AI Data Platform

Build a complete data lakehouse across structured, semi-structured, and unstructured data. Raw data lands in object storage as the bronze layer, ingested where it lives or landed directly without unnecessary duplication. Autoscaling compute with managed Apache Spark refines it into clean, access-controlled data sets in the silver layer. Oracle Autonomous AI Database delivers the curated, AI-ready gold layer that is optimized for analytics, machine learning, and AI agents. One lakehouse. One environment. This means that every user—from data engineer to business user—works with the same trusted data.

• Any sources, any sinks, any language, anywhere
  Catalog and connect to more than 100 sources and sinks, including relational databases, SaaS applications, streaming feeds, and third-party data providers. For example, a data engineer can combine ERP records, customer events, and clickstream data into Apache Spark transformations, land trusted gold tab
• Visual data pipeline creation
  Design no-code and pro-code ETL and ELT pipelines that orchestrate ingestion, transformation, and enrichment from bronze to gold in a single workflow.
• Interoperability with open file formats
  Store and query data in open formats, including Apache Iceberg and Delta Lake directly on Oracle Cloud Infrastructure (OCI) Object Storage, with transactional consistency, time-travel queries, and schema evolution. Interoperate with Apache Spark, Apache Flink, and any engine compatible with Iceberg or Delta Lake. No proprietary lock-in, no rewrites when tools change.
• Best-in-class engines for data preparation and queries
  Data engineers can run large-scale data transformations in Apache Spark. Data scientists can build ML models. Business users can perform ad-hoc exploration using SQL and Oracle AI Database against the same data. Each user can pick the right engine for their workload without copying or remodeling data.
• Converged analytics across data and databases
  Query structured tables, semi-structured JSON, geospatial data, and AI vector embeddings across object storage and connected databases under one, role-based access control model. For example, a business user can validate a new revenue projection by joining the lakehouse activity streams with the authoritative customer records in Oracle AI Database in a single SQL query.
• AI-ready gold layer
  Machine learning pipelines, retrieval-augmented generation (RAG) knowledge bases, and AI agents consume curated, business-validated data. AI engineers ground models on the same authoritative facts that drive operational reporting, not raw or intermediate copies.
• Autonomous gold layer operations
  Gain the benefits of a self-tuning, self-scaling, and self-repairing warehouse with automatic indexing, query optimization, and workload management. Deliver consistent, subsecond performance across billions of rows during peak hours without manual tuning. Scale compute independently from storage to serve hundreds of concurrent users.
• First-class, first-party integration
  Bring Oracle Fusion Applications data, including ERP, HCM, supply chain, and CX, directly into the lakehouse. Prepare Fusion data in Apache Spark–powered notebooks for AI agent grounding, historical reporting, and predictive analytics on the operational system of record.
• Data sharing
  Share live lakehouse data with internal teams, partners, and Oracle applications without copying or moving it. Data sharing helps to keep every consumer on the most current, authoritative version.

Oracle AI Data Platform gives data scientists and ML engineers a fully managed environment to help build, train, track, and operationalize machine learning models directly over lakehouse data. The full MLOps lifecycle—distributed Spark training, experiment tracking, model registry, and catalog-published deployment—runs in a customer-managed and customer-governed workspace with no infrastructure to manage.

• ML pipelines and workflows: Orchestrate end-to-end ML pipelines with reusable components—data preparation, feature engineering, training, and evaluation—using AI Data Platform's workflow infrastructure. Build once, run on schedule or on trigger, with role-based access control policies you define and control.
• Experiments and model registry: Track all model training runs with automatic metrics logging, hyperparameter capture, and artifact versioning. Compare experiments, register production models, and manage lifecycle workflows.
• Model publishing and catalog registration: - Publish trained models to the AI data catalog to make them discoverable, versioned, and accessible across agents, applications, and workflows. Lineage tracking and access control policies are applied at registration.
• Apache Spark with GPU: Scale feature engineering and model training across distributed Spark clusters, accelerating workloads with NVIDIA market-leading GPU-powered ETL and ML.
• Data science agents (coming soon): AI agents that autonomously explore data sets, generate hypotheses, write and execute code, and iterate on model training, dramatically reducing time from raw data to production model.

Build AI agents and applications grounded in your enterprise's own data—not generic LLM capabilities. Your agents and apps are access-controlled by your policies and enriched with your business semantics and domain knowledge. AI agents are connected to your AI data catalog, business ontologies, and enterprise systems so they can reason within the context your company actually runs on. Compose multi-agent systems using any foundation model—from no-code visual builders to full pro-code development—and deploy to managed AI compute with built-in observability.

• No-code visual flow builder: Design and compose agents visually with a drag-and-drop canvas. Connect SQL tools, RAG knowledge bases, LLM prompt nodes, and fan out to multiple tools—all without writing a line of code. Switch foundation models from a drop-down without rebuilding the flow.
• Pro-code development: Write agents in Python using the AI Data Platform SDK with full access to LangChain, OCI Generative AI, and any open source library. Every visual flow capability is fully accessible in code; import scikit-learn, LangChain, or any framework alongside the agent SDK.
• Multi-agent systems: Design and orchestrate systems of cooperating AI agents, such as orchestrator agents, specialist sub-agents, and tool-using agents, to help tackle complex, multi-step enterprise workflows autonomously using the A2A protocol.
• Any foundation model: Use any model on OCI, such as Llama, Cohere, Mistral, Grok, and more, or bring your own fine-tuned models. Swap models from the Oracle Cloud Infrastructure (OCI) Compute drop-down menu without rebuilding application logic.
• AIOps and observability: Full observability across the agent lifecycle from development to production. Test agents interactively in the platform’s playground, inspecting tool calls, LLM reasoning, and outputs before deployment. Sessions capture an audit trail with status, duration, inputs/outputs, and per-step event detail across dev, test, and production. Monitor latency, token usage, error rates, and custom business KPIs in real time.

A single, integrated development environment for data engineers, data scientists, and AI developers to collaborate on end-to-end data and AI projects with enterprise-grade role-based access control (RBAC), CI/CD, versioning, and auditability built in. Connect all personas through shared tools, notebooks, and pipelines, all powered by integrated access to the platform's underlying services and catalog.

• Workbench home dashboard: A unified home screen with access to every capability—master catalog, workspaces, workflows, compute, agent flows, and administrative functions, including data sharing, roles, and audit logs. Quick-action tiles drop you directly into AI integration, get data, analyze, or manage access functions.
• Workspaces: Project-scoped environments where teams collaborate on notebooks, pipelines, agents, and experiments. All artifacts are versioned, shared, and access-controlled within the workspace boundary with full role-based access control per project.
• Notebooks, workflows, and agents: All development artifacts are managed in one place, including Jupyter-compatible notebooks, pipeline workflow DAGs, AI agents, and ML experiments. Share, version, and collaborate across the team without context-switching.
• SQL tool and compute management: Run ad-hoc SQL queries directly against catalog tables with compute lifecycle controls built in. Attach, detach, or spin up new AI compute resources from within the IDE—no separate console required.
• CI/CD and Git integration: Native Git integration for versioning notebooks, pipelines, agent definitions, and model configurations. Connect to GitHub, GitLab, or Bitbucket for continuous integration and automated deployment of data-to-AI project artifacts.
• RBAC, auditing, and network isolation: Granular, role-based access control across workspaces, artifacts, and compute. Comprehensive audit logs for every user action. Network isolation with private endpoints helps ensure sensitive workloads never touch the public internet.

Discover, understand, and manage access to all your data and AI assets in a single, unified catalog that spans the full medallion architecture, including bronze ingestion, silver curation, and gold AI-ready data products. Oracle AI Data Platform's AI data catalog connects to Autonomous AI Database, OCI Object Storage, and third-party sources through external catalogs, surfacing rich business meaning through semantic context and ontologies. Every team finds not just what data exists but what it means to the business. Every AI agent automatically inherits that understanding.

• Unified data and AI asset catalog: A single catalog for all data and AI assets, including structured tables, unstructured files, knowledge bases, ML models, feature stores, and agent definitions. The catalog is access-managed with consistent policies across data and AI. It covers the full medallion architecture, enabling fast ingestion, curation, and delivery of data products and AI applications at every layer.
• External catalogs and asset discovery: Connect to external data sources—Autonomous AI Database, OCI Object Storage, and third-party systems—without unnecessarily moving or duplicating data. Automatically discover and catalog structured and unstructured assets with AI-assisted metadata enrichment and lineage tracking from the point of connection.
• Business ontologies and semantic layer: Define domain ontologies and semantic relationships between business concepts. Business glossaries, semantic ontologies, domain taxonomies, and AI-generated synonyms enable users find data by meaning, not table names. AI agents automatically inherit this semantic understanding.
• Zero copy: Query data where it lives. Access and query data without moving or copying it. Connect directly to your existing Oracle Database, Autonomous AI Database, and Exadata and query in place using SQL. Data stays in its authoritative source while the catalog federates access, applies role-based access control, and surfaces it, reducing duplication and lowering cost.
• Zero ETL: Help eliminate ETL pipelines with Oracle GoldenGate for AI-powered, real-time, log-based replication from any source database for high-throughput streaming ingestion directly into the lakehouse. Data is cataloged and AI-ready the moment it lands.
• Volumes: Volumes store unstructured data alongside data assets in the catalog. Attach to knowledge bases to help enable agents and applications to securely retrieve unstructured content, such as documents, PDFs, and images.
• Data lineage: Visualize end-to-end data lineage, including raw ingestion through transformations, ML feature engineering, model training, and AI application serving. Instantly understand the impact of upstream changes across the full data and AI pipeline.

Enterprise AI at scale demands enterprise-grade security, access management, and auditability, applied consistently across every data asset, model, and agent. Oracle AI Data Platform enforces a two-layer security model: Oracle Cloud Infrastructure Identity and Access Management (IAM) for identity and authentication, combined with Oracle AI Data Platform Workbench for fine-grained access control across every catalog asset, workspace, and AI resource, with comprehensive audit logs for full traceability. Manage your entire data and AI estate without bolting security on as an afterthought.

• Two-layer security model: Security operates at two levels: OCI IAM handles identity, authentication, and cloud-level access; AI Data Platform Workbench controls who can discover, read, modify, and use each data and AI asset within the platform. Both policies are defined and managed by the customer. Oracle maintains the framework that enforces them. You gain the advantages of a defense-in-depth architecture without ceding control.
• AI Data Platform Workbench: Granular, role-based access control across workspaces, catalog assets, compute resources, agents, and administrative functions. Roles are applied consistently across data and AI with no gaps between what users can see in the catalog and what they can act on in the platform.
• Audit logging and traceability: Comprehensive audit logs for every user action, data access, agent interaction, and administrative change provide traceability across the platform to support compliance, investigation, and access history reporting.
• Network isolation and private endpoints: Deploy workspaces, compute, and data connections within private VCN subnets with private endpoints. Sensitive workloads never traverse the public internet. Network isolation is enforced at the infrastructure layer.
• OCI Identity and Access Management and security integration: OCI IAM handles identity federation and authentication across the platform. Deep integration with OCI Vault for secrets management, OCI Certificates for TLS, and OCI Security Advisor for posture recommendations provides a unified security control plane across data, AI, and infrastructure.

A centralized registry for discovering and managing AI agents at enterprise scale, including agents built with AI Data Platform and third-party agents as well as MCP servers and tools. The AI registry tracks every agent's identity, capabilities, permissions, versions, and interaction logs, giving platform administrators visibility and control over the growing fleet of AI agents operating across the enterprise.

• Centralized agent registry A unified registry to help you list, version, and manage all your AI agents whether they are built with AI Data Platform or third-party tools. Track each agent's identity, declared capabilities, permission scope, version history, and interaction logs. Discover agents by capability, domain, or team with rich metadata for management, access control, and reuse at enterprise scale.
• Agent-to-agent (A2A) protocol: Enable structured, standardized communication between AI Data Platform agents and third-party agents using the open A2A protocol. Compose complex workflows where specialist agents collaborate with orchestrator agents with clear identity, capability declaration, and permission boundaries enforced at every interaction.
• Model Context Protocol (MCP) servers and tools: Register and expose MCP servers and tools. Agents dynamically discover and invoke tools at runtime, including database queries, REST APIs, and custom business functions, without hardcoded bindings. Tool permissions are managed through the registry policies you define consistent with how agent access is controlled.

Give nontechnical users access to the full power of your enterprise data through self-service analytics, curated AI agents, and AI-powered insights. AI capabilities are embedded directly in the business workflows where decisions happen. No technical skills required. It’s enterprise AI democratized.

• Analytics in business workflows: Oracle Analytics Cloud embeds world-class analytics directly into the applications your teams use every day, including Oracle Fusion ERP, HCM, and CX. Ask questions in natural language, surface AI-generated narratives, and share centrally managed dashboards within the workflow.
• A unified conversational interface (coming soon): A single pane of glass to discover, query, and collaborate with your organization's AI agents. Agent Hub interprets business user requests, finds and invokes the right agent, and presents results in context—all through natural language. No technical skills required.
• Curated AI agent library: Browse a curated library of approved agents—internal agents built by your data teams and vetted third-party agents—with descriptions, example prompts, and usage guidance for common business tasks.
• Managed access and security: Users can manage every agent interaction and analytics query by the same RBAC policies as the underlying data. Users only see agents and data they're authorized to access. It’s enterprise security without extra configuration.