Retrieve content under the user’s permissions
Connect your application to SharePoint using the authentication and permissions approved by your organization. Apply those permissions before selecting passages for the model request.
Keep source identifiers with the retrieved content. They allow the application to show where an answer came from and let a support agent open the original document.
Build the model request
Combine the question, a bounded set of relevant passages and instructions for how to handle missing information. Send the request through TextCortex using a model available to your account.
Use an EU-hosted route when required for the material. The regional review should also cover your retrieval service, logs and storage.
Ground and validate the answer
Ask the model to cite only the supplied source identifiers. Validate those identifiers before rendering links, and handle cases where the documents do not support an answer.
Keep support actions such as sending a reply or changing a ticket outside the generation step. Add authorization and review at the point where an external action is executed.
Evaluate model choices without changing retrieval
Use the same approved passages to compare GPT, Claude, Gemini, DeepSeek or other supported models. Score source support, answer usefulness and request cost.
The knowledge-agent guide describes this architecture, and the API quickstart shows how to connect the model client.