Microsoft expands Foundry with Fireworks

Microsoft used Build 2026 to add more of the production stack around Microsoft Foundry, turning the platform into a tighter package for model hosting, retrieval and agent operations. In posts published June 2, the company said Foundry now includes Managed Compute, broader Foundry IQ capabilities and expanded support for Fireworks AI on Foundry. Microsoft framed the changes as part of a push to give developers one place to run models, connect tools and ground agents in enterprise and web data. The announcements came through Microsoft’s Foundry blog, Azure product pages and Build session materials. ### So what actually changed in Foundry this week? Microsoft said in its June 2 “Build Edition” recap that Foundry’s model and compute options expanded with Fireworks AI on Foundry and Managed Compute, alongside new tuning and evaluation features. The same post said Foundry IQ broadened into a larger “knowledge plane,” adding serverless retrieval, new knowledge sources, Web IQ and security updates. (devblogs.microsoft.com) Microsoft’s Build recap also placed those additions next to hosted runtimes, toolboxes, memory and governance tools such as ASSERT and the Agent Control Specification, or ACS. That packaging matters because Microsoft is no longer presenting model access, retrieval and policy controls as separate products; it is describing them as parts of one development loop inside Foundry. That is Microsoft’s characterization in the June 2 post. (devblogs.microsoft.com) ### Where does Fireworks fit into this? Microsoft announced Fireworks AI on Microsoft Foundry in public preview on March 11. In that launch post, Yina Arenas, a corporate vice president for Microsoft Foundry, said the integration was meant to bring “high-performance open model inference into Azure” and give customers “a single, trusted control plane” to evaluate, deploy, customize and operate open models. (devblogs.microsoft.com) Fireworks, in Microsoft’s description, is the inference layer for open models, while Foundry is the operational layer around it. Microsoft said customers using Fireworks through Foundry can avoid stitching together separate tools, contracts and deployment paths for open-model workloads. ### What is Foundry IQ supposed to do? (azure.microsoft.com) Pablo Castro, Microsoft CVP and distinguished engineer, wrote on June 2 that Foundry IQ is aimed at the “knowledge infrastructure” underneath agent deployments. His post said the new releases let customers provision knowledge bases faster, unify enterprise and external sources, and expose that knowledge through a Foundry IQ MCP server for agent frameworks and MCP-compatible hosts. (azure.microsoft.com) Microsoft said Foundry IQ knowledge bases are generally available, while Foundry IQ Serverless is in preview. The company also said new preview data sources include Work IQ, Fabric IQ, Azure SQL, File Search and MCP, while Web IQ is now available to extend agent context to the web with “zero data retention.” ### Why does the packaging matter for enterprise deployments? (devblogs.microsoft.com) Microsoft’s Azure page for Foundry IQ says the service includes “enterprise-grade security, identity and policy,” document-level access control synchronization and query-time trimming aligned to user permissions. Those details show Microsoft expects customers to use the product as a shared retrieval and grounding layer, not just as a developer convenience feature. (devblogs.microsoft.com) Microsoft’s Build recap makes a similar point on the tooling side. The company said Toolboxes in Foundry offer “one governed endpoint” for tools, skills, MCP clients and enterprise data, while ASSERT, ACS and Rubric are meant to bring policy-driven evaluations and runtime controls closer to deployment. ### What should readers watch next? Microsoft said on June 2 that hosted agents in Foundry Agent Service are expected to reach general availability by early July 2026. (azure.microsoft.com) The company’s June 2 Foundry and Foundry IQ posts, along with the Azure product page for Foundry IQ, are the current source documents for availability, preview status and the list of supported knowledge sources. (devblogs.microsoft.com)