Microsoft Agent Framework Development Services
We design, build and run production agents and multi-agent workflows on Microsoft Agent Framework, in .NET (C#) and Python. We deploy them to Microsoft Foundry, Azure or your own servers, and bring them into Microsoft Teams and Microsoft 365 Copilot.
Registered in Jodhpur since 2022
Dewiride Technologies Private Limited
Agent Framework tutorials
Step-by-step C# videos on our YouTube channel
Microsoft 365 Agents SDK tutorials
Setup and bot-building guides on our developer blog
Open-source projects
GST invoicing, web analytics, a PowerPoint editor and an ERP
What is Microsoft Agent Framework?
Microsoft Agent Framework is Microsoft’s open-source framework for building AI agents and multi-agent workflows in .NET and Python, with its source code on GitHub (opens in a new tab) under the MIT License. Microsoft describes it as the direct successor to Semantic Kernel and AutoGen, created by the same teams.
Version 1.0 for .NET and Python became generally available in April 2026. Microsoft calls it “the production-ready release: stable APIs, and a commitment to long-term support”. That commitment covers the stable packages; many integrations are still in preview or beta.
A Go version is in public preview, and Microsoft says its APIs may change.
Agents
A language model with instructions, tools and memory that decides which tools to call to answer a request. Use one when the route to an answer varies from request to request.
Workflows
A process you define in code or YAML: which agents and functions run, in what order, and where a person approves. Use one when the steps are known and must be repeatable.
Many systems combine the two: a workflow keeps the process predictable, and agents handle the steps that need judgement. Unsure whether you need an agent? See our comparison of agents, chatbots and workflows.
What we build with it
These are the systems we build with Microsoft Agent Framework. Each starts as a small pilot on your own tasks.
Agents that use your tools
Agents call your own C# or Python methods as function tools, and tools on MCP servers, so they reach your systems only through interfaces you control. Each tool gets the narrowest access that does the job.
Multi-agent workflows
Agents working together through Agent Framework’s five built-in orchestration patterns: sequential, concurrent, handoff, group chat and Magentic. Microsoft announced all five as stable in .NET and Python in July 2026, though some options are still experimental. Workflows save checkpoints so a run can resume after an interruption, and can pause for a person’s approval.
Long-running work with the Agent Harness
Microsoft describes the Agent Harness as a batteries-included agent for long-running, multi-step work: it plans, tracks a to-do list and applies standing approval rules to tool calls. Its core has been stable since July 2026; context compaction, background agents, file access and looping are still experimental.
Knowledge agents
Agents that answer from your documents and data, showing their sources. Agent Framework’s context providers add relevant material before each call; the retrieval underneath is the same work as our custom AI on your own data.
Agents in Teams and Microsoft 365 Copilot
Microsoft documents using Agent Framework for the AI logic inside a Microsoft 365 Agents SDK agent, which handles the channels and conversation state. We build on that pattern; see our Teams and Microsoft 365 Copilot agent development.
Durable agents (preview)
The Durable Task extension for Agent Framework (the Durable Extension in Microsoft’s docs) persists agent sessions and checkpoints multi-agent orchestrations and workflows, on Azure Functions or your own compute. Its .NET and Python packages are prerelease.
Migrating from Semantic Kernel or AutoGen
Microsoft is putting most new features into Microsoft Agent Framework rather than Semantic Kernel. In its October 2025 statement on Semantic Kernel (opens in a new tab), it committed to critical bug and security fixes for Semantic Kernel 1.x for at least a year after Agent Framework became generally available in April 2026.
AutoGen is in maintenance mode: its README on GitHub (opens in a new tab) says it gets no new features, is community managed and accepts only bug fixes, security patches and documentation changes.
Microsoft publishes guides for migrating from Semantic Kernel (opens in a new tab) and migrating from AutoGen (opens in a new tab), and our approach builds on them:
Inventory and tests first
We list every agent, plugin, prompt and model call, and write tests that capture today’s behaviour.
Map the concepts
Using Microsoft’s guides, we map each Semantic Kernel or AutoGen construct to its Agent Framework equivalent and flag anything without a direct match.
Migrate in small pieces
One agent or workflow at a time, starting where Agent Framework helps most.
Run old and new side by side
Both versions handle the same requests, and we compare answers, traces and costs before live traffic moves.
Cut over
Once the tests and comparison pass, we switch traffic, keep a route back for a while, then remove the old packages.
Where Agent Framework fits in Microsoft’s agent stack
Microsoft’s agent products overlap, and several of them can be used together.
| Product | What it is | Use it when |
|---|---|---|
| Microsoft Agent Framework | Open-source framework for agents and multi-agent workflows in .NET and Python; Go is in public preview. | You want full control in code over models, tools, workflows and hosting, or are leaving Semantic Kernel or AutoGen. |
| Foundry Agent Service | Managed platform on Microsoft Foundry for building, deploying and scaling agents. Its hosted agents, generally available since July 2026, run your own code, Agent Framework included. | You want Microsoft to operate the agent: managed endpoint, automatic scaling, Microsoft Entra agent identity and observability. |
| Microsoft 365 Agents SDK | SDK for conversational agents in C#, JavaScript and Python that run across channels such as Microsoft 365 Copilot, Teams and websites. | One agent must reach several channels, including websites or custom apps; Agent Framework can supply its AI logic. See our Microsoft 365 Agents SDK development. |
| Teams SDK | SDK for agents and apps in Teams, formerly the Teams AI Library, generally available in C#, TypeScript/JavaScript and Python. Its agents can also reach Microsoft 365 Copilot. Its own AI libraries are deprecated, so it pairs with an AI framework such as Agent Framework. | The agent is built specifically for Teams, such as a collaborative agent in chats, channels and meetings. |
| Copilot Studio | Graphical, low-code studio for building and managing agents and workflows, fully managed by Microsoft. | A business team wants to build and change the agent itself without heavy development work. We recommend it when it fits. |
| Microsoft Agent 365 | Microsoft 365 service, generally available for commercial customers since May 2026, for observing, governing and securing agents across an organisation, including third-party ones. Separate from the Microsoft 365 Agents SDK. | You need one place to observe and govern the agents across your organisation. |
Where your agents run
Microsoft’s hosting guide for Microsoft Agent Framework (opens in a new tab) gives two models: Foundry hosted agents, run by Microsoft, or self-hosting in your own application, plus the Durable Extension for Azure Functions. We choose with you, based on who should operate the agent and whether you can accept prerelease packages.
Foundry hosted agents
Foundry Agent Service runs your agent code with a managed endpoint, automatic scaling, session state, observability and a dedicated Microsoft Entra agent identity. The service has been generally available since July 2026; Agent Framework’s Foundry hosting packages are still prerelease. Hosted agents support Python and C# and run only in listed Azure regions; private networking for them must be configured when the Foundry account is created. Hosting is billed on the CPU and memory used during active sessions, on top of model and tool usage.
Your own app or container
Agent Framework runs inside your own ASP.NET Core or Python application on Azure Container Apps or Azure App Service, as in Microsoft’s tutorials and reference architectures, or on your own servers. Its .NET self-hosting packages are prerelease. See our step-by-step guide to deploying a .NET app to Azure Container Apps (opens in a new tab).
Azure Functions with the Durable Extension (preview)
For Azure Functions triggers, durable execution or long-running orchestration, Microsoft points to the Durable Extension. We keep your agent code separate from the hosting code, so it can move to another host later.
Enterprise controls
What we build into Microsoft Agent Framework agents before production:
- Middleware on agent runs, tool calls and model calls, for logging, security checks and error handling. It is part of Agent Framework’s stable core.
- OpenTelemetry traces, logs and metrics from Agent Framework, which by default follow the latest experimental version of the GenAI semantic conventions. On Foundry, tracing for hosted agents is generally available, with traces stored in Azure Monitor Application Insights.
- Human approval for sensitive actions, such as payments or messages to customers, through human-in-the-loop requests in workflows or the Agent Harness’s tool approval.
- A dedicated Microsoft Entra agent identity for each agent hosted on Foundry, with only the access it needs. Microsoft Entra Agent ID is generally available, and Microsoft states that extending Entra security features to agents requires Microsoft Agent 365.
- Foundry guardrails (formerly content filters): generally available for model deployments, in preview for agents. They cover Foundry Models sold by Azure and agents built in Foundry Agent Service.
- An evaluation set of real tasks, run before every release. Foundry evaluations are generally available, though some evaluators are in preview.
- Stable packages by default. Each design lists any prerelease package it relies on; in Python we install packages individually, because the agent-framework meta-package also installs prerelease integrations.
Your models, your cloud
Microsoft Agent Framework is not tied to one model provider, though only some of its connectors are stable so far:
| Provider | .NET | Python |
|---|---|---|
| OpenAI | Stable | Stable |
| Azure OpenAI | Stable | Stable |
| Microsoft Foundry | Preview | Stable |
| Anthropic | Preview | Beta |
| Amazon Bedrock | Vendor or community IChatClient package | Beta |
| Google Gemini | Vendor or community IChatClient package | Beta |
| Ollama | Vendor or community IChatClient package | Beta |
| Mistral | No Agent Framework package | Beta |
In .NET, any model service with a Microsoft.Extensions.AI IChatClient implementation can back an agent, which is how Bedrock, Gemini and Ollama connect.
We choose a model per task: a more capable one where the reasoning is hard, and a smaller, cheaper one for routine steps such as classification. Each choice is tested against your evaluation set and kept in configuration, so it can change later. Agents call only the model services you approve.
How we work and what it costs
Discovery call
We talk through the process you want to automate, your systems and your constraints, and tell you plainly whether Microsoft Agent Framework, Copilot Studio or something simpler fits. To start, book a free discovery call (opens in a new tab).
Pilot with success criteria
A small pilot on one real process, with success criteria agreed before work starts: the tasks it must handle, the quality it must reach and the cost per run. Scope and timeline go into the statement of work.
Production build
Identity, tracing, guardrails, evaluations and a deployment pipeline, on the hosting you chose.
Handover and support
Code, documentation and a walkthrough for your team, then support Monday to Friday, 9 AM–6 PM IST, as needed or on a monthly retainer.
- Consulting
- From ₹2,500 (US$40) an hour
- Monthly retainer
- From ₹2,00,000 (US$4,000) a month
- Pilots and fixed-scope projects
- Quoted per statement of work
GST is added to Indian invoices. Rate cards by business size and add-ons are on our pricing page.
Learn Agent Framework with us
If your team is new to Microsoft Agent Framework, our Agent Framework tutorials on YouTube go through it step by step in C#, from a first agent to tools, memory and workflows.
Sources
- Microsoft Agent Framework overview (opens in a new tab) — Microsoft Learn
- Microsoft Agent Framework Version 1.0 (opens in a new tab) — Microsoft
- Semantic Kernel to Agent Framework migration guide (opens in a new tab) — Microsoft Learn
- AutoGen to Microsoft Agent Framework migration guide (opens in a new tab) — Microsoft Learn
- Agent Framework’s orchestration patterns reach 1.0 (opens in a new tab) — Microsoft
Last reviewed by Jagdish Kumawat, Founder & Director
Frequently Asked Questions
Is Microsoft Agent Framework (MAF) production-ready?
Should we migrate from Semantic Kernel now?
What about our AutoGen projects?
Should we use .NET, Python or Go?
Does Agent Framework lock us into Azure?
Do we need Foundry Agent Service?
Can our agents work in Teams and Microsoft 365 Copilot?
When is Copilot Studio the better choice?
Does Agent Framework support MCP and A2A?
How do we monitor agents in production?
What does a pilot cost?
Can we meet in person?
Related services
- Microsoft Foundry & Foundry Agent ServiceProjects on Microsoft Foundry: choosing and deploying models, running agents on Foundry Agent Service, and evaluating them before and after release.
- Teams Bots & Microsoft 365 Copilot AgentsAgents and bots that people use inside Microsoft Teams and Microsoft 365 Copilot, built with the Microsoft 365 Agents SDK, including moves off the Bot Framework SDK.
- AI Agent DevelopmentAgents that call your systems, follow your rules and hand over to a person when they should, proved in a small pilot before they go to production.
Tell us what your agent should do
We'll tell you the simplest approach that will work — Microsoft Agent Framework, Copilot Studio or no agent at all — and what a pilot would involve.