NewsBite analysis · AI, Business, Cloud, Data, Security, ToolsAzure Blog

Foundry IQ: Build smarter agents faster with unified knowledge and serverless retrieval

Microsoft's Foundry IQ team announced a set of updates designed to simplify how enterprises ground AI agents in company knowledge and data. The centerpiece is Foundry IQ Serverless, a developer tier in public preview, which bills by Compute Units and scales to zero when idle. This aims to remove the infrastructure burdens of bursty agent workloads and accelerate time-to-production for retrieval-augmented generation (RAG) workflows. In parallel, Microsoft introduced the Foundry IQ MCP server, exposing knowledge bases as a remote MCP service that can be consumed by Claude, ChatGPT, LangChain, and other MCP-compatible hosts. The integration supports a unified knowledge platform that ingests content from Azure Blob Storage, OneLake, web indexes, SharePoint, and other sources, while preserving fine-grained permissions and governance using Purview labeling and customer-managed keys. The blog outlines several kits: Work IQ for internal signals (emails, meetings, files), Fabric IQ for business ontologies and data agents, and File Search for direct file uploads. Microsoft Web IQ provides external retrieval across web, news, images, video, and shopping contexts, designed for LLM workflows with low-latency ranking. Across the board, the emphasis is on production-readiness: stable contracts, predictable performance, enterprise-grade security, and a scalable architecture that reduces the need for source-specific connectors. Microsoft also highlighted benchmark gains in answer quality and recall, and noted that pricing for Foundry IQ Serverless Developer tier will begin in late 2026 with billing only after activation. The announcement positions Foundry IQ as a backbone for agent-driven automation that can align multiple data sources with organizational permissions to improve decision-making.