Oracle APEX adds multi‑LLM Mistral

Oracle’s APEX 26.1 release adds something enterprise developers have been asking for: built-in support for multiple large-language-model providers inside the same low-code platform. In practical terms, that means Mistral is now a first-class option alongside Google Gemini, Anthropic Claude, Ollama and OpenAI-compatible APIs, rather than something teams must wire up themselves through custom integrations. (blogs.oracle.com) Oracle announced APEX 26.1 general availability on May 14, and Oracle’s APEX team detailed the expanded provider support in a May 30 blog post. (blogs.oracle.com) The immediate change is not a new model from Oracle. It is a new abstraction layer inside APEX. Oracle says developers can configure these providers through the same declarative service model already used for OCI Generative AI, OpenAI and Cohere, with existing applications continuing to work without reconfiguration. (blogs.oracle.com) ### Where does Mistral actually show up in APEX 26.1? Oracle’s May 30 product post lists Mistral AI among the new out-of-the-box providers added in APEX 26.1, together with Google Gemini, Anthropic Claude, Ollama and OpenAI-compatible APIs. The same post says the feature is exposed through the Generative AI Services page in Application Builder, where a workspace defines the service once and then reuses it across AI-powered components. (blogs.oracle.com) Oracle’s APEX 26.1 documentation also names Mistral as a supported provider and says APEX maps provider-specific requests into a normalized internal format so applications receive a consistent response shape across models and providers. For Mistral specifically, Oracle says the service follows an OpenAI-like API pattern but has separate endpoints and extra parameters for chat and embeddings. (blogs.oracle.com) ### What changes for developers using more than one model vendor? Oracle says the setup happens through a single configuration page. In that flow, developers name the service, choose the provider type, enter the base URL, select a model and attach credentials stored in APEX Web Credentials. Oracle says those secrets are stored in the credential repository and are not exposed to application code or the browser. (blogs.oracle.com) Oracle’s documentation says that once a service is saved, “every AI-powered component in the app routes through the configured service,” and swapping providers later is “a single configuration update.” That matters for teams comparing models by task, or keeping a backup provider available if price, performance or regional constraints change. (docs.oracle.com) That last point is an inference from Oracle’s design, not a stated company claim. ### Why did Oracle include OpenAI-compatible endpoints and Ollama too? Oracle’s APEX team said OpenAI-compatible APIs are supported to enable “local and custom endpoints such as LM Studio,” while Ollama support covers chat and embeddings for local LLM execution. Oracle’s documentation says local LLM frameworks can run inside customer-controlled infrastructure for sensitive workloads or offline scenarios while keeping the same remote-server-and-credential model inside APEX. (blogs.oracle.com) That means Oracle is not only adding named commercial providers. It is also giving customers a path to use self-hosted or locally exposed models without rebuilding the application layer each time. Oracle frames that as part of making AI integration “declarative and enterprise ready.” (blogs.oracle.com) ### Which parts of APEX use these configured AI services? Oracle says configured AI services can be made available to App Builder features including AI Assistant and Create Page from Natural Language. Oracle’s broader APEX 26.1 release materials also highlight AI Agents and AI Interactive Reports as major additions in this version. (blogs.oracle.com) Oracle’s next step for users is already published in its APEX 26.1 documentation: admins create a Generative AI Service object, choose Mistral or another provider, set the model and endpoint, and then expose that service to application features in the workspace. (blogs.oracle.com)