Share via

Are Files and Vector Stores Shared Across Microsoft Foundry v1 and v2

Meenakshi Kathiresan 110 Reputation points
2026-04-29T16:31:08.0366667+00:00

I’m migrating from the older Azure AI Foundry agent experience (using PersistentAgentClient in the C# SDK) to the newer experience (using ProjectClient).

Within the same Foundry project, I’ve noticed:

  • Files and vector stores created in the older (v1) experience are visible and accessible in the new (v2) experience
  • Agents are not shared between the two experiences

This raises a few questions:

  1. Are files and vector stores officially project-scoped resources that are shared across both v1 and v2 experiences?
  2. Is this behavior guaranteed and supported, or just an implementation detail?
Foundry Tools
Foundry Tools

Formerly known as Azure AI Services or Azure Cognitive Services is a unified collection of prebuilt AI capabilities within the Microsoft Foundry platform


2 answers

Sort by: Most helpful
  1. Karnam Venkata Rajeswari 3,575 Reputation points Microsoft External Staff Moderator
    2026-04-29T16:35:12.5933333+00:00

    Hello @Meenakshi Kathiresan ,

    Welcome to Microsoft Q&A .Thank you for reaching out to us.

    The observed behavior during migration from the v1 experience (PersistentAgentClient) to the v2 experience (ProjectClient) aligns with how resources are structured within the Foundry project model.

    As asked -

    1. Files and vector stores across v1 and v2 Files and vector stores operate as resources associated with the Foundry project and are not owned by a specific agent definition. A vector store is created independently and then attached to agents or conversations as needed. As a result:
      • Files created in v1 remain accessible in v2 when used within the same project
      • Vector stores created earlier can continue to be reused across newer agent workflows
      • The same vector store can be associated with multiple agents or conversations within that project scope
    2. Behavior supportability and expectations The reuse of files and vector stores across different experiences is consistent with the current architecture, where these artifacts are managed as project-level data resources rather than agent-specific entities.

    Agents are not shared between versions as agent definitions are handled differently from data resources. During migration it is suggested that runtime components such as agents and related configurations are version-scoped and must be recreated when moving between experiences.

    This results in:

    • Agents created in v1 not appearing in v2
    • Recreation of agents being required during migration
    • Separation between agent runtime configuration and reusable data resources

    To ensure a stable and forward-aligned migration , please

    • Reuse existing files and vector stores without duplication
    • Recreate agent definitions using the v2 (ProjectClient) model
    • Transition workflows fully to v2 instead of maintaining mixed usage across versions

    If any inconsistencies are observed, please check if the following help

    1. If files or vector stores not visible then
      1. Confirm both workflows are pointing to the same Foundry project endpoint
      2. Validate subscription, tenant, and project alignment
      3. Ensure appropriate role-based access permissions are assigned
    2. For vector store behavior issues
      1. Please verify that the correct vector store is attached in the v2 agent configuration
      2. Then confirm ingestion has completed successfully before querying
      3. Revalidate data indexing if retrieval results appear inconsistent
    3. For agent availability differences
      1. Please confirm expected behavior due to version-specific agent models
      2. Reconfigure tools, bindings, and settings in the v2 agent setup

    The following references might be helpful , please check them out

    Thank you

    Was this answer helpful?

    1 person found this answer helpful.
    0 comments No comments

  2. AI answer

    2026-04-29T16:31:18.28+00:00
    1. Files and vector stores are project-scoped resources in the new Foundry agent experience. Vector stores are created and then attached to agents or conversations, and a single vector store can be attached to multiple agents or conversations within the same project. The vector store itself is not agent-scoped; it is a separate resource that the file search tool uses for keyword and semantic search.
    2. The v1→v2 transition guidance explicitly states there is no in-place migration and no cross-version references for versioned resources such as workspaces, projects, agents, tools, and knowledge bases. Those must be recreated under v2 and cannot reference each other across versions. However, that guidance does not list files or vector stores as versioned resources that must be recreated or that are blocked from cross-version use. The migration guide for the C# AzureAIAgent also shows that vector stores and files are managed via updated APIs (agentsClient.Files.*, agentsClient.VectorStores.*) but does not introduce a version boundary for those resources.

    Given the documented rules:

    • Version-scoped resources (workspaces, projects, agents, tools, knowledge bases) are explicitly separated between v1 and v2.
    • Files and vector stores are treated as shared, reusable assets that can be attached to agents and conversations, and they are not called out as needing recreation or being version-scoped.

    Therefore, the behavior observed—files and vector stores being visible and usable from both the classic and new agent experiences within the same project—is aligned with the documented model of vector stores as project-level resources and is not described as an unsupported or accidental implementation detail in the provided documentation.


    References:

    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.