Foundry Tools

At a glance

What is it?

Foundry Tools, formerly Azure AI Services and before that Cognitive Services, is a set of pretrained AI services that developers call through REST APIs and client SDKs. Learn lists Speech, Translator, Language, Content Understanding, Document Intelligence, Vision, Azure AI Search, Content Safety, Custom Vision and Immersive Reader. Microsoft's architecture guide describes each tool as targeting one capability, so an application can add a feature such as OCR or transcription without building a model, and notes that many tools share the same Azure resource type as Microsoft Foundry while remaining distinct products. Several older services are retired or scheduled for retirement and are not for new applications.¹²³

What does it do?

The tools convert speech to text and text to speech, translate text across more than 100 languages and dialects, recognize entities and detect personal data in text, extract text, key-value pairs and tables from documents, turn documents, images and video into structured data, analyze images (including OCR and face detection), search company content and flag harmful content. Some tools train custom models on your own labeled data, many run in containers on premises, and many offer a free F0 tier for trials.¹³

Who needs it?

Application developers who need a specific, well-defined AI capability inside an app or workflow and do not need to choose models, orchestrate agents or train from scratch. Microsoft's architecture guide recommends considering Foundry Tools before building a custom model, and also names them as common building blocks for retrieval and agent solutions built in Microsoft Foundry.²

What does it need?¹

• An Azure subscription and a Foundry Tools or Foundry resource
• Microsoft Entra credentials or a resource key
• A region where the specific API is offered
• Labeled training data if you plan to build a custom model

What it is not

Foundry Tools is not the place for generative AI orchestration or agents; Microsoft's architecture guide says to work in Microsoft Foundry for models, agents and orchestration. It is not a training platform beyond each tool's customization options; for custom architectures Microsoft points to Azure Machine Learning. The rename changed the brand, not the services. Retired tools such as Language Understanding (LUIS), QnA Maker, Personalizer, Anomaly Detector, Metrics Advisor and Content Moderator are not options for new work.¹²

Availability and licensing. Generally available service family (Learn, checked 26 Oct 2026). Anomaly Detector, Content Moderator, Language Understanding (LUIS), Metrics Advisor, Personalizer and QnA Maker are marked retired and remain only for existing applications. Pricing is per transaction by tier, with a free F0 tier on many tools; new Azure accounts can try Azure free for up to 30 days. No prices are quoted here.³

The problem it solves

How it works

Capabilities

• Speech¹⁴

  Speech to text, text to speech, speech translation and speaker recognition; Microsoft also delivers its MAI-Voice and MAI-Transcribe models through Azure Speech.

  Why it matters: Adds transcription, captions and voice output to apps and voice agents.

  Limits: Model and voice availability varies by region; custom voice training and hosting are billed separately.

• Translator¹

  Machine translation covering more than 100 languages and dialects, including at-risk and endangered ones.

  Why it matters: Supports multilingual apps and content without building translation models.

  Limits: Quality varies by language pair; review output for regulated content.

• Document Intelligence and Content Understanding³⁵

  Extract text, key-value pairs and tables from documents, and convert unstructured documents, images and video into structured data.

  Why it matters: Automates data entry from forms, invoices and other files.

  Limits: Content Understanding is not available in the new Foundry portal at GA and runs from the classic portal.

• Language²

  Named entity recognition, personal data detection and language detection, with custom text classification options.

  Why it matters: Structures text for search, routing and compliance checks.

  Limits: Generative tasks such as open-ended summarization fit Foundry Models better.

• Vision and Custom Vision¹³

  Image and video analysis, OCR and face detection, plus custom image recognition trained on your images.

  Why it matters: Adds visual inspection and image understanding to apps.

  Limits: Custom Vision models need labeled images for training.

• Content Safety¹²

  Detects unwanted or harmful content in user and AI generated text and images.

  Why it matters: A safety layer for apps and agents that handle user content.

  Limits: Microsoft's architecture guide now lists it as Content Safety in Foundry Control Plane, so documentation may use either name.

• On-premises containers¹

  Many tools can run in Docker containers outside Azure datacenters.

  Why it matters: Keeps processing close to data for compliance, security or operational reasons.

  Limits: Not every tool is offered as a container; check the container support list.

Practical use cases

Who uses it

• Direct pojišťovna (part of the Direct group)

  In production

  Direct pojišťovna: AI agents for windshield claims

  15 minutes to about 2 minutes (87% less) (measured)

  Financial services, Czechia

• Air India

  In production

  Air India: AI.g customer service agent on Azure OpenAI

  about 40,000 (2026), up from an average of 30,000 (2024) (measured)

  Travel and hospitality, India

• EY

  Scaling

  EY: Dynamics 365 Sales with agents, and Copilot across the firm

  85,000 in 18 months; goal 100,000 (measured)

  Professional services, Global

Works with

Relationship labels follow Microsoft's documentation. "Alternative approaches" does not mean one is better: each profile says when it fits.

Getting started

1. Pick the right tool

   Use Microsoft's architecture guide to match the task to a tool and confirm no custom model is needed.

   Check: One tool covers the task without custom training.

2. Create a resource on the free tier

   Create a Foundry Tools resource and choose the F0 SKU where the tool offers it.

   Check: You have an endpoint and credentials for the resource.

3. Run a quickstart

   Call the tool with the quickstart in your language using a sample file.

   Check: The API returns the expected JSON result.

4. Harden for production

   Switch to Microsoft Entra authentication and virtual network rules, and pick a paid tier that fits your transaction volume.

   Check: Requests succeed with Entra credentials from your production network.

5. Avoid retired services

   Check that the design does not use Anomaly Detector, Content Moderator, LUIS, Metrics Advisor, Personalizer or QnA Maker.

   Check: The architecture uses only tools listed as available.

Official resources

• Product pageazure.microsoft.com (opens in a new tab)
• Documentationlearn.microsoft.com (opens in a new tab)

Sources

Each statement above links to the source it comes from. Labels say who reported it.

1. What are Foundry Tools? (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
2. AI and machine learning products (Azure Architecture Center) (opens in a new tab) Microsoft · Vendor-reported
3. Foundry Tools product page (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
4. Foundry Models sold by Azure - Microsoft Foundry (opens in a new tab) Microsoft · Vendor-reported
5. Microsoft Foundry portal general availability overview (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
6. What is Microsoft Foundry Agent Service? (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026