PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteOracle Select AI lets you ask an Oracle database a question in plain English and have the database turn that question into SQL, run it, or explain it. It is not a standalone chatbot. It is a database feature, reached through SQL and related interfaces, that sends a prompt to a large language model (LLM) you configure. The convenience is real, but so is the need to review the SQL it writes and the answers it returns.
What Select AI actually is
Select AI is built into Oracle’s database platform and is driven by the DBMS_CLOUD_AI package. You choose the LLM provider and model, store the provider credential in the database, and connect the two through an AI profile. Once a profile is enabled, a user can write a natural-language prompt inside a SELECT statement using the AI keyword, and the database handles the rest.
The useful change is in how a question is expressed. Instead of starting from table names and join conditions, the user describes what they want to know. Select AI reads the database’s schema information, asks the configured model to draft a query, and then runs that query under the user’s own database privileges.
How a prompt becomes SQL
When you ask for SQL generation, DBMS_CLOUD_AI builds an augmented prompt that contains your question plus relevant schema metadata. That metadata can include table and view definitions, column and table comments, and other data-dictionary content. The model uses this to understand which tables and columns exist and how they are named.
Oracle states that actual row and column values from tables and views are not included in this SQL-generation prompt. The model sees the structure of your data, not its contents. The generated statement then executes inside the database, which is why its correctness and scope matter.
The actions and what each one sends to the model
SQL generation is only one of several actions. The table below separates them because the data each one exposes to the LLM is different, and that difference is the point most often misunderstood.
| Action | What is sent to the LLM | What it is for |
|---|---|---|
| SQL generation (natural language to SQL) | Your prompt plus schema metadata: definitions, comments and data-dictionary content. Oracle says row and column values are not included. | Generating, running and explaining SQL against your tables |
narrate |
Results from a generated database query, or content retrieved from a vector store | Turning query results or retrieved text into a natural-language answer |
| RAG (retrieval-augmented generation) | Vector-store content selected by semantic similarity search and added to the prompt | Answering questions from documents or other content you have vectorised |
| Chat | Your natural-language input for a general response | General conversational answers, without the SQL step |
The practical consequence: if you enable narrate on a query that returns sensitive rows, those rows are part of what the model receives. “No database data reaches the model” is therefore too strong a statement. The accurate version is that SQL generation by default shares structure, while narrate and RAG can share results and retrieved content.
Setting up Select AI
Oracle’s getting-started guide for Oracle AI Database 26 gives a three-stage sequence. The steps below expand it with the prerequisites Oracle’s prerequisite guide lists.
Recommended Free Tools
- Check your environment. You need an OCI cloud account, an Autonomous AI Database instance, and a paid API account with a supported LLM provider.
- Create a provider credential. Store the API key or equivalent credential for your chosen provider in the database.
- Grant privileges. The user who will call Select AI needs
EXECUTEonDBMS_CLOUD_AI. - Configure outbound access if required. For most external providers, the database needs outbound network ACL privileges to reach the provider’s endpoint. Oracle’s prerequisite guide states that these are not needed for OCI Generative AI.
- Create and enable an AI profile. The profile names the provider, model and credential that Select AI will use.
- Run a test prompt. Use the
AIkeyword in aSELECTstatement with a simple question against a non-sensitive table, then read the generated SQL before trusting the result.
Work through the steps in this order. Failures at step 2 or 4 are the most common reason a first test does not return anything useful, and they are easier to diagnose before any real data is involved.
Supported providers
Oracle’s prerequisite guide lists these provider categories: OpenAI, OpenAI-compatible providers, Cohere, Azure OpenAI Service, OCI Generative AI, Google, Anthropic, Hugging Face and AWS. Model catalogs, regional availability, language support and pricing are set by each provider and change over time. Oracle’s documentation sets the integration framework; confirm the specific model you plan to use, and where it is hosted, with the provider before you design around it.
Rank #4
Capabilities depend on release
Select AI is documented across several Oracle platforms: Autonomous AI Database Serverless, Dedicated Exadata Infrastructure, Cloud@Customer, Oracle AI Database 26ai and Oracle Database 19c. Oracle directs readers to a capability matrix for release-specific details, and that matrix, not a feature list, should decide what you can use on your system.
The Oracle AI Database 26 feature reference lists the following capabilities. Treat it as a description of that release, not a guarantee that each item exists in every Oracle deployment.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →| Capability | Described in Oracle AI Database 26 materials | Scope note |
|---|---|---|
| Natural language to SQL | Generating, running and explaining SQL | Core Select AI function |
| Vector indexes and RAG | Automated vector-index creation and retrieval-augmented generation | Depends on release and vector-store setup |
| Agent framework | Agent workflows through DBMS_CLOUD_AI_AGENT |
Confirm availability in your release |
| Synthetic data | Synthetic-data generation | Confirm availability in your release |
| Text processing | Summarization and translation | Confirm availability in your release |
| Programmatic access | PL/SQL and Python APIs | Confirm availability in your release |
Checks before you deploy
Use this list to decide whether a Select AI setup is ready for real data:
- Your database release and deployment type appear in the capability matrix for the feature you need.
- The provider, model and region are confirmed directly with the provider, including current terms and charges.
- Each user who will call Select AI has only the privileges their work requires.
- You know whether
narrateor RAG will send query results or vector content to the model, and whether that content is permitted to leave the database. - Schema comments do not contain information you would not want an LLM to see, because they are part of the prompt.
- Someone reads generated SQL before it is trusted for reporting or decisions.
Accuracy and security limits
Oracle’s own usage guidance is direct about the risk. In its words: “Thus, while LLMs are adept at generating useful and relevant content, they also can generate incorrect and false information including SQL queries that produce inaccurate results and/or compromise security of your data.” (Oracle Select AI documentation.)
Generated statements run in the database, so a wrong query produces wrong numbers, and an overly broad query can read more than the question required. Natural-language access does not replace the usual controls: scoped privileges, review of statements, data governance and checking results against a known figure before anyone acts on them. Select AI reduces the effort of writing a query. It does not reduce the responsibility for what the query returns.
Oracle’s Select AI page was marked as updated on 30 September 2026. Because provider support and release features change, verify details against the current Oracle documentation at the time you deploy.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
