Microsoft Foundry Agent Service
Managed runtime for AI agents on Azure
Foundry Agent Service is Microsoft's managed platform for building, deploying and scaling AI agents on Azure. It was first called Azure AI Agent Service, then Azure AI Foundry Agent Service; since late 2025 it has been part of Microsoft Foundry. There are three kinds of agents: prompt agents without code, hosted agents running your own code in a managed container (for example with Microsoft Agent Framework, LangGraph or the OpenAI Agents SDK) and voice agents in preview. Tools include web search, file search, code interpreter, memory, OpenAPI, MCP servers and the A2A protocol. Models come from the Foundry catalog, including OpenAI GPT models, Claude, Llama, Mistral and DeepSeek. For several collaborating agents there were connected agents (classic version only, retiring on 31 March 2027) and workflows in preview, which Microsoft retires on 1 December 2026. Going forward Microsoft recommends Microsoft Agent Framework or A2A for this. Commerzbank runs its customer service assistant Ava on this service.
Good fit if
Focus: Companies on Azure that want to run agents as a managed service with Microsoft identity, networking and compliance
- your project is about Customer service assistants, Internal knowledge and process agents or Agents with your own code in a managed container
- someone on your team writes code
Probably not if
- nobody on your team writes code (2/5 for use without developers)
Key figures
- Use without coding
- 2/5
- Scalability
- 5/5
- Integrations
- 5/5
- Versioning
- 4/5
- Observability
- 4/5
Editorial rating on a scale from 1 to 5. Methodology
Fact sheet
- Category
- Enterprise
- Type
- Managed runtime for AI agents on Azure
- Licence
- Proprietär
- Pricing
- Usage-basedThere is no separate fee for prompt agents and workflows. You pay for model tokens and tools: file search per GB of vector storage per day (1 GB free), code interpreter per session, web search per 1,000 queries. Hosted agents are billed per vCPU-hour and GiB-hour of memory; idle time is not charged. Current rates are on the Azure pricing page.
- Deployment
- Cloud, Azure
- Models
- OpenAI, Anthropic, Meta, Mistral, DeepSeek
- Compliance
- GDPR
- RAG
- Yes
- Agents working as a team
- Built in
- SSO, roles (RBAC)
- SSO yes, RBAC yes
Key features
- Prompt agents without code and hosted agents with your own code
- Tools: web search, file search, code interpreter, OpenAPI, MCP, A2A
- Models from the Foundry catalog, including GPT, Claude, Llama, Mistral, DeepSeek
- Own Entra identity per agent, roles via Azure RBAC
- Tracing in Application Insights, built-in evaluators
- Versions with rollback, private networking (VNet)
Typical use cases
- Customer service assistants
- Internal knowledge and process agents
- Agents with your own code in a managed container
- Connecting existing systems via OpenAPI and MCP
Sources
- Website(opens in a new tab)
- Dokumentation: Überblick(opens in a new tab)
- Multi-Agent-Funktion (Workflows, Abschaltung 01.12.2026)(opens in a new tab)
- Connected Agents (klassisch)(opens in a new tab)
- Migration und Funktionsstatus(opens in a new tab)
- Preise(opens in a new tab)
- Regionen und Limits(opens in a new tab)
- Datenschutz und Sicherheit(opens in a new tab)
- Einsatz bei der Commerzbank (Anbieterangabe)(opens in a new tab)
Checked on 11 Oct 2026
Market notes
Changes to Microsoft Foundry Agent Service by email
We let you know when the status, pricing or license of Microsoft Foundry Agent Service changes, together with the weekly market notes.
Implementation
Get Microsoft Foundry Agent Service implemented
Free for you. Partners only pay for requests they accept.
Alternatives
Similar systems
Similar in category, features or use cases.
- DeveloperMicrosoft Agent FrameworkMicrosoft Agent Framework is Microsoft's open-source framework for agents and multi-agent workflows in Python and .NET. Version 1.0 shipped in April 2026 and is the unified successor of Semantic Kernel and AutoGen. A graph-based workflow engine supports branching, parallelism, checkpoints and human approvals; ready-made patterns exist for sequential, concurrent, handoff, group chat and Magentic-One orchestration. MCP and A2A are supported.View profile
- No-codeMicrosoft Copilot StudioMicrosoft Copilot Studio is Microsoft's low-code environment for building agents with natural language and a graphical editor. Agents can access Microsoft 365, Dataverse and the Power Platform and can be published to Teams, websites and other channels. Billing is based on Copilot Credits. In Microsoft 365 Copilot, the Researcher agent can also call connected agents that administrators enable (connected agents).View profile
- EnterpriseAgentforce (Salesforce)Agentforce is Salesforce's agent platform that companies use to build and run AI agents for customers and employees. The agents use data, workflows and permissions from Salesforce and Data Cloud and can be customized with low-code or pro-code tools. Guardrails, testing and monitoring are built in.View profile
- EnterpriseOracle AI Agent StudioOracle AI Agent Studio is the environment where Oracle customers and partners create and customize agents and agentic workflows for Oracle Fusion Cloud Applications such as ERP, HCM and SCM. Workflows include Multi Agent nodes in which a supervisor agent routes requests to specialized worker agents and returns a consolidated response for requests spanning several topics. According to Oracle, the studio supports MCP and A2A agent cards as well as models from several providers. Oracle states that AI Agent Studio is available to Fusion customers at no additional cost; Custom AI is a separate offering.View profile