Incident sheet · /enterprise-ai

Filed
Read
4 min · 872 words
Severity
Routine

Foundry IQ adds knowledge graphs, Retriever-1 and managed setup

Microsoft is adding knowledge graphs, Microsoft-Retriever-1 and automatic provisioning to Foundry IQ, aiming to ground enterprise agents while reducing retrieval work.

Microsoft adds knowledge graphs, retrieval model to Foundry IQ for better AI grounding
FeedMicrosoft adds knowledge graphs, retrieval model to Foundry IQ for better AI grounding — Pexels / pixabay

Event log

  1. Foundry IQ adds three capabilities: knowledge graphs, Microsoft-Retriever-1 and automatic provisioning.
  2. Knowledge graphs are currently in public preview.
  3. Microsoft-Retriever-1 runs on existing indexes without additional provisioning or document reindexing.
  4. Managed Foundry IQ provisioning will enter public preview in December 2026 under usage-based, serverless pricing.
  5. Microsoft has published no detailed Microsoft-Retriever-1 performance comparisons.

By Tom Whitfield

Three Foundry IQ changes target enterprise agents; managed-resource provisioning will enter public preview in December 2026. Knowledge graphs are already in public preview, Microsoft-Retriever-1 is a new retrieval model, and every new Foundry project will receive a managed Foundry IQ resource.

At the Microsoft Ignite conference, Microsoft presented three additions to Foundry IQ, its knowledge retrieval service, for agents that access, understand and reason over company data. Grounding those agents in accurate, relevant and current business information matters as enterprises automate more workflows.

Knowledge graphs add cited relationships

Foundry IQ's new graphs connect related entities and facts across documents and organize enterprise content into topics. They let agents answer questions requiring evidence from several sources instead of relying on a single matching document.

Developers can add a graph configuration containing their ontology to an existing Foundry IQ knowledge source or import an existing graph. Foundry IQ then maps how documents, contacts and information relate, while recording the source of every connection.

Ashish Chaturvedi, executive research leader at HFS Research, said: “Until now, retrieval could find documents that resembled a question, but it could not follow the connections between them.” His example: standard search may return a regulation and documents mentioning it, leaving developers to connect the rule, supplier and product chain.

Knowledge graphs can let an agent trace that chain and cite each link, reducing the need for developers to write retrieval logic for complex questions. Chaturvedi cited regulatory compliance, supply chain risk, contract analysis and M&A due diligence as likely use cases because each can depend on several sources.

What changes for enterprise teams?

Microsoft-Retriever-1 moves more search planning into a specialized model. It interprets queries, reformulates them with terminology from enterprise documents and decides whether further searches are needed, aiming to reduce search count, LLM reliance, latency and token consumption.

Developers can enable the model on existing Foundry IQ indexes and knowledge bases without extra provisioning or document reindexing. Stephanie Walter, practice lead of the AI stack at HyperFrame Research, said: “It handles decisions such as where to look, what to ask next, and when enough evidence has been gathered.”

Automatic provisioning will remove the separate Azure AI Search setup from the starting line. Each new Foundry project will receive a managed retrieval resource, so developers need not provision or manage Azure AI Search to create knowledge bases and connect agents to enterprise data.

The managed service will handle the underlying retrieval infrastructure, use usage-based, serverless pricing, and enter public preview in December 2026. Once available, Chaturvedi said, teams can prototype grounded agents immediately and pay only for what they use, rather than carry standing charges for unused projects.

Analysts expect the three capabilities to simplify AI application and agent development. The practical gain is less setup code and fewer infrastructure choices before an agent reaches useful work. The plumbing gets friendlier; the bill still needs a controls desk.

Scale, governance and usage economics

Beyond development, the three updates could improve reliability, economics and scalability, according to analysts. Knowledge graphs provide richer context across documents, supporting better-informed decisions and more reliable execution of complex workflows.

Using a specialized retrieval model instead of larger general-purpose models for every task could reduce application overhead and improve responsiveness, Walter said. That could make agent deployment more economical across a wider range of enterprise workflows.

Automatic provisioning also creates a more consistent, governed route for connecting agents to company data. Chaturvedi said common retrieval services reduce homegrown search setups that enterprises must secure and maintain, while usage pricing lowers the cost of experimentation.

Accuracy and cost remain unsettled

The primary concern with knowledge graphs is the accuracy of entities and relationships extracted across documents. Chaturvedi said automated extraction may misread industry jargon, abbreviations or inconsistent names, causing agents to draw incorrect conclusions. Developers must verify actual business relationships and update the graph as documents change.

Graph-based answers can take longer and cost more than conventional searches, making them better suited to complex questions than routine lookups. The economy depends on querying the right graph, not merely having one.

Microsoft-Retriever-1 can add latency and processing cost when it runs iterative searches. Chaturvedi said the benefit depends on the model recognizing when enough evidence has been gathered; Microsoft has published no detailed performance comparisons, so enterprises should test it against their own data. The model's retrieval decisions may also be harder to inspect than developer-written search logic.

For CIOs, the sharper concerns are Microsoft-Retriever-1 and automatic provisioning, particularly cost predictability and dependence. Automatic provisioning may deepen dependence on Microsoft's retrieval stack by making it the default starting point for new Foundry projects. The standard path simplifies governance, though Microsoft becomes the default.

Usage-based pricing can become less predictable when agents run continuously, Walter said. She recommended spending limits and eligibility criteria for the projects permitted to use the service.

Microsoft has addressed three pieces of agent development. It has not removed the need to test retrieval quality, graph accuracy and consumption before deployments expand.

For CIOs, agents need enterprise grounding. The harder choice is which Microsoft dependencies, and which bills, that grounding will permit.

via infoworld.com/feed/ (Source)

Filed under

  • microsoft-foundry-iq
  • enterprise-agents
  • knowledge-graphs
  • microsoft-retriever-1
  • azure-ai-search
Share this article:

More from Tom Whitfield

Tom Whitfield

Show full bio

Market editor covering people, launches and day-to-day practice in enterprise information technology and the leaders who run at CIO Monitor.

185 articles

The newsroom digest

One email a day with the stories that matter. No noise.

By subscribing you agree to the privacy policy.

Cross-references

‹ Previous articleNext article ›