Share
Report
40I’m building a Copilot Studio agent in Teams where users upload DOCX and PDF files for analysis (summarization, extraction, comparison, etc.).
I’ve been using AI Builder so far, but I’m running into practical limitations:
- File size constraints
- Token limits for larger documents or batches of files
- Limited flexibility when users upload many documents in a single session
I’m exploring alternative architectures and would appreciate guidance from others who’ve solved this at scale.
A few specific questions:
- Dataverse as file storage – Is Dataverse a recommended approach for storing uploaded documents for Copilot Studio scenarios?
- Are there best practices for using Dataverse file columns vs alternatives (e.g., SharePoint, Azure Blob + references)?
- Cost considerations – How expensive does Dataverse become for unstructured file storage at scale?
- In practice, does Dataverse file storage get costly compared to external storage options?
- Processing patterns – For large or many files, is the preferred pattern to:
- Pre‑process documents outside Copilot Studio (e.g., chunking, indexing, embeddings), then
- Let Copilot Studio orchestrate over processed outputs rather than raw files?
I have the same question (0)
All responses (
Answers (
