Snowflake and Copilot Studio connect via native data and inference links
What happened
- Microsoft made Snowflake a Copilot Studio knowledge source available on May 19, 2025, letting agents query selected Snowflake tables in real time. - Snowflake says its Cortex Agents integration lets Microsoft 365 Copilot and Teams users query structured and unstructured Snowflake data in chat. - Microsoft lists Snowflake knowledge in Copilot Studio release plans, while Snowflake provides setup guides for Cortex Agents in Teams and Microsoft 365 Copilot.
Why it matters
Microsoft and Snowflake have been building two related paths into the same enterprise workflow: direct data access from Copilot Studio into Snowflake tables, and Snowflake-hosted Cortex Agents surfaced inside Microsoft 365 Copilot and Teams. Microsoft’s 2025 release plan says Copilot Studio can connect directly to Snowflake as a knowledge source, while Snowflake’s documentation says Cortex Agents can answer questions over both structured and unstructured data inside Microsoft interfaces. Together, those pieces show how enterprise agents are being wired to reach business data without forcing users to leave the Microsoft stack. The integration matters because Copilot Studio and Microsoft 365 Copilot are increasingly being used as front ends for internal agents, while Snowflake is positioning Cortex as the inference layer over enterprise data. Microsoft says makers can select specific Snowflake tables as knowledge sources and let agents query, join and filter them from natural-language prompts. Snowflake says its Cortex Agents combine Cortex Analyst for structured data and Cortex Search for unstructured data, then expose the result through Microsoft surfaces. ### Which part is native inside Copilot Studio, and which part still runs in Snowflake? Microsoft’s release plan for 2025 Wave 1 says Snowflake became a Copilot Studio knowledge source with general availability on May 19, 2025. The document says makers can connect Copilot Studio directly to Snowflake data with a no-code interface, choose specific tables, and let the agent answer grounded questions without manual integration or scripting. It also says the data remains in Snowflake while Copilot provides real-time responses. Snowflake describes a separate but adjacent path. Snowflake’s quickstart says Cortex Agents orchestrate across structured and unstructured sources by using Cortex Analyst for SQL generation and Cortex Search for unstructured retrieval, then connect that agent into Microsoft Copilot through OAuth. Snowflake says the goal is to let customers build GenAI services in Snowflake and use Microsoft Copilot as the interface. (learn.microsoft.com) ### What data can these agents actually pull together? Snowflake’s quickstart uses a sales-intelligence example built from two source types: structured sales data and unstructured sales-call data. The guide says the combined agent can answer plain-text questions by drawing on both. That is the clearest documented example of the “structured plus unstructured” pattern behind the integration. (snowflake.com) Microsoft’s documentation on unstructured knowledge sources shows how Copilot Studio handles files and external content more broadly. Microsoft says uploaded files and connected content from OneDrive, SharePoint, Salesforce, ServiceNow, Confluence and Zendesk are ingested into Dataverse, chunked, vector-indexed and retrieved as relevant pieces when a user asks a question. (snowflake.com) ### How does this fit with the broader Microsoft-Snowflake partnership? Snowflake and Microsoft announced a Power Platform connector in November 2024 that enabled bidirectional access between Dataverse and Snowflake’s AI Data Cloud. Snowflake said at the time that developers and business users could build applications on Microsoft Power Platform with Snowflake data and do so without moving or copying data. Christian Kleinerman, Snowflake’s EVP of Product, said the companies were giving users tools to build AI applications “without the need to move or copy data.” (learn.microsoft.com) That earlier connector matters because Dataverse is Microsoft’s enterprise data platform for agents, and Copilot Studio sits inside the same Power Platform family. The newer Snowflake knowledge-source support in Copilot Studio extends that connection from app building into agent grounding. ### Where do permissions and traceability become the hard part? Snowflake’s Teams and Microsoft 365 Copilot documentation says data sent between Snowflake and Microsoft crosses the Snowflake service boundary and is then governed by the customer’s terms with Microsoft. (businesswire.com) That makes the handoff between retrieval, permissions and downstream use a concrete design issue, not an abstract one. Microsoft’s knowledge-source documentation says Copilot Studio retrieves the most relevant chunks matching a user query from indexed content, while the Snowflake quickstart says Cortex Agents identify tools and execute them across structured and unstructured sources. In practice, that means enterprises need to know which table, document chunk or search result shaped an answer before they trust agents to prepare meetings or trigger follow-up work. That traceability requirement is an inference from how both products describe retrieval and tool use. (docs.snowflake.com) ### Where can users see this next? Snowflake says Cortex Agents for Microsoft Teams and Microsoft 365 Copilot are available through Microsoft AppSource, and Microsoft lists Snowflake knowledge-source support in Copilot Studio’s release documentation. Snowflake’s developer quickstart and Microsoft Learn pages are the main public setup references for teams testing the connection today. (docs.snowflake.com) (learn.microsoft.com)
Key numbers
- Microsoft made Snowflake a Copilot Studio knowledge source available on May 19, 2025, letting agents query selected Snowflake tables in real time.
- Snowflake says its Cortex Agents integration lets Microsoft 365 Copilot and Teams users query structured and unstructured Snowflake data in chat.
- Microsoft lists Snowflake knowledge in Copilot Studio release plans, while Snowflake provides setup guides for Cortex Agents in Teams and Microsoft 365 Copilot.
- Microsoft and Snowflake have been building two related paths into the same enterprise workflow: direct data access from Copilot Studio into Snowflake tables, and Snowflake-hosted Cortex Agents surfaced inside Microsoft 365 Copilot and Teams.
What happens next
- Microsoft’s 2025 release plan says Copilot Studio can connect directly to Snowflake as a knowledge source, while Snowflake’s documentation says Cortex Agents can answer questions over both structured and unstructured data inside Microsoft interfaces.
- Microsoft’s release plan for 2025 Wave 1 says Snowflake became a Copilot Studio knowledge source with general availability on May 19, 2025.
- Snowflake said at the time that developers and business users could build applications on Microsoft Power Platform with Snowflake data and do so without moving or copying data.
Quick answers
What happened in Snowflake and Copilot Studio connect via native data and inference links?
Microsoft made Snowflake a Copilot Studio knowledge source available on May 19, 2025, letting agents query selected Snowflake tables in real time. Snowflake says its Cortex Agents integration lets Microsoft 365 Copilot and Teams users query structured and unstructured Snowflake data in chat. Microsoft lists Snowflake knowledge in Copilot Studio release plans, while Snowflake provides setup guides for Cortex Agents in Teams and Microsoft 365 Copilot.
Why does Snowflake and Copilot Studio connect via native data and inference links matter?
Microsoft and Snowflake have been building two related paths into the same enterprise workflow: direct data access from Copilot Studio into Snowflake tables, and Snowflake-hosted Cortex Agents surfaced inside Microsoft 365 Copilot and Teams. Microsoft’s 2025 release plan says Copilot Studio can connect directly to Snowflake as a knowledge source, while Snowflake’s documentation says Cortex Agents can answer questions over both structured and unstructured data inside Microsoft interfaces. Together, those pieces show how enterprise agents are being wired to reach business data without forcing users to leave the Microsoft stack. The integration matters because Copilot Studio and Microsoft 365 Copilot are increasingly being used as front ends for internal agents, while Snowflake is positioning Cortex as the inference layer over enterprise data. Microsoft says makers can select specific Snowflake tables as knowledge sources and let agents query, join and filter them from natural-language prompts. Snowflake says its Cortex Agents combine Cortex Analyst for structured data and Cortex Search for unstructured data, then expose the result through Microsoft surfaces. Which part is native inside Copilot Studio, and which part still runs in Snowflake? Microsoft’s release plan for 2025 Wave 1 says Snowflake became a Copilot Studio knowledge source with general availability on May 19, 2025. The document says makers can connect Copilot Studio directly to Snowflake data with a no-code interface, choose specific tables, and let the agent answer grounded questions without manual integration or scripting. It also says the data remains in Snowflake while Copilot provides real-time responses. Snowflake describes a separate but adjacent path. Snowflake’s quickstart says Cortex Agents orchestrate across structured and unstructured sources by using Cortex Analyst for SQL generation and Cortex Search for unstructured retrieval, then connect that agent into Microsoft Copilot through OAuth. Snowflake says the goal is to let customers build GenAI services in Snowflake and use Microsoft Copilot as the interface. (learn.microsoft.com) What data can these agents actually pull together? Snowflake’s quickstart uses a sales-intelligence example built from two source types: structured sales data and unstructured sales-call data. The guide says the combined agent can answer plain-text questions by drawing on both. That is the clearest documented example of the “structured plus unstructured” pattern behind the integration. (snowflake.com) Microsoft’s documentation on unstructured knowledge sources shows how Copilot Studio handles files and external content more broadly. Microsoft says uploaded files and connected content from OneDrive, SharePoint, Salesforce, ServiceNow, Confluence and Zendesk are ingested into Dataverse, chunked, vector-indexed and retrieved as relevant pieces when a user asks a question. (snowflake.com) How does this fit with the broader Microsoft-Snowflake partnership? Snowflake and Microsoft announced a Power Platform connector in November 2024 that enabled bidirectional access between Dataverse and Snowflake’s AI Data Cloud. Snowflake said at the time that developers and business users could build applications on Microsoft Power Platform with Snowflake data and do so without moving or copying data. Christian Kleinerman, Snowflake’s EVP of Product, said the companies were giving users tools to build AI applications “without the need to move or copy data.” (learn.microsoft.com) That earlier connector matters because Dataverse is Microsoft’s enterprise data platform for agents, and Copilot Studio sits inside the same Power Platform family. The newer Snowflake knowledge-source support in Copilot Studio extends that connection from app building into agent grounding. Where do permissions and traceability become the hard part? Snowflake’s Teams and Microsoft 365 Copilot documentation says data sent between Snowflake and Microsoft crosses the Snowflake service boundary and is then governed by the customer’s terms with Microsoft. (businesswire.com) That makes the handoff between retrieval, permissions and downstream use a concrete design issue, not an abstract one. Microsoft’s knowledge-source documentation says Copilot Studio retrieves the most relevant chunks matching a user query from indexed content, while the Snowflake quickstart says Cortex Agents identify tools and execute them across structured and unstructured sources. In practice, that means enterprises need to know which table, document chunk or search result shaped an answer before they trust agents to prepare meetings or trigger follow-up work. That traceability requirement is an inference from how both products describe retrieval and tool use. (docs.snowflake.com) Where can users see this next? Snowflake says Cortex Agents for Microsoft Teams and Microsoft 365 Copilot are available through Microsoft AppSource, and Microsoft lists Snowflake knowledge-source support in Copilot Studio’s release documentation. Snowflake’s developer quickstart and Microsoft Learn pages are the main public setup references for teams testing the connection today. (docs.snowflake.com) (learn.microsoft.com)
