AI agents powered by reasoning models can take minutes to work through complex problems \u2014 deep research, multi-step analysis, lengthy content generation. In a traditional request-response pattern, that means your client sits idle waiting for a connection that may time out, or worse, fails silently and loses all progress. Background responses in Microsoft Agent Framework let you offload these long-running operations so your application stays responsive and resilient, regardless of how long the agent takes to think.<\/p>\n
With background responses, you start an agent task and get back a continuation token<\/strong> instead of blocking until completion. Your application can poll for results on its own schedule, resume interrupted streams from exactly where they left off, and handle network hiccups without restarting work from scratch. This matters whether you’re building internal tools for enterprise workflows or customer-facing products where reliability is non-negotiable. Background responses are available in both .NET and Python.<\/p>\n When you enable background responses and send a request to an agent, what happens depends on whether the underlying agent and model support background processing:<\/p>\n The continuation token captures the current state of the operation. When it’s Background responses are a good fit when your agent tasks involve:<\/p>\n The simplest approach is to start a background run, then poll until the result is ready.<\/p>\n Key points:<\/strong><\/p>\n For a more responsive experience, you can stream results in real time while still benefiting from background processing. Each streamed update carries a continuation token, so if the connection drops, you can pick up right where you left off.<\/p>\n Key points:<\/strong><\/p>\n One of the practical benefits of background responses is resilience to connectivity issues. Because the agent continues processing server-side regardless of what happens to the client connection, you get fault tolerance without building it yourself:<\/p>\n For production applications, consider storing continuation tokens persistently (in a database or cache) so operations can survive not just network blips but full application restarts. This is especially valuable for enterprise scenarios where agent tasks may run for several minutes.<\/p>\n An enterprise compliance team uses an agent to generate regulatory filings. These documents require the model to reason through complex guidelines and can take several minutes. With background responses, the application submits the request, shows the user a progress indicator, and polls for the result \u2014 no risk of the request timing out mid-generation.<\/p>\n A financial services agent performs deep analysis across market data, SEC filings, and news. The reasoning model needs time to synthesize information from multiple sources. Background responses let the application kick off the analysis, free up the UI, and notify the user when results are ready.<\/p>\n A data engineering team runs agents over large datasets to extract insights. Each item may require extended reasoning. Background responses allow the pipeline to submit tasks in parallel, poll for completions, and handle individual failures without losing progress on the rest.<\/p>\n Background responses are fully supported when used with the Responses API for OpenAI<\/a><\/strong> and Azure OpenAI<\/a><\/strong> providers \u2014 via The amount of time you can access response data is subject to the data retention policy of the underlying service.<\/p>\n Background responses are available in both the .NET and Python SDKs. To get started:<\/p>\n We’re always interested in hearing from you. If you have feedback or questions, reach out to us on the GitHub discussion boards<\/a>. And if you’ve been enjoying Agent Framework, give us a \u2b50 on GitHub<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":" Handling Long-Running Operations with Background Responses AI agents powered by reasoning models can take minutes to work through complex problems \u2014 deep research, multi-step analysis, lengthy content generation. In a traditional request-response pattern, that means your client sits idle waiting for a connection that may time out, or worse, fails silently and loses all progress. […]<\/p>\n","protected":false},"author":157200,"featured_media":5207,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[78,143,34],"tags":[],"class_list":["post-5202","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-net","category-agent-framework","category-python-2"],"acf":[],"blog_post_summary":" Handling Long-Running Operations with Background Responses AI agents powered by reasoning models can take minutes to work through complex problems \u2014 deep research, multi-step analysis, lengthy content generation. In a traditional request-response pattern, that means your client sits idle waiting for a connection that may time out, or worse, fails silently and loses all progress. […]<\/p>\n","_links":{"self":[{"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/posts\/5202","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/users\/157200"}],"replies":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/comments?post=5202"}],"version-history":[{"count":0,"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/posts\/5202\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/media\/5207"}],"wp:attachment":[{"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/media?parent=5202"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/categories?post=5202"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/agent-framework\/wp-json\/wp\/v2\/tags?post=5202"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}How It Works<\/h2>\n
\n
null<\/code> (.NET) or None<\/code> (Python), the operation is complete \u2014 either the response is ready, the operation failed, or further input is required.<\/p>\nWhen to Use Background Responses<\/h2>\n
\n
Non-Streaming: Poll for Completion<\/h2>\n
.NET<\/h3>\n
AIAgent agent = new AzureOpenAIClient(\r\n new Uri(\"https:\/\/<myresource>.openai.azure.com\"),\r\n new DefaultAzureCredential())\r\n .GetResponsesClient(\"<deployment-name>\")\r\n .AsAIAgent();\r\n\r\nAgentRunOptions options = new()\r\n{\r\n AllowBackgroundResponses = true \/\/ Enable background responses\r\n};\r\n\r\nAgentSession session = await agent.CreateSessionAsync();\r\n\r\n\/\/ Start the run\u2014 may complete immediately or return a continuation token\r\nAgentResponse response = await agent.RunAsync(\r\n \"Write a detailed market analysis for the Q4 product launch.\", session, options);\r\n\r\n\/\/ Poll until complete\r\nwhile (response.ContinuationToken is not null)\r\n{\r\n await Task.Delay(TimeSpan.FromSeconds(2));\r\n\r\n options.ContinuationToken = response.ContinuationToken;\r\n response = await agent.RunAsync(session, options);\r\n}\r\n\r\nConsole.WriteLine(response.Text);\r\n<\/code><\/pre>\nPython<\/h3>\n
import asyncio\r\nfrom agent_framework.openai import OpenAIResponsesClient\r\n\r\nagent = OpenAIResponsesClient(model_id=\"o3\").as_agent(\r\n name=\"researcher\",\r\n instructions=\"You are a helpful research assistant.\",\r\n)\r\n\r\nsession = await agent.create_session()\r\n\r\n# Start a background run\r\nresponse = await agent.run(\r\n messages=\"Write a detailed market analysis for the Q4 product launch.\",\r\n session=session,\r\n options={\"background\": True},\r\n)\r\n\r\n# Poll until the operation completes\r\nwhile response.continuation_token is not None:\r\n await asyncio.sleep(2)\r\n response = await agent.run(\r\n session=session,\r\n options={\"continuation_token\": response.continuation_token},\r\n )\r\n\r\nprint(response.text)\r\n<\/code><\/pre>\n\n
null<\/code>\/None<\/code>, you have the final result.<\/li>\n<\/ul>\nStreaming with Resumption<\/h2>\n
.NET<\/h3>\n
AgentRunOptions options = new()\r\n{\r\n AllowBackgroundResponses = true \/\/ Enable background responses\r\n};\r\n\r\nAgentSession session = await agent.CreateSessionAsync();\r\n\r\nAgentResponseUpdate?latestUpdate = null;\r\n\r\nawait foreach (var update in agent.RunStreamingAsync(\r\n \"Write a detailed market analysis for the Q4 product launch.\", session, options))\r\n{\r\n Console.Write(update.Text);\r\n latestUpdate = update;\r\n\r\n \/\/ Simulate a network interruption\r\n break;\r\n}\r\n\r\n\/\/ Resume from exactly where we left off\r\noptions.ContinuationToken = latestUpdate?.ContinuationToken;\r\nawait foreach (var update in agent.RunStreamingAsync(session, options))\r\n{\r\n Console.Write(update.Text);\r\n}\r\n<\/code><\/pre>\nPython<\/h3>\n
session = await agent.create_session()\r\n\r\nlast_token = None\r\n\r\n# Start streaming with background enabled\r\nasync for update in agent.run(\r\n messages=\"Write a detailed market analysis for the Q4 product launch.\",\r\n stream=True,\r\n session=session,\r\n options={\"background\": True},\r\n):\r\n last_token = update.continuation_token\r\n if update.text:\r\n print(update.text, end=\"\", flush=True)\r\n\r\n # Simulate a network interruption\r\n break\r\n\r\n# Resume the stream using the last continuation token\r\nif last_token is not None:\r\n async for update in agent.run(\r\n stream=True,\r\n session=session,\r\n options={\"continuation_token\": last_token},\r\n ):\r\n if update.text:\r\n print(update.text, end=\"\", flush=True)\r\n<\/code><\/pre>\n\n
Built-In Fault Tolerance<\/h2>\n
\n
Use Cases<\/h2>\n
Document Generation and Review<\/h3>\n
Research and Analysis<\/h3>\n
Batch Processing Pipelines<\/h3>\n
Best Practices<\/h2>\n
\n
Supported Agents<\/h2>\n
ChatClientAgent<\/code> in .NET and Agent<\/code> in Python. The A2AAgent<\/code> has limited support for background responses, with improvements planned for an upcoming release.<\/p>\nLearn More<\/h2>\n
\n