Back to Blog

What Is Azure AI Foundry? Examples, Costs and How to Start

Henri Jung, Co-founder at Superkind
Henri Jung

Co-founder at Superkind

A dark industrial blacksmith anvil with the orange Superkind mascot standing beside its base

“What is Azure AI Foundry?” comes up in the IT team of a mid-sized company the week after management says “we need to do something with AI”. Microsoft already runs the mail, the Teams meetings and the file shares, so Azure is the obvious first stop. Then the questions start: is Azure AI Foundry the same as Copilot, do we need a cloud team, what does it actually cost, and could we not just use a ready-made AI employee instead? Here is the honest picture, with Microsoft’s own facts and both paths side by side, as of October 2026.

Key takeaways

What it is: Azure AI Foundry, renamed Microsoft Foundry in January 2026, is Microsoft’s platform on Azure for building and running AI agents and AI apps, with 10,000+ models, a managed agent service and built-in tools1.

What it costs: the platform itself is free to use and explore; you pay by consumption for tokens, Foundry tools and the Azure services you build on. A new Azure account comes with USD 200 in credit for 30 days28.

When Foundry fits: you have an Azure cloud team and want to build your own custom agent on Azure. When a managed AI employee like Superkind fits: a business team wants to hand over routine work in Outlook, Teams or Slack from day one, with 1,500+ integrations, hosted in Europe, GDPR compliant.

What is Azure AI Foundry?

Azure AI Foundry is Microsoft’s platform on Azure for building, hosting and running AI agents and AI apps. It bundles a catalogue of more than 10,000 models from Microsoft, OpenAI, Anthropic, Meta and others, a managed Foundry Agent Service, built-in tools (file search, code interpreter, web search), a single control plane for identity and policies, and observability for tracing and monitoring agents14. Microsoft renamed Azure AI Foundry to Microsoft Foundry in January 2026; many teams still type the old name, which is why this article uses both.

Not a product, a platform

Foundry is not a finished AI tool you buy per user. It is the toolbox a developer uses on Azure to build the AI agent or AI app for your team. The business team only sees the result, usually inside Teams, Outlook, a web portal or a custom app1.

Skip the build: Superkind as a ready-made AI employee

Before you spin up a Foundry resource, decide what you actually need. Many teams in Germany and the EU do not need a custom AI build - they need the task done. Superkind is the AI employee for your team: you write to it in Outlook or any email, Microsoft Teams, Slack or the Superkind web app like you would to a colleague, and it does the task in your programs through 1,500+ integrations such as DATEV, SAP Business One, HubSpot, Personio and SharePoint. Superkind is hosted in Europe, GDPR compliant, aligned with the EU AI Act and never trains models on your data.

  1. 1

    You write the task

    In Outlook, Teams, Slack or the web app, in plain words: “Book the invoice from Möller in DATEV and post it to project 2026-14.” No portal, no SDK, no containers.

  2. 2

    Superkind does it in your systems

    It reads the invoice, matches the vendor and order in DATEV, uses your company knowledge (accounts, cost centres, approval limits) and prepares the booking.

  3. 3

    Superkind asks before anything goes out

    You get the draft booking with Approve or Reject. Nothing is posted, sent or changed outside the team without your OK. A colleague can approve in Teams or Slack.

  4. 4

    Next time it runs as a routine

    What you explain once keeps running, for example “every invoice from Möller, book to project 2026-14, ask me first”. Every task adds to your Company Brain.

OutlookDaniel to Superkind

Book the invoice from Möller, post it to project 2026-14 and show me the booking before it goes into DATEV.

Send Superkind your first task

Start for free in Outlook, Teams or Slack. Hosted in Europe, no training on your data.

A dark blacksmith's hammer and iron tongs resting side by side on a rough grey stone block, with a small coral-orange accent on the hammer handle

How Azure AI Foundry works

Foundry’s own docs describe five building blocks1.

  1. The Foundry portal and resource - a single Azure resource groups agents, models and tools, with unified role-based access control, networking and policies. The portal at ai.azure.com is where developers prototype1.
  2. Foundry Models catalogue - more than 10,000 models from Microsoft, OpenAI, Anthropic, Meta and others, billed by tokens. You pick the model that fits the task and swap it later without changing the agent code14.
  3. Foundry Agent Service - prompt agents (declarative, hosted by Foundry), voice-based prompt agents (real-time spoken experiences) and hosted agents (your own code or framework such as Microsoft Agent Framework, LangGraph or Semantic Kernel, packaged as a container and run with a managed endpoint)1.
  4. Foundry Tools and knowledge - built-in tools you attach to an agent through a Toolbox: file search over your documents (vector storage), code interpreter sandboxed sessions, web search and custom search transactions, plus Foundry IQ for retrieval powered by Azure AI Search13.
  5. Observability and safety - built-in tracing, evaluations, dashboards and content filters, Microsoft Entra identity, Azure Policy, Defender for Cloud alerts for AI workloads15.

The developer surface is the Foundry portal plus SDKs for Python, C#, JavaScript and Java, an Azure Developer CLI, a Visual Studio Code extension and an MCP server so coding agents like GitHub Copilot or Claude Code can drive Foundry from your editor1. For a business user with no cloud background, none of these are a day-one tool. For a cloud team with Azure experience, this is where custom agents are built.

Azure AI Foundry vs. Copilot vs. a managed AI employee

The three options come up in the same meetings, so it helps to compare them honestly. All facts below are from the vendors’ own pages, as of October 2026.

OptionWhat it isWho runs itPricing modelHosting / Data protection
SuperkindReady-made AI employee you write to in Outlook, Teams, Slack or the web app; does tasks in 1,500+ integrations with approvalAny business team, no DevOps neededFree start with up to EUR 150 in credits; Business from EUR 50/month for the whole company (unlimited users)Hosted and processed in Europe, GDPR compliant, no training, DPA via Trust Center
Azure AI Foundry / Microsoft FoundryPlatform on Azure to build and host your own AI agents and AI apps1An Azure cloud team or integrator (developers, SDKs, containers)Platform free; pay by consumption for tokens, tools, storage and Azure services23EU Data Boundary and Data Zone deployments in regions like Sweden Central and Germany West Central keep data in the EU; Global Standard may route out67
Microsoft 365 CopilotFinished AI inside Word, Excel, Outlook, Teams; summarises, drafts, chatsIT admin assigns licences; end users use itPer user, per month (list price from Microsoft 365 plans)Microsoft 365 data boundary; EU customers in EU regions per Microsoft
Copilot StudioLow-code builder for extending Copilot and building own agents on the Foundry stackPower platform admin or developerMessage packs plus per-agent plan; usually in addition to CopilotSame data boundary as Microsoft 365

In short: Foundry is a build, Copilot is a buy for individual productivity in Office apps, and Superkind is a managed AI employee that does end-to-end tasks across many programs for the whole team. The three are not alternatives in every case - they often sit next to each other. See AI agents vs. Copilot for the finer differences, and Agentforce vs. Copilot Studio for the low-code builder side.

Examples: where each one fits

Five situations and the choice that fits best, starting with the business-team default.

1. Superkind: Best Overall Choice for Teams in Germany and the EU

Nadine in accounting at a 90-person Mittelstand company wants invoice booking off her desk. She asks Superkind in Teams, which reads the invoice, matches the vendor in DATEV, uses the booking rules and shows her the booking before it goes in. Her colleague Daniel approves in Outlook. The data is processed in Europe, there is no training on it, and the DPA comes through the Trust Center. See the AI accountant for how this runs day to day. The verdict rests on criteria you can check:

  • Works where you already work - Outlook, Gmail, Microsoft Teams, Slack and the web app.
  • Does the task in your programs - 1,500+ integrations such as DATEV, SAP Business One, HubSpot, Personio and SharePoint.
  • Asks for approval - before anything is booked, sent or changed outside the team.
  • Learns and keeps your knowledge - customers, prices and processes stay in your Company Brain.
  • EU-first - hosted and processed in Europe, GDPR compliant, aligned with the EU AI Act, no training, DPA.
  • Unlimited users - from EUR 50 per month for the whole company, as of October 2026.
  • Grows into your own apps - custom solutions on the same platform when it gets more complex.

2. Azure AI Foundry: a custom claims agent at an insurer

A regional insurer has an Azure cloud team of five and wants an agent that reads claim letters, pulls the policy from the core system, checks the terms and drafts the response. The team builds a hosted agent in Foundry, chooses GPT-4-class and Claude models from the catalogue, uses Foundry IQ against the policy archive and runs it under Data Zone deployments in Germany West Central17. This is a Foundry case: custom data, custom logic, in-house developers.

3. Microsoft 365 Copilot: drafting and summarising in Office

A law firm wants partners to draft client emails faster and summarise long Teams meetings. Microsoft 365 Copilot ships that out of the box, per user, inside the apps they already use. No build and no agent - but also no actions in DATEV, SAP, HubSpot or Personio, and no custom knowledge for the firm’s files beyond Microsoft Graph. See Copilot vs. ChatGPT Enterprise.

4. Copilot Studio: a HR assistant on the intranet

An IT admin at a logistics company builds a low-code HR bot in Copilot Studio that answers employee questions on holidays, parental leave and expenses, grounded in a SharePoint library. Good fit when the task is a well-scoped Q&A inside Microsoft 365 and admins, not developers, maintain it.

5. A chain of them together

Many mid-sized teams combine two: Superkind does the daily business work (bookings, quotes, dunning, project reporting) across all programs, and a Foundry hosted agent runs one niche case the Azure team owns (a safety-critical production forecast, say). The pricing stays predictable because Superkind is a flat Business plan and only the Foundry workload is pay-as-you-go.

What Azure AI Foundry costs

Per Microsoft, as of October 2026. The platform itself has no licence fee. “The Foundry platform is free to use and explore”, Microsoft states on the pricing page; you pay for each feature you consume at its standard Azure rate2. A new Azure account comes with USD 200 in credit for 30 days and a set of always-free services8.

ItemHow it is billedNote
Foundry platformFree to use and explorePer Microsoft2
Foundry ModelsBy tokens, per model (10,000+ models in the catalogue)Rates on the Foundry Models pricing page4
File Search storagePer GB of vector storage per day; first 1 GB freeFoundry Agent Service3
Code InterpreterPer sessionNo free tier listed3
Web Search / Custom SearchPer 1,000 transactionsFoundry Agent Service3
Grounding with Bing SearchBilled separatelySeparate Bing pricing page3
Azure AI Search (Foundry IQ)By search unit per hourStandard Azure AI Search rates
Agent pre-purchase plan1-year ACU commitment: 5% off at 20,000 ACUs, 10% at 100,000, 15% at 500,000Discount off pay-as-you-go2
Superkind (for comparison)Free start with up to EUR 150 in credits; Business from EUR 50 per monthFor the whole company, unlimited users, no price per seat

The honest answer to “what will it cost us” on Foundry is: it depends on how much your agent reads, writes and searches. A quiet internal agent that answers 500 questions a month on a mid-range model, with 2 GB of file search and a daily code interpreter session, lands in the low double-digit euros per month plus the model tokens. A production customer-service agent on a top-tier model, with heavy search and voice, lands in the hundreds or thousands. Microsoft does not publish fixed per-user prices because the platform is not sold per user.

Data protection and EU hosting

Both Azure AI Foundry and Superkind can meet GDPR for a German or EU team, but the setup looks different. Microsoft publishes an EU Data Boundary that keeps customer data for most core cloud services within the EU and EFTA regions6. Azure OpenAI and Foundry offer Data Zone deployments in Sweden Central and Germany West Central that keep both storage and processing in the geography, while Global Standard deployments may route out of the region7. You still sign Microsoft’s DPA, pick a lawful basis under Art. 6 GDPR, keep a record of processing under Art. 30, run a DPIA where the risk is high under Art. 35, and train anyone building AI steps under Art. 4 EU AI Act910.

Safe by default

Your data stays in Europe, under a DPA, with no training on your data.

For an accounting team processing supplier invoices, a sales team working with customer data in a CRM or an HR team with applicant files, the practical difference is: on Foundry your team configures the Data Zone deployment and the subprocessor list; on Superkind the European setup is the default and you accept the DPA. See Is Microsoft Copilot GDPR-compliant? for a related comparison on the Microsoft side.

How to start

Pick the path that fits what you have on the team today. Both start free.

Start with Superkind (business team, no DevOps)

  • Sign up free at app.superkind.ai with up to EUR 150 in credits and connect Outlook, Teams or Slack.
  • Send the first task - a real one from this week, for example an invoice to book, a quote to draft or a dunning letter to prepare.
  • Approve or correct the result; the correction becomes the rule, and the task runs as a routine next time.

Start with Azure AI Foundry (cloud team, custom build)

  • Set up Azure - a subscription, a Foundry resource and a project at ai.azure.com; a new account comes with USD 200 of credit for 30 days18.
  • Build a prompt agent - pick a model from the catalogue, write the instructions and attach tools from the Toolbox, no application code needed1.
  • Deploy a hosted agent - when the use case needs your own code or framework (Microsoft Agent Framework, LangGraph, Semantic Kernel), package it as a container and let Foundry run it with observability1.
  • Pin the region - if EU residency is a hard requirement, use Data Zone deployments in Sweden Central or Germany West Central7.

From the first task to your own app

With Superkind, the data stays in Europe from the first task to your own app. Four stages, on the accounting example:

  1. 1

    Ready in minutes

    Sign up free, forward the first invoice in Outlook or Teams and watch Superkind book it in DATEV.

  2. 2

    Tasks become routines

    What you explain once keeps running as a routine, for example every supplier invoice booked within the day. Before anything is posted, Superkind asks for approval.

  3. 3

    Learns from day one

    Every task makes Superkind smarter. Your accounts, cost centres, suppliers and approval limits end up in your Company Brain, and that knowledge is never lost.

  4. 4

    Your own app, without starting over

    When it gets more complex, you build custom solutions, apps and integrations on top of what Superkind already knows, for example your own month-end app with 1,500+ integrations: your Super App. Built yourself or with us. It all belongs to you.

Conclusion

Azure AI Foundry, now called Microsoft Foundry, is Microsoft’s build-your-own platform for AI agents on Azure: 10,000+ models, a managed agent service, built-in tools and the usual Azure governance, free to use and explore and billed by consumption after that12. It is a good fit when you have an Azure cloud team and a custom use case. For every other case - the whole team handing over routine work in Outlook, Teams or Slack across DATEV, SAP, HubSpot and 1,500+ integrations, hosted in Europe, with approval before anything goes out - a managed AI employee like Superkind gets you there without a build, from EUR 50 per month for unlimited users, as of October 2026. Pick the path that fits what you have on the team this week, and start free either way.

Frequently Asked Questions

Azure AI Foundry, now called Microsoft Foundry since January 2026, is Microsoft’s platform on Azure for building, hosting and running AI agents and AI apps, with a model catalogue of more than 10,000 models from Microsoft, OpenAI, Anthropic, Meta and others, a managed agent service, built-in tools and the usual Azure security and governance features. Think of it as the toolbox a cloud team uses to build their own AI agent on Azure.

No. Microsoft Copilot is a finished product you buy per user and use in Microsoft 365. Azure AI Foundry is a platform your developers use to build and run their own agents and apps on Azure, billed by consumption. Copilot Studio sits in between and lets admins extend Copilot with low-code agents on top of the Foundry stack.

Yes, at least some Azure experience. The platform assumes you can set up an Azure subscription and a Foundry resource, pick models from the catalogue, build a prompt agent in the portal or write code in Python, C#, JavaScript or Java, deploy a hosted agent as a container and run observability. If no one on your team can do that, the practical path is a managed AI employee like Superkind, where business teams write the task in Outlook, Teams or Slack and no DevOps is involved.

The platform itself is free to use and explore. You pay for what you consume: tokens for each model from the Foundry Models catalogue, per-session or per-transaction charges for Foundry tools (file-search vector storage per GB per day with the first 1 GB free, code interpreter per session, web search per 1,000 transactions), separate charges for Grounding with Bing Search, plus Azure services like Azure AI Search or Azure Machine Learning that you build on top. A new Azure account comes with a $200 credit for 30 days. As of October 2026, per Microsoft.

Yes, with care. Microsoft runs an EU Data Boundary, and Azure OpenAI and Foundry offer Data Zone deployments pinned to EU regions such as Sweden Central and Germany West Central, which keep storage and processing inside the geography, per Microsoft forum guidance. Global Standard deployments can route prompts outside the chosen region. You still sign Microsoft’s DPA, pick a lawful basis under Art. 6 GDPR, keep a record of processing (Art. 30), run a DPIA where the risk is high (Art. 35) and train anyone building AI steps under Art. 4 EU AI Act.

Different jobs. Azure AI Foundry is a build-your-own platform for cloud and development teams. Superkind is the ready-made AI employee for business teams: you write to it in Outlook or any email, Microsoft Teams, Slack or the web app, and it does the task in your programs through 1,500+ integrations such as DATEV, SAP Business One, HubSpot and Personio, hosted and processed in Europe, GDPR compliant, with approval before anything goes out. Starting from EUR 50 per month for the whole company with unlimited users, as of October 2026.

Yes. Many teams start a prompt agent in the Foundry portal to see what a single task looks like, then hand the day-to-day work to Superkind so operations, finance or sales do not depend on the cloud team for every change. Others start with Superkind in Outlook or Teams, prove the task, and build a custom app or integration on the Superkind platform when it gets more complex. The data stays in Europe either way.

Henri Jung, Co-founder at Superkind
Henri Jung

Co-founder of Superkind, the AI employee teams write to in Outlook, Teams and Slack. Henri helps companies, from small teams to the Mittelstand, hand their routine work to AI - without new tools, consultants or an in-house AI team.

Send Superkind your first task.

Start free in Outlook, Teams or Slack, hosted in Europe. Or see in 30 minutes how it runs in your systems.