11,000+ Model Catalog
Deploy leading frontier models—including OpenAI (GPT-4o, o1), Anthropic Claude, Meta Llama, and Microsoft Phi—via serverless API endpoints or managed GPU-isolated compute.
Azure AI Foundry Scale and Govern Your Autonomous AI Agent Fleet
Microsoft Foundry (formerly Azure AI Foundry) unifies over 11,000 foundational models and stateful agent runtimes under a single platform. With Pargesoft's deep expertise in Dynamics 365 and Microsoft security architecture, you can securely open your D365 data to AI without altering existing security or authorization structures. Kickstart your intelligent transformation by ensuring your agents generate insights while strictly adhering to corporate privacy labels and access permissions, keeping your data completely secure.
Plan Your Secure AI Transformation
Azure AI Foundry
What is Microsoft Foundry?
Microsoft Foundry (formerly Azure AI Studio and Azure AI Foundry) is Microsoft's unified enterprise AI platform designed to build, ground, and govern custom AI applications and autonomous agents at scale. Built directly under the Microsoft.CognitiveServices resource provider, it eliminates infrastructural fragmentation by grouping model gateways, agent hosting, vector indexing, and compliance tracing under a single governance plane. As Pargesoft, we leverage our quarter-century of enterprise ERP and integration experience to adapt this platform seamlessly and securely into your existing business processes, maximizing the value of your AI investments.
Foundry Agent Service
Host and scale code-first autonomous agents. Execute stateful, multi-turn loops using the next-generation, server-side Responses API (Agents v2).
Foundry IQ Layer
Connect agents directly to SharePoint, Microsoft Fabric OneLake, and Azure Blob Storage. Automate chunking and indexing for highly accurate RAG workflows.
Centralized Fleet Control
Audit, trace, and protect your entire multi-cloud agent fleet. Manage security postures with Microsoft Entra Agent ID, Purview DLP, and Defender for Cloud.
Enterprise Architecture Hierarchical Resource Architecture:
Hierarchical Resource Architecture:
Resource, Project, and Assets
Scaling enterprise AI requires balancing centralized security controls with developer agility. Microsoft Foundry solves this by structuring workloads into a strict three-tier resource hierarchy.
Foundry Resource Subscription Boundary
Managed as Microsoft.CognitiveServices/accounts, this defines the administrative boundary. Global configurations like BYO-VNet, Customer-Managed Keys (CMK) for encryption, and shared regional model quotas are configured exclusively at this level.
Foundry Projects Development Scopes
Created as sub-resources, projects isolate teams (e.g., Finance, HR). Developers can prototype and deploy agents without repeating security audits or VNet configurations, securely reusing parent model deployments.
Project Assets Scoped Workload Assets
Resources that live strictly inside a single project, including prompt templates, fine-tuning datasets, agent versions, and OpenTelemetry logs. This separation prevents data mixing and simplifies GDPR/HIPAA compliance.
Security-Driven Separation of Concerns in B2B AI
To prevent developers from accidentally exposing endpoints to the public internet or misconfiguring GPU SKUs, Microsoft Foundry separates the Control Plane (management) from the Data Plane (execution).
Control Plane IT & Security Administrators
Administrators configure Azure Policy, private network routing, and global safety filters at the top-level resource. Developers are blocked from altering outbound firewalls or disabling responsible AI guardrails, protecting the enterprise boundary.
Data Plane AI Engineering Teams
Developers are assigned the Foundry User role. This provides data-plane access to run model inference, manage project-level files, build agents, and trace performance within designated projects without administrative permissions.
Agent Architecture
Selecting the Right Agent Architecture: Prompt Agents vs. Hosted Agents
Microsoft Foundry supports diverse agentic runtime patterns based on your system design constraints, development workflow, and infrastructure requirements. By collaborating with Pargesoft consultants, you can determine the most suitable architecture for your enterprise, ensuring cost efficiency and performance optimization from day one.
Prompt Agents
Configuration-First & Serverless
- Infrastructural Principle: Completely serverless, zero-maintenance runtime fully managed and scaled by Microsoft.
- Execution Pattern: The agent's instructions, model mappings, and built-in tools are defined declaratively via portal UI or YAML.
- Best-Fit Scenario: Quick deployment of document search assistants, standard HR/IT Q&A bots, and low-latency workflows.
Hosted Agents
Code-First & Custom Orchestration
- Infrastructural Principle: Fully managed serverless container workloads running on a customer-dedicated, secure container runtime.
- Execution Pattern: Code-first agents built using advanced frameworks (LangGraph, Semantic Kernel) are packaged into container images.
- Best-Fit Scenario: Long-horizon, multi-agent workflows with explicit graph cycles and complex handoff or escalation loops.
Enterprise Knowledge Layer
Automating Enterprise RAG with Foundry IQ
Traditional RAG architectures require establishing separate vector databases, building custom data chunking pipelines, and managing synchronization—demanding weeks of engineering effort. Built on Azure AI Search, Foundry IQ is a managed enterprise knowledge layer that eliminates this complexity.
Foundry IQ unifies SharePoint, Microsoft Fabric OneLake, Azure Blob Storage, and external web sources into a single Knowledge Base. It fully automates embedding generation and incremental indexing, exposing them through a single unified endpoint. Pargesoft's data engineering team ensures your scattered corporate data is seamlessly migrated into this unified knowledge layer, guaranteeing that your AI agents always operate with the most accurate and up-to-date information.
Workflow
The 4-Stage Agentic Retrieval Pipeline
Unlike legacy search patterns that rely on single-keyword queries, Foundry IQ utilizes an advanced Agentic Retrieval engine. Powered by a large language model (LLM), it optimizes context retrieval through a 4-stage dynamic pipeline.
Systems architects can adjust the retrieval reasoning effort (minimal, low, or medium) to balance latency budgets against reasoning quality.
-
Query Deconstruction
The LLM analyzes the user's complex intent and decomposes it into multiple targeted sub-queries.
-
Parallel Hybrid Execution
Sub-queries are executed simultaneously across indexed sources leveraging hybrid search (semantic vector + BM25).
-
Semantic Reranking
Retrieved document chunks are contextualized and prioritized using advanced semantic rerankers.
-
Consolidation & Citation
The engine filters redundant data and passes a clean, aggregated context string with precise inline citations.
Data Security
Preventing Data Leaks with Permission-Aware Grounding
When grounding AI agents in corporate data, there is a risk of data leakage. If an agent accesses a global index containing confidential salary records, it may disclose that information to unauthorized employees.
Foundry IQ mitigates this at the query-time layer via Permission-Aware Grounding:
- Delegated Authentication: Every search query executes within the caller's Microsoft Entra ID context.
- Dynamic ACL Sync: The engine synchronizes and validates SharePoint and Fabric permissions in real time.
- Query-Time Filtering: Unauthorized documents are filtered out before the retrieval pipeline outputs data.
Compliance Boundaries
Advanced Fleet Governance with Microsoft Purview DLP Integration
In global B2B environments, scaling an autonomous agent fleet introduces complex regulatory challenges. Organizations must prevent proprietary data leaks, enforce geo-specific compliance boundaries (such as GDPR, HIPAA, and ISO 27001), and maintain strict auditability over every large language model (LLM) interaction.
Microsoft Foundry integrates natively with the Microsoft Security Stack to establish a comprehensive compliance boundary around your generative AI workloads. By bridging model inference with enterprise-grade data governance, organizations can safely democratize AI tools across global departments without compromising corporate IP or violating regional data residency mandates. Pargesoft's security architects configure these complex DLP and Purview integrations from end-to-end, fully aligned with local regulations (such as GDPR), entirely eliminating enterprise risk.
Real-Time Data Loss Prevention (DLP)
Purview APIs intercept prompts and agent responses in real time, detecting and blocking sensitive data before it is sent to external model providers.
Sensitivity Label Protection
Foundry RAG agents honor Purview Sensitivity Labels. Users must have explicit EXTRACT and VIEW rights to receive answers grounded in encrypted documents.
Microsoft Defender for Cloud & SIEM
Defender provides real-time threat protection for Foundry Tools, alerting administrators to jailbreaks. All telemetry is centralized into Microsoft Sentinel.
Frequently Asked Questions (FAQ)
Yes. Standalone Azure OpenAI resources can be upgraded directly via the portal or Azure CLI. The upgrade preserves your endpoint URI, API keys, and model deployments while unlocking advanced project containers.
The next-gen Responses API stores and manages execution state server-side. It natively supports deep reasoning models, multi-agent delegation, and context compaction, preventing token redundancy.
Every agent deployed in Foundry is automatically issued a durable, unique Entra Agent ID at build time. This identity tracks ownership and usage history across the agent's lifecycle.
Every Foundry project automatically applies a unique project tag to its underlying resource usage. FinOps teams can filter costs in Azure Cost Management by these tags, enabling seamless chargeback.
Build the AI Infrastructure of the Future Today with Pargesoft
Combine the boundless potential of Microsoft Foundry with Pargesoft's quarter-century of enterprise ERP, cloud architecture, and integration experience. Let's securely connect your existing data to frontier models and rapidly deploy custom autonomous agents that are fully GDPR compliant. If you are ready to turn Generative AI from a trial project into a strategic asset with measurable ROI, our expert engineers are ready to listen.
Schedule a Free Discovery Meeting