Foundry Tools
Foundry Tools are pre-built AI APIs for documents, speech, vision, language, and translation, which an agent can call instead of you building those capabilities yourself. They were previously Azure AI Services, and are now part of the Foundry platform and available to agents as tools.
Service Catalogue
| Tool | Capabilities | Use For |
|---|
| Document Intelligence | OCR, layout detection, table extraction, key-value pairs, custom models | Invoices, contracts, forms, receipts |
| Speech | Speech-to-text, text-to-speech, real-time transcription, voice cloning (MAI-Voice-2) | Voice agents, transcription, accessibility |
| Vision | Image analysis, OCR, spatial analysis, face detection | Image classification, document scanning |
| Language | NER, summarisation, PII detection, sentiment, key phrase extraction | Text analytics, compliance, content moderation |
| Content Understanding | Multimodal analysis (video, audio, documents) | Media processing, content indexing |
| Translator | Real-time translation, 100+ languages, document translation | Multilingual apps, content localisation |
Document Intelligence
Extract structured data from unstructured documents at scale.
| Feature | Description |
|---|
| Prebuilt models | Invoice, receipt, ID, tax form, health insurance |
| Layout | Tables, paragraphs, headers, page structure |
| Custom models | Train on your document types |
| Add-on features | Barcode, formula, font extraction |
from azure.ai.documentintelligence import DocumentIntelligenceClient
from azure.identity import DefaultAzureCredential
client = DocumentIntelligenceClient(
endpoint="https://my-foundry.cognitiveservices.azure.com/",
credential=DefaultAzureCredential()
)
result = client.begin_analyze_document(
"prebuilt-invoice",
document_url="https://storage.blob.core.windows.net/invoices/sample.pdf"
).result()
for invoice in result.documents:
print(f"Vendor: {invoice.fields['VendorName'].content}")
print(f"Total: {invoice.fields['InvoiceTotal'].content}")
Document Intelligence docs
Speech
| Feature | Status | Description |
|---|
| Speech-to-text | GA | Real-time and batch transcription |
| Text-to-speech | GA | Neural voices, custom voice |
| MAI-Transcribe-2 | Preview | Microsoft's speech-to-text model with content biasing |
| MAI-Voice-2 | Preview | Multilingual TTS with voice cloning |
| Voice Live | GA | End-to-end voice agent (integrated with Agent Service) |
| Multichannel | GA (July 2026) | Stereo input audio processing |
Speech Service docs
Vision
| Feature | Description |
|---|
| Image analysis | Captions, tags, objects, people |
| OCR | Text extraction from images |
| Spatial analysis | People counting, zone monitoring |
| Face | Detection, verification, grouping |
Vision docs
Language
| Feature | Description |
|---|
| Named Entity Recognition | People, places, organisations, dates |
| PII Detection | Personal data identification and redaction |
| Summarisation | Extractive and abstractive |
| Sentiment Analysis | Document and sentence level |
| Key Phrase Extraction | Important terms and concepts |
| Text Analytics for Health | Medical entity extraction |
New in July 2026: Foundry playgrounds for language detection, text PII detection, conversation PII detection, and Text Analytics for Health.
Language Service docs
Content Understanding
Multimodal content analysis - process video, audio, and documents with a single API.
| Input | Output |
|---|
| Video | Scene detection, transcription, visual descriptions |
| Audio | Transcription, speaker identification, content classification |
| Documents | Layout, entities, classifications |
Content Understanding docs
Translator
| Feature | Description |
|---|
| Text translation | 100+ languages, real-time |
| Document translation | Preserve formatting |
| Custom translator | Domain-specific terminology |
| Transliteration | Script conversion |
Translator docs
Integration with Agents
All Foundry Tools are available as agent tools via Toolboxes:
agent = client.agents.create(
model="gpt-5",
name="document-processor",
tools=[
{"type": "document_intelligence"},
{"type": "language", "capabilities": ["pii_detection", "summarization"]}
]
)
Key Takeaways
- Check for a Foundry Tool before building a capability: document extraction, speech, vision, language, and translation are already solved.
- Document Intelligence covers invoices, contracts, forms, and receipts, with custom models for your own document types.
- Every tool is available to agents through Toolboxes, so extracting a document or redacting PII becomes a single tool call.
- MAI-Transcribe-2 and MAI-Voice-2 are in preview. Speech-to-text, text-to-speech, and Voice Live are GA.
Next: Microsoft Agent Framework - Foundry hosts agents; Agent Framework is how you write the ones that need custom logic.