Learn how to host your agents on Microsoft Foundry
We’ve just wrapped up our series on how to Host your agents on Foundry, a three-part livestream event where we demonstrated how to deploy and host Python AI agents using Microsoft Foundry:
- Deploying Python agents via Foundry Hosted agents with the Azure Developer CLI
- Creating hosted agents using the Microsoft Agent Framework, including integrating with Foundry IQ and managing multi-agent workflows
- Developing hosted agents with LangChain + LangGraph, introducing built-in features like Bing Web Search
- Conducting quality and safety evaluations: bulk evaluations, scheduling, continuous assessments, establishing guardrails, and red-teaming
All the resources from our series are available for further learning, and you can find them linked below:
- Recordings of each livestream
- PowerPoint slides perfect for reviewing or sharing with your own community
- Open-source code samples ready for you to try in your own Microsoft Foundry project
¿Hablas español? Consulta la versión en español de la serie.
Got more questions? Join us for the weekly Python+AI office hours on Foundry Discord.
During our first session, we deployed agents created with the Microsoft Agent Framework (the evolution of Autogen and Semantic Kernel). We started small, introducing Foundry tools like Code Interpreter, integrating the agent with enterprise data using Foundry IQ, and extending it to include multi-agent workflows. We also made use of the Foundry UI to interact with the hosted agent, testing its capabilities in the playground and examining the traces of its reasoning and tool usage.
In our second session, we showcased deploying agents that utilise popular open-source libraries, LangChain and LangGraph. We began with a basic agent, added Foundry tools like Bing Web Search, and then advanced to deploying more complicated agents using the LangGraph orchestration framework. Again, we engaged with the hosted agent via the Foundry UI, testing it within the playground and observing its reasoning and tool interaction traces.
In the third session, we focused on ensuring our AI agents deliver high-quality results while operating safely and responsibly. We started by discussing what constitutes high-quality outputs, utilising built-in evaluators to assess overall task adherence. We also created custom evaluators for specific domain checks. With Foundry hosted agents, we facilitated bulk evaluations on demand, established scheduled evaluations, and even set up continuous assessments for select live agent traces. In addition, we explored safety systems that can be integrated with agents and conducted audits for potential safety risks. To align with an organization’s objectives, we configured custom policies and guardrails that could be shared across agents. Lastly, we safeguarded against unsafe outputs from adversarial inputs by running automated red-teaming scans on agents, even scheduling these to occur regularly.
Frequently Asked Questions
- What is Microsoft Foundry?
- Microsoft Foundry is a platform designed for deploying and managing AI agents, enabling developers to create sophisticated solutions backed by Microsoft technologies.
- How can I access the series materials?
- You can find video recordings, PowerPoint slides, and code samples linked in the content above. They’re free to access and great for learning!
- What are the benefits of hosting AI agents on Foundry?
- Hosting AI agents on Foundry allows for seamless integration with Microsoft tools, extensive customization options, and robust evaluations to ensure quality and safety.
- Can I join the weekly office hours even if I missed the livestreams?
- Absolutely! The weekly Python+AI office hours are open to everyone, providing a great opportunity to get your questions answered and connect with others in the community.
Share this content:
Discover more from Qureshi
Subscribe to get the latest posts sent to your email.
