Oracle adds AI-provider connections and broader database checks to GoldenGate
The update separates provider credentials from model settings and automates verification infrastructure. Teams still need to configure connections, comparison jobs and target columns.
Oracle’s October 23, 2026 GoldenGate update expands two distinct operator workflows: Replicat can generate and write embeddings during replication, while Veridata compares source and destination records, with MySQL- and PostgreSQL-family connections newly supported. A Resource Manager template provisions one Veridata Server and two Agents, but operators must still configure connections, table pairs, and comparison jobs. AI-provider access also requires network connectivity and the right Vault and Secrets policies, so selecting a provider in the menu is not sufficient.
01
With GoldenGate 26ai or later, Replicat can call @AISERVICE during apply and map the returned embedding into a target column.
02
GoldenGate separates provider connections from AI Model connections; the latter select a model, set a model key and input-character limit, and can reuse provider credentials.
03
Provider setup requires network connectivity and the necessary Vault and Secrets access policies, which vary by provider.
Oracle is giving GoldenGate users more ways to check copied database records and generate AI embeddings as data moves. Its October 23, 2026, update describes new AI-provider connections, MySQL and PostgreSQL verification support, and an automated setup template. The changes address both preparing data for AI applications and checking that source and destination records match.
Provider access is separate from model choice
GoldenGate’s connection menu now groups OCI Generative AI, OpenAI, Google Gemini and Voyage AI under AI Services and Models. A provider connection holds the details needed to access the service. An AI Model connection specifies which model the replication pipeline will use.
That model connection can reuse an existing provider connection or accept provider details directly. It also sets a model key and maximum input characters. Network connectivity and the required Vault and Secrets access policies must be prepared before configuring the provider; the menu entry alone is not a working connection.
The access requirements differ by provider. OCI Generative AI uses tenancy and user selection, a region, an API-key secret and a key fingerprint. Oracle lists API-key authentication, an API-key secret and a Base URL for OpenAI, Google Gemini and Voyage AI.
Embeddings are generated as records are applied
With GoldenGate 26ai and higher, Replicat—the process that applies replicated data—can call a configured model through @AISERVICE. It then maps the returned vector embedding into a target column. Oracle suggests evaluating this approach for product descriptions, generating embeddings while those descriptions are replicated rather than handling them afterward.
The underlying function is not itself new with this update. Oracle’s August 30 @AISERVICE reference already describes invoking a model when records are applied and using its result in the destination mapping. Its example uses OpenAI’s text-embedding-3-small model and requires adding an embedding column to the target table before using that mapping.
Automated provisioning does not finish verification setup
On the verification side, Oracle adds MySQL and PostgreSQL family connections. Exact database services and versions remain subject to its support matrix. The service is powered by GoldenGate Veridata: Agents retrieve source and target data, while a Server coordinates comparison jobs and identifies differences.
The new Resource Manager service template provisions one Veridata Server and two Agents automatically. It supplies a prebuilt Terraform configuration, and Oracle recommends it within the Data Verification deployment-creation screen. That removes repeated resource-creation steps, but operators still have to define the connections and comparisons that make the environment useful.
From template to comparison job
Create a Resource Manager stack from the template, review its configuration and run an apply job to provision the Server and Agents.
Assign database connections to the appropriate Agents, then assign GoldenGate connections for those Agents to the Server. Open the Veridata console and verify the connections.
Configure Compare Pairs to associate source tables with target tables, organize those pairs into Groups, and create jobs to run and monitor the comparisons.
Reader comments
Newest comments first. Replies stay oldest first.