Oracle has introduced Autonomous AI Vector Database, a fully managed service on Oracle Cloud Infrastructure (OCI) built for semantic search, retrieval-augmented generation (RAG) and AI agents. The launch was announced on the Oracle Database blog on November 17, 2026, together with a step-by-step getting started guide.
The service aims at developers who would otherwise reach for a separate, vector-only database. It offers a simple API in front of Oracle AI Vector Search, while the engine underneath is Autonomous AI Database. That means teams get easy vector calls at the start and the full Oracle engine later, without moving data.
What Oracle Announced
Oracle positions the product as a bridge between quick pilots and production AI apps. Its main parts are:
- Three ways to code: a Python SDK, REST APIs, and SQL or PL/SQL for teams that prefer the database directly.
- Ready-made schema: creating a vector table also creates the vector column, metadata storage and indexes, using Oracle's recommended design.
- Index control: default vector and metadata indexes are built for you, and HNSW or IVF parameters can be tuned later.
- Search features: semantic search, filtering on vectors and metadata together, a choice of distance metric, and reranking with built-in or uploaded functions.
- Embeddings your way: load an embedding model into the database so vectors are made on insert, or bring vectors made by any outside model.
Free, Developer and Full Tiers
When you create an instance in the OCI Console, you pick one of three sizes. Always Free gives a small fixed shape, and the usual limit of two Always Free Autonomous AI Databases per tenancy applies.
The Developer option is a cheap, fixed 4 ECPU shape and is billed. The full version lets you set the ECPU count, storage and auto scaling for production loads. Both smaller options can be upgraded in place to the full service with an upgrade button, so a prototype does not need to be rebuilt.
How the Developer Workflow Looks
The getting started guide walks through a short path. You open Analytics and AI in the OCI Console, create the instance, and set two sets of credentials: the ADMIN account and a separate vector application user. A new Vector Database Console then handles model loading, table creation, test searches and OAuth tokens.
On the client side, the SDK package is oracle-vecdb on PyPI and needs Python 3.10 or newer. A single call creates a vector table and names which JSON key in the metadata should be embedded. Another call upserts records, and the database turns the chosen text into vectors with the loaded model.
Queries take plain text, a top-k value and optional filters, such as limiting results to a set of states. In the sample, the default table gets an IVF vector index plus a metadata index, which can be swapped for a custom strategy as data grows.
What It Means for Developers
The main point is that Oracle now has an entry product that looks and feels like the popular vector-only services. Python developers can work with tables, upserts and queries without writing SQL or designing a schema. That lowers the barrier for AI teams who never planned to use an Oracle database.
The bigger advantage comes later. Because the data lives in Autonomous AI Database, the same rows can be joined with business tables, spatial data or graph data when an app outgrows plain similarity search. Oracle's own example filters store results by distance from a customer, which is hard to do in a vector-only store without extra services.
Oracle developers who already use AI Vector Search in 26ai will not gain new engine features here. What they get is a faster start and a managed REST layer.
If you are planning agent memory or RAG on Oracle, our guides on agent memory with Oracle AI Database and choosing a chunk size for embeddings cover the design choices that still matter.
One practical caution: the quick start uses a username and password stored in a local file. For real apps, switch to OAuth tokens and a secrets manager, and keep credential files out of source control.
How to Get Started
Sign in to an OCI tenancy, or create a Free Tier account, and create an instance from the Autonomous AI Vector Database page. Choose Always Free to explore without cost, or Developer if you need the fixed 4 ECPU shape. Load an embedding model in the Vector Database Console and copy the REST endpoint from the details page.
Then install the SDK with pip, create a table, load a few records and run a first query. Oracle also links sample apps, notebooks and an API reference from the product page.
Conclusion
Autonomous AI Vector Database gives Oracle a simple, code-first front door for vector search, with a free tier and an in-place upgrade path. Its real value is that the data stays inside a full Oracle database, ready to mix with relational, spatial and graph data when an AI app needs more than similarity search.
Sources
- Oracle Database Blog: Autonomous Vector Database
- Oracle Database Blog: Getting Started with Autonomous AI Vector Database
See my articles more often when you search Google.
