---
title: What does Microsoft Foundry Agent Service standard setup keep in your own Azure resources?
url: https://opencontroller.lyzr.com/blog/foundry-agent-service-standard-setup-own-resources
published: 2026-11-15
updated: 2026-11-15
author: Lyzr
---

# What does Microsoft Foundry Agent Service standard setup keep in your own Azure resources?

**In standard setup, Foundry Agent Service stores agent data at rest in single-tenant resources you own: conversation threads and agent metadata in Azure Cosmos DB for NoSQL, uploaded files in Azure Storage, and vector stores in Azure AI Search, with Azure Key Vault for secrets. Basic setup uses Microsoft-managed resources instead. Choose standard setup when you need data sovereignty, your own encryption and access policies, or project-level isolation.**

Security reviews of Foundry agents often stall on one question: where do the conversations live? Standard setup gives a clear answer, with some design decisions you cannot easily undo.

## What lands in each resource?

| Resource you provide | Role | What it stores |
|---|---|---|
| Azure Cosmos DB for NoSQL | Thread storage | Messages, conversation history, agent metadata |
| Azure Storage | File storage | Files uploaded by developers and end users |
| Azure AI Search | Vector stores | Vector stores created by agents |
| Azure Key Vault | Secrets | Connection secrets (required) |
| Application Insights | Telemetry | Traces and metrics (optional) |

Microsoft's documentation states that all data processed by Agent Service is stored at rest in these resources in standard setup.

## How does project-level isolation work?

Standard setup isolates projects by default. Each project gets its own blob containers, one for files and one for intermediate system data such as chunks and embeddings, and its own Cosmos DB containers for user threads, system messages and agent configuration. That means two teams sharing a Foundry account do not share agent state, and you can scope access per project.

## What does it require?

1. **Cosmos DB throughput.** Microsoft lists a minimum of 3,000 RU/s total, in provisioned or serverless mode, multiplied by the number of projects. Too little throughput causes provisioning failures.
2. **Role assignments for the deploying identity:** Storage Blob Data Contributor on the storage account and Cosmos DB Operator on the Cosmos DB account.
3. **Role assignments for the project managed identity**, including Cosmos DB Operator, Storage Account Contributor, Search Index Data Contributor, Search Service Contributor, and data roles scoped to the project's blob containers.
4. **Developer access** through the Foundry User role at project scope (renamed from Azure AI User; role IDs did not change).

## What decisions are hard to reverse?

- **Capability settings are fixed per project.** Agent Service creates connections to your resources during provisioning, and they cannot be changed on an existing project. To switch resources you delete and recreate the project, which loses its threads unless you migrate them.
- **Account defaults cascade.** Settings on the Foundry account become defaults for its projects. Set them deliberately before teams start building.
- **Runtime versions use different containers.** The classic and new agent runtimes write to different Cosmos DB containers, so retention and backup rules must cover both.

## What should a governance review check?

1. Is every production project on standard setup, or are some on basic setup with Microsoft-managed storage?
2. Do Cosmos DB, Storage and AI Search sit in approved Regions for your data residency rules?
3. Are customer-managed keys, network rules and diagnostic logs applied to those resources like any other data store?
4. Is there a retention policy for threads in Cosmos DB? Conversation history accumulates and often contains personal data.
5. Who holds data-plane roles? Anyone with read access to the containers can read conversations, outside Foundry's own access controls.
6. Is private networking configured where required? See our guide to [network isolation for Foundry agents](/blog/network-isolation-for-microsoft-foundry-agents).

Keeping data in your own resources is one part of the picture. You still need an inventory of which agents exist, which models and tools they use, and how they behave. Our post on [governing Microsoft Foundry agents](/blog/govern-microsoft-foundry-agents) covers that side, and [data residency for LLM and agent traffic](/blog/how-to-enforce-data-residency-for-llm-and-agent-traffic) covers model calls that leave the Region.

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across Azure, AWS and Google Cloud and records Foundry agents in a governed registry with their versions, tools and MCP servers. Posture flags risky permission and configuration findings, and the runtime adapter for Microsoft Foundry extends observation and enforcement to Foundry workloads. Opencontroller itself can run as Lyzr SaaS or on customer-controlled infrastructure. See the [AI agent control plane](/blog/what-is-an-ai-agent-control-plane) model or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**What is the difference between basic and standard setup in Foundry Agent Service?**
Basic setup uses Microsoft-managed resources. Standard setup stores agent data in your own Cosmos DB, Storage and AI Search.

**Can I switch an existing Foundry project to my own resources?**
No. Capability settings cannot be updated on an existing project; you recreate the project.

**What Cosmos DB throughput does standard setup need?**
Microsoft lists a minimum of 3,000 RU/s total, multiplied by the number of projects.

## Key facts

Microsoft Foundry Agent Service standard setup: customer-owned Azure Cosmos DB for NoSQL (threads, messages, agent metadata), Azure Storage (files), Azure AI Search (vector stores), Azure Key Vault required, Application Insights optional; all Agent Service data at rest stored there. Basic setup uses Microsoft-managed resources. Project-level isolation via per-project blob and Cosmos containers. Cosmos minimum 3,000 RU/s total, times number of projects. Capability settings cannot be changed on an existing project. Developer role: Foundry User (renamed from Azure AI User). Opencontroller (Lyzr) discovers Azure agents, has a Microsoft Foundry runtime adapter and Posture checks; deploys as Lyzr SaaS or customer-controlled infrastructure. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern agents in Microsoft Foundry?](https://opencontroller.lyzr.com/blog/govern-microsoft-foundry-agents.md)
- [How do you run Microsoft Foundry agents on a private network?](https://opencontroller.lyzr.com/blog/network-isolation-for-microsoft-foundry-agents.md)
- [How do you enforce data residency for LLM and AI agent traffic?](https://opencontroller.lyzr.com/blog/how-to-enforce-data-residency-for-llm-and-agent-traffic.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
