Azure AI Search (Foundry IQ)
Azure AI Search is Microsoft’s fully managed, cloud-hosted search service for connecting enterprise data to AI. Microsoft’s current docs title it “Foundry IQ (Azure AI Search)”, so expect both names. Agents and LLMs use it to get grounded answers from enterprise and web content.
Two retrieval modes (Microsoft Learn, updated 2026-11-17)
| Classic search | Agentic retrieval | |
|---|---|---|
| Target | One predefined index | A knowledge base over one or more knowledge sources |
| Query plan | None: a single request and response | LLM-assisted or user-provided: plan, split into subqueries, retrieve in parallel, semantically rerank, merge |
| Response | Ranked documents | An LLM-written answer or raw source data, with an activity log and references |
| Status | Generally available | Generally available, some capabilities in preview; region restrictions apply |
Capabilities
- Full-text, vector, hybrid and multimodal queries over indexed or remote content.
- AI enrichment during indexing: chunking, vectorising (integrated vectorisation) and LLM-assisted transformations. Only JSON documents can be indexed, loaded by push or pulled by indexers.
- Sources include Azure Blob Storage, Cosmos DB, SharePoint and OneLake.
- Relevance tuning, facets, filters (including geo-spatial), synonyms and autocomplete.
- Security: Microsoft Entra ID, Private Link, role-based access and document-level access control (native permission ingestion is generally available for supported Blob, ADLS Gen2 and indexed SharePoint scenarios; some others are preview).
- SDKs for .NET, Java, JavaScript and Python, plus REST and the Azure portal.
- Billing offers a Dedicated or a Serverless pricing model; see Microsoft’s pricing page for rates.
Microsoft’s advice: LLMs and agents do not require vectors, so add them only for similarity search or content that can be homogenised into vectors.
Where it fits
It is the managed Azure option beside self-run opensearch and Elasticsearch, and above library-level faiss. See vector-databases-compared for the dedicated databases. See retrieval-augmented-generation-overview, hybrid-search-and-rank-fusion and microsoft-foundry-note.