Foundry IQ
Foundry IQ is the knowledge layer behind Foundry agents: SLA-backed retrieval that unifies enterprise data sources behind one endpoint, built on Azure AI Search. It matters because it replaces the RAG plumbing most projects otherwise build by hand, and agents in Foundry Agent Service consume it directly.
What Foundry IQ Replaces
Without Foundry IQ, building RAG requires: chunking, embedding, indexing, retrieval pipeline, source integration, permission sync - all custom code.
With Foundry IQ: connect data sources, create a knowledge base, wire it to your agent. Foundry handles ingestion, indexing, retrieval, and permissions.
Architecture
IQ Components
| Component | Status | Description |
|---|---|---|
| Foundry IQ Knowledge Bases | GA | SLA-backed retrieval over connected data sources |
| Foundry IQ Serverless | Preview | No infrastructure provisioning required |
| Work IQ | Preview | Grounding in Microsoft 365 data (email, docs, calendar) |
| Fabric IQ | Preview | Grounding in Fabric data (semantic models, ontology) |
| Web IQ | Limited access | Live web grounding, sub-200ms, zero data retention |
| Agentic Retrieval | GA | Multi-step retrieval with reasoning |
| File Search | GA | Direct file upload and search |
Knowledge Base Setup
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential
client = AIProjectClient(
credential=DefaultAzureCredential(),
endpoint="https://my-foundry.services.ai.azure.com"
)
# Create knowledge base connected to Azure AI Search
kb = client.knowledge_bases.create(
name="product-docs",
data_sources=[
{"type": "azure_blob", "connection": "my-blob-connection"},
{"type": "sharepoint", "connection": "my-sp-connection"}
]
)
# Wire to agent
agent = client.agents.create(
model="gpt-5",
name="docs-agent",
tools=[{"type": "foundry_iq", "knowledge_base_id": kb.id}]
)
MCP Access
Knowledge bases are queryable via MCP (Model Context Protocol). Any MCP-compatible client can connect:
Endpoint: https://my-foundry.services.ai.azure.com/mcp
Data Pipeline Features
- Layout-aware ingestion - understands document structure (headers, tables, lists)
- Image serving - extracts and serves embedded images
- Broad SharePoint indexing - libraries, lists, pages
- Permission sync - respects source ACLs via Purview
- Sensitivity labels - governance flows through from source to retrieval
Foundry IQ vs Custom RAG
| Dimension | Foundry IQ | Custom RAG |
|---|---|---|
| Setup time | Minutes | Days-weeks |
| Infrastructure | Managed | Self-managed |
| Permission sync | Automatic (Purview) | Manual |
| Multi-source | Built-in | Custom per source |
| SLA | Yes | Depends on implementation |
| Flexibility | Opinionated | Full control |
| Best for | Standard enterprise knowledge | Custom retrieval logic, non-Azure sources |
Key Takeaways
- Foundry IQ turns days or weeks of custom RAG plumbing into a managed knowledge base you connect and wire to an agent.
- Permission sync through Purview means retrieval respects source ACLs without custom code.
- Knowledge bases are queryable over MCP, so clients outside Foundry can use them too.
- Build custom RAG only when you need retrieval logic Foundry IQ does not support, or sources outside Azure. Otherwise the managed option wins on setup time and SLA.
- Serverless, Work IQ, and Fabric IQ are in preview and Web IQ is limited access. Knowledge Bases, Agentic Retrieval, and File Search are GA.
Next: Foundry Tools - IQ gives agents knowledge; Foundry Tools give them pre-built capabilities over documents, speech, and images.