Microsoft Agent Framework

Microsoft Agent Framework

Trigger On

• building or reviewing .NET code that uses Microsoft.Agents.*, Microsoft.Extensions.AI, AIAgent, HarnessAgent, AgentSession, or Agent Framework hosting packages
• choosing between ChatClientAgent, Responses agents, hosted agents, custom agents, Anthropic agents, workflows, or durable agents
• adding the batteries-included Microsoft.Agents.AI.Harness surface for planning, todos, compaction, file memory/access, tool approvals, skills, shell execution, or background agents
• authoring preview-era Microsoft.Agents.AI.Workflows.Declarative* packages or wrapping a workflow with workflow.AsAIAgent()
• adding tools, MCP, A2A, OpenAI-compatible hosting, AG-UI, DevUI, background responses, or OpenTelemetry
• migrating from Semantic Kernel agent APIs or aligning AutoGen-style multi-agent patterns to Agent Framework
• using Anthropic Claude models (haiku, sonnet, opus) via AnthropicClient or through Azure Foundry with AnthropicFoundryClient

Workflow

1. Decide whether the problem should stay deterministic. If plain code or a typed workflow without LLM autonomy is enough, do that instead of adding an agent.
2. Choose the execution shape first: single AIAgent, batteries-included HarnessAgent, explicit programmatic Workflow, workflow-as-agent wrapper, declarative workflow when YAML portability is explicitly required, Azure Functions durable agent, ASP.NET Core hosted agent, AG-UI remote UI, or DevUI local debugging.
3. Choose the agent type and provider intentionally. Prefer the simplest agent that satisfies the threading, tooling, and hosting requirements.
4. Keep agents stateless and keep conversation or long-lived state in AgentSession. Treat the session as opaque provider-owned state, serialize it through the owning agent, and never accept a raw service conversation ID as an end-user authorization boundary.
5. Add only the tools and middleware that the scenario needs. Narrow the tool surface, require approval for side effects, and treat MCP, A2A, and third-party services as trust boundaries.
6. For workflows, model executors, edges, typed RequestPort boundaries, checkpoints, shared state, and human-in-the-loop explicitly rather than hiding control flow in prompts.
7. Prefer Responses-based protocols for new remote/OpenAI-compatible integrations unless you specifically need Chat Completions compatibility.
8. Use durable agents only when you truly need Azure Functions serverless hosting, durable thread storage, or deterministic long-running orchestrations.
9. Verify preview status, package maturity, docs recency, and provider-specific limitations before locking a production architecture.

Current Upstream Notes

• .NET 1.20.0 fixes Foundry-hosted workflow cancellation and duplicate AgentHost port binding, preserves Responses logprobs, and adds a timeout for background-agent wait-for-first-completion. Re-test cancellation, timeout, recovery, and streamed metadata with the selected provider.

• Current AG-UI hosted web-search samples use Responses. Update the Cosmos connector name from CommunityToolkit.VectorData.CosmosNoSql to CommunityToolkit.VectorData.AzureCosmosDB when following the migrated sample. Retired OpenAI Assistants integration tests were removed; choose an active provider API for new work.

• dotnet-1.19.0 adds persisted routing and sessions, resilient/steerable hosted agents, AG-UI forwarding, and experimental agent hooks. It makes a breaking move to the MCP 2026-07-28 Tasks extension; update both peers and resume tests together.

• Current AG-UI hosting uses Microsoft.Agents.AI.Hosting.AGUI.AspNetCore with AddAGUIServer() and MapAGUIServer(...); the client uses AGUI.Client, and conversation state flows through AgentSession. Do not copy older AddAGUI/MapAGUI or AgentThread hosting examples into current applications.

• The bundled August 2026 snapshot contains 172 pages. Start with the September review for current checkpoint and AG-UI guidance, then load only the relevant topic reference. Verify language and package maturity against the linked live page.

Architecture

flowchart LR
  A["Task"] --> B{"Deterministic code is enough?"}
  B -->|Yes| C["Write normal .NET code or a plain workflow"]
  B -->|No| D{"One dynamic decision-maker is enough?"}
  D -->|Yes| O{"Needs a packaged long-task runtime?"}
  O -->|No| E["Use an `AIAgent` / `ChatClientAgent`"]
  O -->|Yes| P["Use `HarnessAgent` with scoped capabilities"]
  D -->|No| F["Use a typed `Workflow`"]
  F --> G{"Needs durable Azure hosting or week-long execution?"}
  G -->|Yes| H["Use durable agents on Azure Functions"]
  G -->|No| I["Use in-process workflows"]
  E --> J{"Need a remote protocol or UI?"}
  P --> J
  F --> J
  J -->|OpenAI-compatible HTTP| K["ASP.NET Core Hosting.OpenAI"]
  J -->|Agent-to-agent protocol| L["A2A hosting"]
  J -->|Web UI protocol| M["AG-UI"]
  J -->|Local debug shell| N["DevUI (dev only)"]

Core Knowledge

• AIAgent is the common runtime abstraction. It should stay mostly stateless.
• AgentSession owns conversation state. Create, reuse, serialize, and restore it through the owning agent; keep service IDs server-side and verify user or tenant ownership before resumption.
• AgentResponse and AgentResponseUpdate are not just text containers. They can include tool calls, tool results, structured output, reasoning-like updates, and response metadata.
• ChatClientAgent is the safest default when you already have an IChatClient and do not need a hosted-agent service.
• HarnessAgent is a prerelease packaged ChatClientAgent composition for tool loops, planning, todos, compaction, memory, approvals, telemetry, and optional file, web, shell, skill, loop, or background-agent capabilities. Enable only what the task needs.
• Microsoft Foundry Agents is the canonical Azure-hosted persistent-agent surface. Azure OpenAI Responses is the app-composed Azure option for tool approval, code interpreter, file search, web search, and MCP.
• Workflow is an explicit graph of executors and edges. Use it when the control flow must stay inspectable, typed, resumable, or human-steerable.
• workflow.AsAIAgent() is the escape hatch when a complex workflow needs to present a normal agent surface. It keeps sessions, streaming, and agent response APIs, but the workflow start executor still needs chat-message-compatible input.
• AgentWorkflowBuilder provides high-level factory methods such as BuildConcurrent for common agent orchestration patterns. Use it when you need concurrent or sequential agent pipelines without writing custom executor classes.
• Sequential orchestration passes the previous agent's full input-and-response conversation forward by default. Choose response-only context deliberately when later stages should not inherit the entire conversation.
• Current .NET workflow execution uses InProcessExecution.RunStreamingAsync(...). For sensitive agent tools, wrap the function with ApprovalRequiredAIFunction, listen for RequestInfoEvent with ToolApprovalRequestContent, and send the external approval response back through the run.
• Handoff is a mesh-style transfer of task ownership between agents, not a primary-agent tool call. In the current C# docs it requires locally tool-capable agents; Python-only autonomous handoff, approval, or checkpoint examples are not evidence of equivalent .NET APIs.
• Declarative workflows are now a documented surface, but the .NET package/runtime story is still preview-heavy and narrower than programmatic workflows. Use YAML when portability and operator-editable orchestration matter; keep deeply custom .NET control flow programmatic.
• Hosting layers such as OpenAI-compatible HTTP, A2A, and AG-UI are adapters over your in-process agent or workflow. They do not replace the core architecture choice.
• Durable agents are a hosting and persistence decision for Azure Functions. They are not the default answer for ordinary app-level orchestration.
• Prefer canonical middleware, tool, integration, migration, support, and upgrade pages from the local docs index when exact signatures or maturity matter.