trained model to new data	Data Flow Apply Model

Strategic Direction (Per Oracle's Roadmap)¶

The ecosystem is moving from "navigate to insight" → "talk to your data":

"Analytics is no longer something users navigate. It's something they talk to."

Key trends:

AI Assistant becoming the default entry point
Domain Agents proliferating (each business area has its own)
MCP Server enabling embedded analytics in any app
Voice-first mobile experiences
Generative AI reducing time from question to insight

Adoption Roadmap (Recommended)¶

For organizations adopting OAC's AI capabilities:

Phase 1 — Foundation (Month 1-2) - Solidify Semantic Model (consistent metrics) - Index key datasets for AI Assistant - Train users on Explain and Ask

Phase 2 — AI Assistant (Month 2-3) - Roll out AI Assistant to power users - Collect feedback on common queries - Refine Subject Area metadata for AI

Phase 3 — Domain Agents (Month 3-6) - Identify 3-5 high-value domains (Sales, Finance, HR) - Build agents using R.T.C.C.O.E framework - Iterate Supplemental Instructions weekly based on user feedback

Phase 4 — MCP & Integration (Month 6+) - Set up MCP Server - Pilot with one external use case (Slack bot, custom app) - Expand to additional automations

Governance Across the Ecosystem¶

A governance checklist that applies to ALL AI surfaces:

[ ] Semantic Model defines KPIs consistently
[ ] RLS rules cover all sensitive data
[ ] AI Assistant indexed datasets reviewed for sensitive columns
[ ] AI Agent SI includes "do not fabricate" constraint
[ ] MCP Server users use enterprise LLMs (zero retention)
[ ] Audit logs reviewed for unusual queries
[ ] User feedback loop active for AI improvements