Builders constructing agent fleets hold hitting the identical sample: the agent logic is prepared, however the data infrastructure beneath is advanced to do nicely. Attending to manufacturing means fixing for stability, scale, information entry, reply high quality, safety, and content material ingestion abruptly. At this time, we’re enabling builders to have quicker influence by simplifying the enterprise data platform.
Your organization’s IQ, powered by Microsoft IQ, is the collective intelligence locked in paperwork, emails, conferences, operational information, and the stay net. That is the place your true aggressive edge lives. Foundry IQ grounds brokers with the data from these sources and repeatedly improves based mostly on your small business objectives.
The bulletins in the present day are designed to assist prospects provision data bases quicker, unify enterprise and exterior sources, and expose that data by the Foundry IQ Mannequin Context Protocol (MCP) server for any agent framework or MCP-compatible hosts.
What’s new
Foundry IQ Serverless in preview: Provision on the spot, no-friction context retrieval with scale to zero pricing. Developer tier now out there in public preview with extra coming quickly. Docs | Create a Foundry IQ useful resource
New data sources in preview: Floor brokers throughout Work IQ, Cloth IQ (together with Information brokers and Ontology), File Search, Azure SQL, and MCP by a multi-source data base, with no customized integrations required. Docs | Cookbook
Internet IQ in Foundry IQ is now out there: Lengthen agent context to the net, honoring writer preferences, and market information with sub-165 ms latency and nil information retention. Weblog | Web site
Foundry IQ data bases are typically out there: Ship manufacturing brokers on a totally SLA-backed data layer with steady APIs, compliance certifications, and the Foundry IQ MCP server for any MCP-compatible host. Docs | Quickstart
Agentic retrieval high quality enhancements: The newest updates to the agentic retrieval engine enhance reply efficiency throughout datasets, effort tiers, and mannequin sizes whereas spending fewer tokens. Weblog | Quickstart
Information pipeline updates in preview: Automated layout-aware ingestion of paperwork, picture enrichment, and broader SharePoint indexing floor brokers in full paperwork, not simply uncooked textual content. Weblog | Quickstart
Safety updates in preview: New controls for encryption, permissions sync, and sensitivity-label governance hold enterprise coverage intact as content material flows into brokers. Weblog | Quickstart
Foundry IQ Serverless in public preview
We all know agent workloads are bursty and event-driven: an agent would possibly execute a whole lot of steps in seconds, then go idle for hours. Serverless eliminates infrastructure friction: no clusters to handle, no capability to order, no idle prices. Go from zero to manufacturing quick, with on the spot retrieval-augmented era (RAG) and state-of-the-art retrieval high quality in-built.
Foundry IQ Serverless (Developer tier) is obtainable in public preview. You might be billed for compute assets and storage used, and the service scales to zero when idle.
Serverless tiers use Compute Items (CU) to measure useful resource consumption, together with CPU utilization, reminiscence and storage I/O. Utilization is calculated every minute in increments of 0.25 CUs.
For big-scale serverless deployments, contact us for added choices.
Functionality
Developer tier
Compute utilization
$0.24 CU / hour
Listed storage
As much as $0.29 GB / month; GB value is area dependent
Listed storage per index
1 GB / index
Indexes per service
30 indexes / service
Companies per subscription per area
5 providers / subscription / area
Billing is predicted to start in late 2026 with particulars offered a minimum of 30 days prematurely. Clients utilizing Serverless Developer received’t be charged earlier than billing is enabled. Present Compute Unit measurements are estimates solely and topic to vary earlier than billing is enabled.
Subsequent steps: create a Foundry IQ Serverless useful resource within the Foundry portal.
New data sources in preview
How do I give an agent entry to organizational data and structured enterprise information with out constructing a customized connector for each system?
Bringing enterprise data into agent workflows usually means stitching collectively customized integrations throughout every information supply. Builders should account for various information codecs, permission fashions, retrieval patterns, and source-specific logic earlier than an agent may reliably use that data.
Foundry IQ simplifies this by bringing enterprise content material and structured techniques right into a single data base for multi-source, agentic retrieval. Builders may give brokers entry to that data with out constructing and sustaining separate connectors or source-specific retrieval methods.
New data sources in preview:
Work IQ brings organizational alerts like emails, conferences, information, and Groups messages into one enriched, AI-ready supply, all whereas respecting person permissions. Brokers can reply questions on how the group operates, what selections have been made, and what’s high of thoughts for groups.
Cloth IQ lets brokers question information brokers and firm ontologies: formal fashions of enterprise entities, relationships, and guidelines linked to stay information in OneLake and a specialised semantic layer. This returns structured solutions alongside unstructured doc context for a question.
File Search permits you to straight add information to a data base.
Azure SQL brings structured relational information right into a data base.
MCP Server connects data served over the Mannequin Context Protocol.
Subsequent steps: use the Foundry IQ Forgebook to check out further data sources.
Microsoft Internet IQ in Foundry IQ now out there
When a solution wants contemporary, real-world context, how do I attain the open net with out paying a latency or compliance penalty?
Microsoft Internet IQ is obtainable in restricted entry by the Foundry IQ MCP data supply. It offers brokers entry to exterior retrieval throughout net, information, pictures, video, and procuring sources whereas honoring writer preferences. It’s designed for giant language mannequin (LLM) workflows relatively than conventional search pages, with industry-leading low-latency rating.
Mixed with Foundry IQ, brokers can plan, search, purpose, and synthesize solutions that draw on each inside data and real-world exterior context in a single retrieval engine.
Subsequent steps: learn the weblog announcement for Microsoft Internet IQ.
Data bases in Foundry IQ are typically out there
What does it truly take to maneuver a prototype into manufacturing?
Manufacturing means ensures: steady contracts, predictable efficiency, and safety that holds underneath audit. Foundry IQ data bases and choose data sources, and safety capabilities are typically out there: with full SLA protection, compliance certifications, steady APIs, and enterprise-grade community isolation with identification and coverage enforced by default.
What’s included in GA:
Data bases: agentic retrieval references, output and exercise logs, the Foundry IQ MCP server, and minimal retrieval reasoning effort.
Foundry IQ MCP server: exposes Foundry IQ data bases as a distant MCP server, making them accessible from any MCP-compatible host or consumer, together with Claude, ChatGPT, LangChain, and the Microsoft Agent Framework. Community isolation, document-level safety, cross-source rating, and agentic retrieval all work over the open MCP customary, making it out there for the broader agent ecosystem.
Data sources: Azure Blob Storage (with a standing API to examine indexing progress), search indexes, Internet, and OneLake.
Safety: community isolation and managed identification assist.
“We’ve been utilizing Foundry IQ in our analysis and prototyping work, and the reusable data base strategy has minimize numerous the setup overhead we’d usually anticipate. With the ability to floor brokers in trusted enterprise content material from day one, with out rebuilding retrieval logic every time, has made early-stage experimentation noticeably quicker and better high quality.” — Jane Chen, Lead AI Developer, Baringa Companions
Subsequent steps: use the Mastering Foundry IQ cookbook to get began constructing with the Foundry IQ MCP server.
Agentic retrieval high quality enhancements
The newest retrieval enhancements improved our reply high quality benchmarks by as much as 20%, throughout our evaluated datasets, effort tiers, and mannequin sizes. In comparison with single-shot RAG, data bases improved recall by as much as 54%.
Foundry IQ improved its iterative agentic retrieval loop to batch queries extra successfully, floor extra related passages through semantic ranker, and apply server-side token caching to scale back redundant consumption throughout multi-turn conversations. This leads to meaningfully fewer tokens spent with out sacrificing reply high quality, whereas beating earlier benchmarks on reply high quality.
Subsequent steps: learn our weblog for extra on the most recent evaluations and Foundry IQ benchmarks.
Safety updates in preview
How do I hold enterprise information permissions intact as content material flows into brokers?
Safety belongs on the information layer, not approximated in utility code. A number of safety capabilities at the moment are in preview, together with cross-tenant customer-managed keys (CMK) utilizing federated identification credentials — eliminating shared secrets and techniques — Purview sensitivity-label auditing, incremental SharePoint permissions sync, APIM assist for Foundry mannequin integrations, and surfacing Purview sensitivity labels inside data sources so label-based entry controls are honored finish to finish.
Non-public connectivity between Foundry IQ and Foundry merchandise, through Shared Non-public Hyperlink and Community Safety Perimeter, is usually out there.
“By integrating Foundry IQ, we offer a managed, permission-aware enterprise context layer that connects advertising and marketing and model data into each agent to allow them to entry the correct data, on the proper time, with the correct governance.” — Andrei Pop, Director of PM, Innovation, Sitecore
Subsequent steps: learn extra concerning the newest Foundry IQ safety bulletins.
Information pipeline updates in preview
How do I make sure that brokers are grounded in the entire doc (tables, diagrams, and pictures) not simply the uncooked textual content?
Ingestion high quality units the ceiling on retrieval high quality. New information pipeline capabilities in preview embrace first-class SharePoint indexing for ASPX pages and Lists alongside doc libraries and doc enrichment to course of pictures plus serve them at question time in data bases, so brokers and customers can reference authentic visuals and ask follow-up questions on them. We’re additionally introducing Azure Content material Understanding chunking with picture verbalization — a layout-aware ingestion pipeline that converts diagrams, charts, and scanned pictures into significant textual content so brokers are grounded in full, semantically correct representations of supply paperwork.
Subsequent steps: learn Foundry IQ’s information pipeline deep dive weblog publish.
Get began in the present day
Construct as soon as, reuse in every single place: Foundry IQ allows you to floor a number of brokers with the identical data base, connecting and unifying information from wherever. Foundry IQ is designed for agent workloads to ship higher outcomes out of your firm’s IQ. With Foundry IQ, speed up agent supply, ship context with out blind spots, and guarantee each reply respects your group’s safety by default.
The best solution to discover Foundry IQ is thru the Microsoft Foundry portal. From there you’ll be able to create a data base, entry the documentation, and observe the Microsoft Foundry Be taught programs, all in a number of clicks.
Make sure to try the most recent information from Foundry IQ at Microsoft Construct 2026:
BRK246: Foundry IQ: Gasoline brokers with enterprise data and agentic retrieval
LAB532: From information to context: agent-ready data with Foundry IQ
BRK240: Construct context-aware brokers: From information to selections with Microsoft IQ
DEM331: Flip APIs, instruments, and information into actual agent velocity
