| name | foundry-iq-knowledge-layer |
| description | This skill should be used when designing Microsoft Foundry IQ knowledge bases, reusable enterprise grounding, federated knowledge APIs, permission-aware retrieval, knowledge source governance, and shared context layers for agents. |
Foundry IQ Knowledge Layer
Foundry IQ is the Microsoft-native answer for reusable, governed enterprise knowledge. Use it when multiple agents need a shared, permission-aware knowledge layer instead of each agent owning separate RAG plumbing, indexes, source connectors, and retrieval policies.
When to Activate
Activate this skill when:
- Designing Foundry IQ knowledge bases for agents or copilots.
- Reusing the same enterprise knowledge across multiple agents, apps, or workflows.
- Grounding agents over mixed sources such as SharePoint, OneLake, Azure data stores, web content, or customer-managed indexes.
- Preserving ACLs, sensitivity labels, tenant boundaries, and user permissions during grounding.
- Deciding whether to use Foundry IQ, Azure AI Search, SharePoint grounding, File Search, or Fabric Data Agent.
- Standardizing a knowledge API or knowledge-base catalog for an enterprise agent platform.
Do not activate this skill for adjacent work owned by other skills:
- Direct custom index design, chunking, scoring, and retrieval traces:
azure-agentic-retrieval.
- Structured business metrics and semantic models:
fabric-data-agent.
- Managed identities, OBO, RBAC, and tenant boundaries:
azure-identity-for-agents.
- Tool catalog and action governance:
foundry-tool-governance.
Core Concepts
Knowledge bases are a product boundary, not just a prompt trick. The goal is to make enterprise knowledge reusable, governed, and consistent across agents while keeping the user's effective permissions intact.
Use Foundry IQ as the default when:
| Need | Why Foundry IQ fits |
|---|
| Shared knowledge across agents | One knowledge base can serve many agents and applications |
| Permission-aware grounding | Enterprise ACLs, labels, and tenant controls remain central |
| Mixed source estates | Federated and indexed sources can be presented through one knowledge API |
| Reduced RAG duplication | Teams avoid rebuilding ingestion, vectorization, query planning, and governance per agent |
| Enterprise control plane | Knowledge ownership, refresh, and reuse become platform concerns |
Foundry IQ complements lower-level retrieval. Azure AI Search remains the right choice when the team needs direct index control, custom ranking, enrichment pipelines, or domain-specific retrieval tuning.
Practical Guidance
- Define knowledge bases by durable business domain, not by single agent.
- Treat each knowledge base as a governed product with owners, source lists, permission semantics, refresh expectations, and citation requirements.
- Put reusable enterprise grounding in Foundry IQ before building per-agent RAG pipelines.
- Use Azure AI Search behind or beside Foundry IQ when custom indexing and retrieval tuning are required.
- Route analytical questions to Fabric Data Agent or Fabric semantic models when the answer requires computation over governed measures.
- Keep identity decisions explicit: service identity for platform operations, OBO or user context for permissioned answers.
- Evaluate retrieval quality, groundedness, permission filtering, and citation accuracy before promoting a knowledge base to production agents.
Examples
Example: enterprise support platform
Knowledge base: "Product Support"
Sources: SharePoint troubleshooting guides, public docs, known-issue database, escalation playbooks
Users: support agents, customer-facing copilots, engineering triage agents
Use Foundry IQ as the shared knowledge layer.
Use Azure AI Search only for the known-issue database if it needs custom indexing or ranking.
Use Entra OBO when the answer depends on the user's support role or customer assignment.
Example: choosing the grounding path
One uploaded PDF for one agent: Foundry File Search.
Shared enterprise policy knowledge: Foundry IQ.
Custom document index with tuned scoring: Azure AI Search.
Power BI-aligned sales metrics: Fabric Data Agent.
Microsoft 365 documents with user permissions: SharePoint grounding or Foundry IQ, depending on availability and reuse needs.
Guidelines
- Prefer Foundry IQ for reusable enterprise knowledge bases.
- Keep knowledge-base ownership separate from agent ownership.
- Preserve permissions before optimizing relevance.
- Require citations and source identifiers for answers grounded in enterprise knowledge.
- Version and test knowledge-base changes before attaching them to production agents.
- Document when a knowledge base federates sources versus indexes copied content.
- Keep volatile product setup steps in deployment docs so the skill remains stable as Foundry IQ evolves.
Gotchas
- Reusable does not mean universal: Broad knowledge bases can leak intent and reduce relevance. Scope by business domain and permission boundary.
- Permissions are retrieval behavior: If the user cannot access the source, the knowledge layer must not retrieve it for that user.
- Foundry IQ does not remove evaluation: You still need groundedness, citation accuracy, freshness, and access-control tests.
- Product surfaces may move: Foundry IQ is an evolving product surface. Verify current APIs, connector coverage, and preview status before hard-coding architecture.
- Fabric IQ and Foundry IQ are different layers: Use Fabric for governed analytical semantics; use Foundry IQ for reusable enterprise knowledge grounding.
Integration
- azure-agentic-retrieval - Provides lower-level Azure AI Search and retrieval-design patterns when Foundry IQ needs custom indexes or tuning.
- fabric-data-agent - Owns governed analytical answers and semantic-model grounding.
- azure-identity-for-agents - Owns OBO, managed identity, RBAC, and tenant-boundary design.
- responsible-ai-guardrails - Evaluates groundedness, prompt injection risks, and policy adherence.
- core memory-systems - Owns the generic retrieval and memory architecture.
- core context-optimization - Owns context selection, compression, and token budget tradeoffs.
References
Skill Metadata
Created: 2026-05-24
Last Updated: 2026-05-24
Author: Agent Skills for Context Engineering Contributors
Version: azure-0.1.0
