Oracle launches managed vector database with free entry and in-place upgrades
Autonomous AI Vector Database handles search setup and database maintenance. Free and Developer deployments have fixed compute sizes; configurable production capacity comes with the full version.
Oracle’s new managed vector database offers a route from a small deployment to configurable production capacity without migrating the database: both starter tiers can be upgraded in place. The trade-off is that Always Free and Developer compute allocations cannot be resized; Developer starts at four ECPUs and is paid, while the free option shares a two-database-per-tenancy limit with other Always Free Autonomous AI Databases. Oracle is targeting teams that want managed database operations as they move AI search and retrieval applications beyond a pilot.
01
The Python toolkit can generate searchable vectors automatically when text is inserted, or developers can provide vectors made with their preferred embedding model.
02
Developers can use Python, REST, or SQL and customize schemas and indexes, including supported HNSW and IVF options.
03
Search supports similarity, metadata filters, and reranking; Oracle’s example also filters products by whether a store is within five miles.
Oracle is promising an easier start for AI search without a separate database migration when applications grow. On November 17, 2026, it introduced Autonomous AI Vector Database, a fully managed service for search by meaning, retrieval-augmented generation and AI-agent applications. Free and Developer deployments can upgrade in place, but their starting compute sizes are fixed.
From stored text to searchable meaning
The service extends Oracle AI Vector Search with interfaces for Python, REST APIs and SQL, the language used to query databases. Oracle positions it as a place to build knowledge repositories for retrieval-augmented generation, or RAG: finding relevant material that an AI application can use when answering a request.
Its Python toolkit creates the underlying structures for vectors, descriptive metadata and search indexes, rather than requiring developers to design them first. In Oracle’s getting-started example, a developer loads an embedding model, then selects a text field to turn automatically into a searchable vector representation whenever data is inserted.
That removes a separate embedding-generation step from the demonstrated loading workflow. Developers can also supply vectors generated by an embedding model of their choice. Automatic setup is not mandatory: they can customize the schema and indexing strategy, including the supported HNSW and IVF search-index options.
Search can combine similarity with filters on other attributes, then rerank the results using built-in or uploaded functions. Oracle illustrates the business-data connection with a store search: a query could retrieve relevant products while using location information and spatial-distance functions to restrict results to stores within five miles of a customer.
Three deployment choices, one upgrade path
Oracle’s launch-day provisioning guide distinguishes three configurations. Always Free has a small, fixed compute allocation. Developer has a fixed allocation of four ECPUs, Oracle’s compute-capacity unit, and incurs charges. The full version lets users configure compute and storage and select optional automatic scaling for production workloads.
Both starter configurations have an upgrade button for moving to the full version in place. Provisioning requires an Oracle Cloud Infrastructure account with permission to create an instance. Oracle says creation typically takes a few minutes; the database is ready when its status changes from Provisioning to Available.
Managed maintenance, configurable access
Underneath the vector interfaces is Autonomous AI Database. Oracle says the service draws on that platform’s automated provisioning, patching, backups and scaling, with built-in encryption, access controls and observability. Those are the concrete operational features behind its pitch that developers can take a search application beyond a pilot without managing the database infrastructure themselves.
For Python developers, the entry point is the oracle-vecdb package. The database console supplies the REST endpoint needed to configure the client and tools for loading embedding models, creating vector tables, running searches and generating OAuth tokens. Oracle’s walkthrough uses username-and-password authentication for simplicity, while also supporting OAuth.
pip install oracle-vecdb
Managed infrastructure does not remove the application’s credential responsibilities. Oracle explicitly tells developers not to commit their local credentials file to source control and to use an approved secrets-management solution for deployed applications.
Sources
blogs.oracle.comGetting Started with Autonomous AI Vector Database
Reader comments
Newest comments first. Replies stay oldest first.