oracle-ai-developer-hub

TanStack Chat + Oracle 26ai Select AI

A chat application built with TanStack Start and TanStack AI that connects to an Oracle 26ai Autonomous Database via Select AI. The outer LLM (Claude, GPT-4o, Gemini, or Ollama) decides when to query the database and asks questions in plain English. The inner LLM inside Oracle (Select AI) translates those questions to SQL, executes the query, and returns a narrated answer.

Browser  ->  TanStack Start  ->  TanStack AI chat handler
                                        |
                                  Outer LLM (Claude / GPT-4o / etc.)
                                        | (tool call: execute_query)
                                  oracledb thin client (Node.js)
                                        | (TLS)
                                  Oracle 26ai Autonomous DB
                                        | (DBMS_CLOUD_AI.GENERATE)
                                  Select AI  ->  Anthropic Claude (in-DB)
                                        |
                                  SQL generated + executed inside DB
                                        |
                                  Natural language answer returned

Prerequisites

Node.js (current LTS recommended)
pnpm (corepack enable or install globally)
An Oracle Cloud account with an Always Free Autonomous AI Database provisioned (26ai, ATP workload, mTLS off / TLS only)
An Anthropic API key (used both by the app’s outer LLM and by Select AI inside the database)

Quick start

pnpm install
cp .env.example .env.local
# Fill in your API keys and Oracle connection details in .env.local
pnpm dev

The app serves at http://localhost:3000.

Environment variables

Copy .env.example to .env.local and fill in the values:

Variable	Required	Description
`ANTHROPIC_API_KEY`	At least one LLM key	Anthropic API key (checked first)
`OPENAI_API_KEY`		OpenAI API key (fallback)
`GEMINI_API_KEY`		Google Gemini API key (fallback)
`ORACLE_USER`	Yes	Database user (default: `shoestore`)
`ORACLE_PASSWORD`	Yes	Password for the database user
`ORACLE_CONNECTION_STRING`	Yes	TLS connection string from OCI Console
`ORACLE_AI_PROFILE`	Yes	Select AI profile name (default: `SHOESTORE_AI`)

The chat route picks the first available LLM provider in order: Anthropic, OpenAI, Gemini, then local Ollama.

Getting the Oracle connection string

Go to the OCI Console > Autonomous Database > your database
Click DB Connection
Select the TLS tab (not mTLS)
Copy the connection string for the _low service (least resource priority, fine for a demo)

Database setup

1. Seed the database

Run seed.sql as ADMIN (first block creates the user and grants), then as SHOESTORE (tables + data). The easiest way is to paste it into Database Actions > SQL Worksheet in the OCI console.

Alternatively, using SQLcl:

# As ADMIN — creates user, grants, network ACL
sql admin/<password>@'<TLS connection string>'
@seed.sql

The seed script creates:

products (20 rows) – shoes across 5 categories (running, casual, hiking, basketball, skateboarding)
customers (15 rows) – spread across US cities/states
transactions (40 rows) – purchases with completed, returned, and pending statuses
Table and column comments that Select AI uses to understand the schema

2. Configure Select AI

Run setup-selectai.sql as the SHOESTORE user. Before running, replace <your-anthropic-api-key> with your actual Anthropic API key:

sql shoestore/<password>@'<TLS connection string>'
@setup-selectai.sql

This creates:

An Anthropic credential stored encrypted in the database
A Select AI profile (SHOESTORE_AI) pointing at the three tables
A verification query to confirm it works

Select AI actions reference

Action	What it does	When to use
`narrate`	Generates SQL, executes it, returns natural language summary	Default for our chat tool
`runsql`	Generates SQL, executes it, returns raw rows	When the app needs structured data
`showsql`	Returns the generated SQL without executing	Debugging / transparency
`chat`	General conversation with the LLM (no SQL)	Off-topic questions

How it works

When a user asks a data question in the chat:

The outer LLM (TanStack AI) decides to call the execute_query tool with a plain English question
The app sends that question to Oracle via DBMS_CLOUD_AI.GENERATE using the oracledb thin client
Select AI (the inner LLM inside Oracle) generates SQL from the question using table/column comments as context
The SQL executes inside the database and Select AI narrates the results
The narrated answer returns to the outer LLM, which incorporates it into its response

The outer LLM never writes SQL. Table/column comments are the prompt engineering for Select AI – the more descriptive the comments, the better the generated SQL.

Demo queries to try

“How many products do we carry?”
“What’s our most expensive shoe?”
“Who are our customers from Oregon?”
“What are the top selling brands by revenue?”
“Show me all returned orders”
“Which customer has spent the most money?”
“What category of shoes sells the best?”
“Are there any pending orders?”
“What should a customer from Gresham buy based on what’s popular in their area?”
“Compare running shoe sales vs hiking shoe sales”

Building for production

pnpm build
pnpm preview

Testing

pnpm test

Cleanup

To tear down Select AI config without dropping the database, run as SHOESTORE:

BEGIN
  DBMS_CLOUD_AI.DROP_PROFILE(profile_name => 'SHOESTORE_AI', force => true);
END;
/

BEGIN
  DBMS_CLOUD.DROP_CREDENTIAL(credential_name => 'ANTHROPIC_CRED');
END;
/

To drop the user entirely, run as ADMIN:

DROP USER shoestore CASCADE;

Tech stack

Framework: TanStack Start + TanStack Router (file-based routing)
AI: TanStack AI with Anthropic/OpenAI/Gemini/Ollama adapters
Database: Oracle 26ai Autonomous DB with Select AI (DBMS_CLOUD_AI)
DB Client: oracledb v6+ thin mode (pure JS, no Oracle Client install needed)
Styling: Tailwind CSS
Language: TypeScript

License

Licensed under the Universal Permissive License (UPL), Version 1.0.

See LICENSE for more details.

ORACLE AND ITS AFFILIATES DO NOT PROVIDE ANY WARRANTY WHATSOEVER, EXPRESS OR IMPLIED, FOR ANY SOFTWARE, MATERIAL OR CONTENT OF ANY KIND CONTAINED OR PRODUCED WITHIN THIS REPOSITORY, AND IN PARTICULAR SPECIFICALLY DISCLAIM ANY AND ALL IMPLIED WARRANTIES OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE. FURTHERMORE, ORACLE AND ITS AFFILIATES DO NOT REPRESENT THAT ANY CUSTOMARY SECURITY REVIEW HAS BEEN PERFORMED WITH RESPECT TO ANY SOFTWARE, MATERIAL OR CONTENT CONTAINED OR PRODUCED WITHIN THIS REPOSITORY. IN ADDITION, AND WITHOUT LIMITING THE FOREGOING, THIRD PARTIES MAY HAVE POSTED SOFTWARE, MATERIAL OR CONTENT TO THIS REPOSITORY WITHOUT ANY REVIEW. USE AT YOUR OWN RISK.