Power Automate AI advisor uses SharePoint and Claude

Key insights

• The demo uses SharePoint-hosted client documents as the advisor’s controlled information source.
• Power Automate acts as the orchestration layer between the document source and Claude.
• The described design includes prompt text, access codes and usage logs.
• An HTML chat widget provides the user-facing interface in the demonstrated architecture.

Microsoft published a 10-minute demo showing how to build a domain-specific AI advisor with Power Automate, SharePoint and external large language model APIs. Presenter OluwaMayowa Ogbeide describes an advisor that uses SharePoint-hosted client documents and Claude, with Power Automate coordinating the process. The main takeaway is architectural: the model is paired with a controlled document set rather than being presented as a general-purpose source of answers.

What's new

The video description identifies the elements in the demonstration: client documents hosted in SharePoint, prompt text, access codes, usage logs, an HTML chat widget, Power Automate and Claude. Together, these form a proposed pattern for a focused AI advisor. The description says the system answers questions from a controlled set of documents instead of giving generic responses.

That description establishes the components, but not a full build guide. It does not specify the flow actions, how SharePoint content is selected, the API request format, or how answers are displayed or checked. Those details matter to anyone trying to reproduce the design, so the video should be treated as an architectural demonstration rather than a complete implementation specification.

Why it matters

For makers and consultants, the example is relevant because it brings together familiar Power Platform and SharePoint components with an external model. SharePoint is the document host in the described design, while Power Automate coordinates the connection to the model. The chat widget gives users a way to ask questions, and access codes and usage logs are named as parts of the approach.

For business users, the intended difference is the source of the answer: the advisor is described as working from a defined document set, rather than responding generically. That framing can be useful where the question concerns a particular body of client information. However, the description does not establish that answers are always accurate, cite their sources, or respect each user’s document permissions.

For administrators, the use of an external model raises implementation questions that the published description does not resolve. It does not say how credentials are protected, what information is sent to Claude, or how access codes relate to SharePoint permissions. Nor does it identify licensing, rollout stage, tenant requirements or regional availability. The demo therefore gives no basis for assuming a particular deployment will be available or approved in every organization.

Readers evaluating the workflow can compare it with other Power Automate examples, while recognizing that this video’s description is the only source here for the advisor’s specific design.

How it works

The published description supports a high-level outline, but it does not provide a numbered sequence of flow steps. At that level, the named components suggest the following relationship: users interact through an HTML chat widget; SharePoint holds the client documents; and Power Automate connects the interaction with Claude. Prompt text, access codes and usage logs are also included in the described setup.

1. A user asks a question through the HTML chat widget, according to the described architecture.
2. The advisor uses SharePoint-hosted client documents as its controlled information source.
3. Power Automate serves as the orchestration layer connecting the advisor’s process to Claude, an external model.
4. Prompt text, access codes and usage logs are among the other elements named in the demo description.

This is a component-level summary, not a verified account of the flow’s internal sequence. The available description does not say whether the flow searches, filters or retrieves particular documents for each question, or how it formats information for the model. It also does not explain what happens when the documents do not contain an answer, when an API call fails, or when a user lacks access.

Limitations and open questions

The primary limitation is the amount of implementation detail available in the published material. The description names the architecture and its components, but does not include a complete technical transcript or detailed build instructions. That means readers cannot use this information alone to reproduce the flow or assess its exact behavior.

Several operational details remain unclear: how documents are chosen and updated, whether the advisor returns citations, how usage is logged, and how access codes are issued or managed. The description also does not discuss data handling by the external model, security controls, error handling, or the costs of operating the integration. These are questions to investigate before using a similar pattern with real client material, not capabilities confirmed by the demo description.

The video is useful as a concise overview of a possible combination of Power Automate, SharePoint and an external LLM. It does not establish a Microsoft product feature, a supported reference architecture, or a general availability claim. Readers should consult the video itself for the demonstrated context and seek implementation specifics before treating the outline as a deployment plan.

Sources

• Microsoft’s domain-specific AI advisor demo

Frequently asked questions

Which model does the demo use?

The description names Claude as the external model used in the demonstration.

What does Power Automate do in the design?

The description identifies Power Automate as the orchestration layer. It does not provide the flow’s exact actions or configuration.

Does the material explain how SharePoint documents are selected?

No. It says the advisor uses SharePoint-hosted client documents, but does not explain searching, filtering or retrieval.

Does the video establish licensing or availability?

The available description does not specify licensing, rollout stage, tenant requirements or regional availability.