Oracle Data Science Agent Reaches General Availability: Conversational Machine Learning Comes to Autonomous AI Database 26ai
Oracle has reached an important milestone for its AI and database platform. On June 23, 2026, the Oracle Data Science Agent became Generally Available for Oracle Autonomous AI Database Serverless 26ai, bringing a conversational machine learning assistant directly into the database. No separate Python or R environments are required, and most machine learning workflows can remain entirely inside the database without moving data to external analytics platforms. Through a guided chat interface, users can perform the entire data science lifecycle — from data profiling to production-ready inference SQL — while keeping data in the database.
Unlike general-purpose AI assistants that rely on external notebooks or third-party machine learning platforms, the Oracle Data Science Agent operates entirely within Oracle Autonomous AI Database. Built on Oracle Machine Learning (OML), it enables organizations to create, evaluate, and deploy machine learning models without exporting data from the database environment.
This is no longer a preview or a limited-access feature. With General Availability, Oracle is making conversational, in-database machine learning available for production use, fundamentally expanding who can build machine learning models inside Oracle Database and how quickly they can do it.
What the Data Science Agent Does
The Data Science Agent is embedded within Oracle Machine Learning (OML) and provides end-to-end data science lifecycle support through natural language interactions. Users can accomplish tasks that traditionally required specialized tooling, scripting expertise, and significant setup time.
The core capabilities include:
- Data Profiling: The agent can analyze datasets, identify distributions, detect data quality anomalies, and summarize key statistical characteristics using conversational prompts.
- Feature Preparation: Users can request feature engineering, transformations, and data preparation through natural language, with the agent generating the appropriate SQL and Oracle Machine Learning (OML) operations while guiding users through the workflow.
- Model Training: The agent supports in-database algorithms for classification, regression, anomaly detection, and clustering, including Oracle’s in-database implementation of the widely used XGBoost algorithm. Models are trained entirely within the database, leveraging Autonomous Database compute resources.
- Model Evaluation: Once trained, the agent presents evaluation metrics, confusion matrices, and performance summaries, helping users interpret and compare model performance.
- Inference SQL Generation: One of the most production-oriented capabilities is the automatic generation of deployment-ready SQL for scoring and inference, enabling immediate integration into applications and reporting pipelines.
Two Modes of Operation: Guided and Delegated
Oracle designed the agent with flexibility in mind, recognizing that users range from curious business analysts to experienced data scientists. The Data Science Agent supports two distinct interaction modes:
- Interactive Guided Mode: The agent walks users step-by-step through each phase of the machine learning workflow, prompting for additional information when needed, presenting options, and explaining decisions along the way. This mode is ideal for users who are newer to machine learning or who want granular control over each stage.
- Fully Autonomous Delegated Execution: For experienced practitioners or well-defined use cases, users can delegate the entire workflow to the agent. Describe the objective, point to the data, and the agent can orchestrate profiling, feature preparation, algorithm selection, model training, evaluation, and inference SQL generation with minimal user intervention.
This dual-mode approach is a pragmatic design decision. It makes the tool accessible without oversimplifying the experience for experts, while remaining approachable for newcomers.
Governance and Security: Built for the Enterprise
Any machine learning capability that runs inside a production database must address security, auditability, and access control. Oracle has incorporated several enterprise-grade governance features:
- Database Privilege Integration: The Data Science Agent operates entirely within existing Oracle Database privilege and role frameworks. Users can only access data and perform operations that their database privileges permit, eliminating privilege escalation and reducing shadow IT risks.
- Action Logs and SQL Visibility: Every action the agent performs is logged, and every SQL statement it generates is visible to the user. The generated SQL and workflow remain transparent, allowing DBAs and compliance teams to review exactly what was executed and when.
- Persistent Conversation History: Conversation history is retained, enabling reproducibility and auditability of machine learning workflows. If a model needs to be rebuilt or a decision reviewed months later, the conversational context remains available.
- Conversation Objects Catalog: A scoped catalog system controls access to conversation-generated objects, ensuring that models, datasets, and artifacts created through the agent adhere to organizational access policies.
This governance model is particularly significant because it enables organizations to broaden access to machine learning capabilities without compromising the security and governance framework that Oracle DBAs have spent years building and maintaining.
Why This Matters for Oracle Professionals
The Data Science Agent represents a meaningful shift in how Oracle positions machine learning within its ecosystem. For years, in-database machine learning through OML has been powerful but required specialized knowledge of PL/SQL APIs, OML4Py, or OML4R. The Data Science Agent significantly reduces that barrier.
For DBAs, this means machine learning workloads can remain inside the database, where they can be governed, monitored, and managed using existing operational processes. Organizations can reduce their dependence on external Spark clusters, Jupyter notebooks connected to production databases, and ad hoc data exports.
For developers, the generated inference SQL provides a direct path from experimentation to production. Models built through conversational workflows generate SQL that can be incorporated into applications, views, and scheduled jobs with minimal additional work.
For business analysts, this is arguably the biggest opportunity. The ability to build, evaluate, and deploy machine learning models using natural language, with little or no code, brings predictive analytics within reach of a much broader audience.
Practical Takeaway
If you are running or planning to run Oracle Autonomous AI Database Serverless 26ai, the Data Science Agent is now generally available. A practical first step is to identify a well-defined classification or regression use case with clean, accessible data already stored in your Autonomous Database. Begin with Guided Mode, inspect the generated SQL and workflow, validate the model metrics, and review the governance logs before moving to production workloads. This approach provides a solid understanding of both the platform’s capabilities and its governance model before broader adoption.
Oracle is making a clear strategic statement: machine learning should run where the data already resides. By bringing conversational AI directly into Autonomous AI Database, Oracle is lowering the barrier to building and deploying in-database machine learning while preserving the governance, security, and operational model that enterprises expect.