Answered

Copilot Studio | SharePoint List as Knowledge Source 2,048 Row Limit

(1) Share

Report

Posted on by HB-09041311-0

Hi Community,

I am exploring the "SharePoint List as a knowledge source" feature in Microsoft Copilot Studio and came across a point of confusion.

As per the official documentation on requirements and quotas:
Reference Link:

https://learn.microsoft.com/en-us/microsoft-copilot-studio/requirements-quotas
It mentions that- "SharePoint list queries only return data from the first 2,048 rows of data."

However, in the upcoming feature (currently in preview as of April 2026, planned for GA in May 2026):

Reference Link:

https://learn.microsoft.com/en-us/power-platform/release-plan/2026wave1/microsoft-copilot-studio/add-sharepoint-lists-as-knowledge-source#feature-details
It states that- "When you select a specific list, the agent can access all rows and columns."

Could you please help clarify the following:

Will the new SharePoint list knowledge source (once generally available) removes the 2048 row limitation?
What should be considered for production use during the preview phase until this feature reaches GA?

Any clarification or guidance would be really helpful.

Thanks in advance!

Categories:

Copilot Studio skills development

General topics

I have the same question (0)

All responses (2)

Answers (2)

Sort by

Verified answer

rezarizvii 352 on at

Like (2)

Report
Copy link

Link copied!

Hi, hope you are doing well.

I'd say you shouldn't think of it as the 2,048 limit being fully “removed”. It’s more that Microsoft is changing how the data is handled. Instead of hitting that hard query cap, the system indexes the list and retrieves relevant pieces when needed. So technically it can “access all rows”, but in practice it still depends on retrieval, ranking, and token limits. You’re not going to get the entire list back in one go.

As for production during preview, I’d be a bit careful. It works, but results can be inconsistent with larger lists, and indexing isn’t always instant. If the data is critical or needs to be exact every time, better to use an action (direct query) instead of relying only on the knowledge source. For lighter, search-style scenarios it’s fine, but I wouldn’t fully trust it just yet.

Reza M. Rizvi
LinkedIn | Email
If this solved your issue, please mark it as ✅ Accepted Answer. If it helped, feel free to give it a 🩷 Like!

Was this reply helpful? Yes No
Verified answer

Sayali Microsoft Employee on at

Like (3)

Report
Copy link

Link copied!

Hello HB-09041311-0 ,

The apparent discrepancy between the current 2,048‑row limitation and the upcoming SharePoint List knowledge source capability in Copilot Studio stems from a shift in how data is retrieved rather than an actual removal of limits. Today, when SharePoint Lists are used as knowledge sources, the agent relies on ingestion‑based grounding—meaning the list data is embedded and indexed into a semantic snapshot at ingestion time. Due to model token and grounding constraints, only the first ~2,048 rows are processed, which is why this limit is documented in the current requirements. However, the new preview feature (April 2026, planned GA in May 2026) introduces runtime query‑based grounding, allowing the agent to dynamically fetch only the relevant rows during each query instead of embedding the entire list in advance. As a result, the agent can effectively access all rows and columns in a SharePoint list, since it retrieves data live using APIs rather than relying on ingestion. While this approach bypasses the previous limit, it introduces new production considerations during preview—such as runtime latency, API throttling, and query reliability—so organizations should consider implementing fallback mechanisms (e.g., Power Automate–based list retrieval) until the feature reaches GA and becomes fully supported for enterprise workloads.
Reference Document-Add SharePoint lists as a knowledge source | Microsoft Learn

-----------------------------------------------------------------------------------------------------------------------------------------------
Your feedback is important to us. Please rate us:

🤩 Excellent 🙂 Good 😐 Average 🙁 Needs Improvement 😠 Poor

Was this reply helpful? Yes No