Web Agent for SharePoint

App Features

Start from the template, extend
for your organization

This is a generic web agent pattern for SharePoint: an SPFx web part with an AI assistant for analysis, upload, and follow-up. The included starter template wires in common office formats (for example .docx and .pdf) as a baseline—you can extend or replace extraction, prompts, and steps to match your file types, policies, and line-of-business needs. Content is sent to Azure OpenAI with your document guide (where to route files, types, and tagging hints). You get structured suggestions: narrative text, type, suggested upload URL (SharePoint library path), tags, and a short reason—then confirm with Upload. After a successful upload, a planner agent (strict JSON) powers chat: it can suggest new field values, show Field / Current / Proposed, and Update on the last item. Works on modern SharePoint and in Microsoft Teams when the host supports the web part. Developers can change the implementation to call Copilot for Microsoft 365 (or other Microsoft 365–native AI surfaces your tenant provides) so AI requests are handled inside your organization’s Microsoft 365 boundary instead of sending prompts to a separate Azure OpenAI endpoint you operate outside the SharePoint experience.

Download now

Analyze & stage upload

»» The starter flow drops a file, extracts text, and calls Azure OpenAI with your document guide and a structured prompt. Out of the box, that path includes example extractors for common office types (such as .docx and .pdf)—swap or add parsers, different AI steps, or inputs for a fully custom agent aligned to your org.
»» The model returns a reply plus fields such as type, suggested upload URL (server-relative library path), tags, and reason.
»» The UI stages a pending upload and shows an Upload action so you confirm before any file is written to SharePoint.

Upload to SharePoint

»» On confirm, the Document service uses PnPjs to upload the file to the suggested library and applies metadata where it maps to editable list columns (with optional follow-up for field mapping).
»» The upload confirmation path is a fast path: it runs the tool directly without an extra LLM call when you use the Upload button or type upload.

Planner chat & metadata updates

»» After upload, a planner-style agent uses a strict JSON contract: responseText, optional userOption (e.g. "update"), and proposedMetadata when proposing field changes.
»» You can ask for more tags, clarifications, or field changes; when the planner proposes an update, the chat shows an Update control (or you can type update) to apply values to the last uploaded item—also a fast path with no extra LLM round trip on commit.

Stack & production notes

»» Think of the package as a starting point for a SharePoint-native agent: the same UI and agent plumbing can be extended beyond the default document-drop experience for organizational or industry-specific workflows. UI: React + TypeScript, SPFx (see your solution version in package.json), SharePoint I/O via @pnp/sp.
»» Admins set AI API endpoint, API key, and the long-form document guide in web part properties (optional welcome text and out-of-scope fallback are supported).
»» Keys in browser properties and network calls can be visible to page editors; for production, prefer a server-side proxy (e.g. Azure Function or API Management) with Entra ID and keys in Key Vault / managed identity.
»» The sample routes AI through configurable endpoints; in your fork you can retarget the same flows to Copilot for Microsoft 365 and related Microsoft Graph–backed entry points (subject to license and product availability) so users’ AI interactions need not leave the SharePoint and broader Microsoft 365 trust surface you already govern.
»» Support for deployment and usage is available on request.

SharePoint · SPFx · AI