Hammad's Foundry Updates Portal

Your personal guide to the latest and greatest in Microsoft Foundry and the Microsoft Agent Framework — the unified enterprise AI platform for building, deploying, and operating models, agents, and AI apps at scale. Auto-refreshed weekly.

🏭 Agent Factory for the enterprise ☁️ Azure-native · Cross-cloud 🔐 Entra ID · RBAC · Governance built-in 🧪 Evals & Red Teaming as first-class ops
11,000+
Models in catalog
1,400+
Tool connectors
1 SDK
Agent Framework (Py / .NET)
70k+
Organizations building on Foundry
🗞️ Auto-updated weekly

What's new in Foundry & Agent Framework

This feed refreshes itself: a GitHub Actions workflow scans the Microsoft Foundry blog and Agent Framework releases every week, summarizes what changed, and commits the results here.

Loading latest updates…

The Big Picture

One platform, five pillars

Microsoft Foundry is an agent-first AI factory that spans cloud, edge, and Microsoft 365 — a shift from a model-centric toolbox to a complete platform for agentic apps. Everything below rolls up into these pillars.

01

Foundry Models

Unified catalog of frontier & open models — OpenAI, Anthropic, Grok, Mistral, Llama, Phi, Hugging Face — with routing, fine-tuning & provisioned throughput.

02

Foundry Agent Service

Managed runtime to build, host, orchestrate, and scale single- and multi-agent systems with identity, memory, and tools.

03

Foundry IQ & Tools

Unified knowledge/RAG layer plus a catalog of 1,400+ connectors and MCP tools that ground agents in enterprise data and actions.

04

Foundry Control Plane

Day 2 home base: observability, evaluations, red teaming, guardrails, policy, cost and fleet governance across every agent & model.

05

Foundry Local

On-device inference runtime for Windows, macOS & edge — run models and agents offline with the same tooling.

💡 Why it matters: Instead of stitching together a model host, an orchestration library, a vector DB, an eval harness, and a monitoring stack from five vendors, Foundry gives you the full lifecycle — build → deploy → operate → govern — under one roof, with Azure identity, networking, compliance, and billing already wired in.
Pillar 01 · Models

Foundry Models — every frontier model, one endpoint

The model catalog is the front door: sold-direct-by-Azure frontier models plus thousands of partner and open-weight models, all deployable behind a consistent API with unified auth, quota, and billing.

🗂️

Unified Catalog

OpenAI (GPT-5 family, o-series, Sora), Anthropic Claude, xAI Grok, Mistral, Meta Llama, Cohere, DeepSeek, Microsoft Phi, plus 10k+ Hugging Face models — searchable, benchmarked, and leaderboard-ranked in the portal.

🔀

Model Router GA

Automatically routes each request to the best-fit model for quality vs. cost — no code changes. Swap or add underlying models without redeploying your app.

🎛️

Fine-Tuning & RFT

Supervised fine-tuning, distillation, and reinforcement fine-tuning on frontier models; bring datasets, run training jobs, and deploy tuned variants under the same governance.

📦

Flexible Deployment

Global Standard (serverless, pay-per-token), Data Zone & regional deployments for residency, and Provisioned Throughput (PTU) for guaranteed latency at scale.

🧩

Bring Your Own Model

Register custom or open-weight models behind the Foundry gateway so home-grown models get the same routing, monitoring, and policy enforcement as catalog models.

📊

Benchmarks & Leaderboards

Compare models on quality, cost, and throughput with built-in benchmarking before you commit — then A/B in production with evals.

Spotlight

Anthropic Claude in Foundry GA

Frontier Claude models are first-party citizens in Foundry — making Azure the only cloud where you can mix OpenAI, Anthropic, and open models behind one governed endpoint. Claude also powers agent scenarios across Microsoft 365 Copilot and Copilot Studio.

ModelBest forContextNotes
Claude Opus 4.xDeepest reasoning, complex agentic coding, long-horizon tasks~200K tokensFlagship tier; highest capability & cost
Claude Sonnet 4.5 / 4.6The workhorse — agents, coding, computer use, office tasks~200K tokensBest capability-to-cost balance; most popular for agents
Claude Haiku 4.5High-volume, low-latency: chat, classification, sub-agents~200K tokensMost cost-efficient; great for fan-out multi-agent work
🏗️

How it works: two hosting paths

  • Hosted on Azure: inference runs entirely on Azure infrastructure (e.g., East US 2, Sweden Central) with Entra ID auth, RBAC, data-zone residency options, and Azure SLAs — recommended for production.
  • Hosted on Anthropic: Foundry endpoint proxies to Anthropic's infrastructure — earliest access to brand-new models/features before Azure-hosted GA.
  • Same Foundry APIs & SDKs (Python, C#, TypeScript, Java) either way — including the native Anthropic Messages API shape.
💳

Enterprise mechanics

  • Billing via Azure Marketplace using Claude Consumption Units (CCUs) mapped to token pricing — shows up on your Azure bill.
  • Spend can count toward your MACC (Microsoft Azure Consumption Commitment).
  • Prompt caching & tool streaming supported for cost/latency wins in agentic loops.
  • Zero Data Retention option for regulated, high-sensitivity workloads.
  • Works with Model Router, evals, red teaming, tracing, and content safety like any other Foundry model.
🤝 Why this matters: Multi-model is the default strategy now. Route coding & computer-use agents to Claude Sonnet, deep research to Opus or GPT-5, and high-volume triage to Haiku or mini models — all in one platform, one bill, one compliance boundary. No separate Anthropic account required.
Pillar 02 · Agents

Foundry Agent Service — agents as managed cloud resources

The Agent Service turns "an agent" from a script on someone's laptop into a durable, secured, observable cloud resource: model + instructions + tools + knowledge + memory + identity, exposed as an API.

⚙️

Declarative Prompt Agents

Define agents with instructions, model, tools & knowledge in the portal, SDK, or a versioned agent.yaml — deploy in minutes without managing infra.

🧠

Agent Memory Preview

Persistent, context-aware memory across sessions and long-running workflows — agents remember users, preferences, and prior task state.

🛠️

Tools & MCP

Built-in tools (Bing grounding, code interpreter, file search, browser/computer use) plus Model Context Protocol servers and 1,400+ Logic Apps connectors — Stripe, Atlassian, Databricks, ServiceNow, and more.

🪪

Entra Agent ID

Every agent gets a first-class Microsoft Entra identity — auditable, permissioned, and governable like any employee or service principal. No more shared API keys.

🕸️

Multi-Agent Workflows GA

Compose agents into orchestrated workflows — sequential, concurrent, handoff — with durable state, retries, and human-in-the-loop approval steps.

🌐

Cross-Cloud & A2A

Agents connect to data and services in Azure, AWS, and GCP, and interoperate with external agents via the open Agent-to-Agent (A2A) protocol.

Spotlight

Hosted Agents — bring your code-first agents into Foundry

Hosted agents let you take an agent you wrote in any framework — Microsoft Agent Framework, LangGraph, CrewAI, OpenAI Agents SDK — package it as a container, and run it inside Foundry as a managed, serverless resource.

💻

1 · Build locally

Write your agent in Python/.NET with the framework of your choice.

→
📦

2 · Containerize

Docker build → push to Azure Container Registry.

→
🚀

3 · Create hosted agent

Register the image as an agent version in your Foundry project.

→
📈

4 · Operate

Invoke via the standard Responses API; scale, trace, evaluate & govern in the Control Plane.

💎

The value of bringing agents into Foundry

  • Zero infra ops: no AKS clusters, ingress, or autoscaling to manage — Foundry runs the containers serverlessly.
  • Instant enterprise-grade: Entra identity, VNet/private networking, RBAC and audit inherited automatically.
  • Unified observability: your custom agent shows up in the same traces, dashboards, and eval pipelines as declarative agents.
  • One invocation surface: every agent — prompt-based or code-first — is callable through the same OpenAI-compatible Responses API.
  • Fleet governance: security teams see and control all agents in one registry instead of shadow deployments.
🧾

Anatomy of a hosted agent

# agent.yaml — versioned agent definition
name: contoso-support-agent
kind: hosted
image: contosoacr.azurecr.io/support-agent:v3
protocol: responses      # OpenAI-compatible
cpu: 1
memory: 2Gi
env:
  - name: MODEL_DEPLOYMENT
    value: "claude-sonnet-4-5"

Versioned, reviewable, CI/CD-friendly — agents become artifacts you promote through environments like any other software.

Open Source SDK

Microsoft Agent Framework GA

The open-source SDK + runtime for building agents and multi-agent workflows in Python and .NET. It's the convergence of Semantic Kernel (enterprise durability) and AutoGen (cutting-edge orchestration) into one production-ready framework — both predecessors are now in maintenance mode.

🤖

Agents

LLM-driven, tool-using, autonomous. Ideal for open-ended tasks where the model decides the path.

  • Any Foundry model (OpenAI, Claude, local) as the brain
  • Function tools, MCP servers, OpenAPI, A2A messaging
  • Pluggable memory & threads (Azure SQL, Redis, vector stores)
  • Middleware for auth, filtering & telemetry
🧭

Workflows

Explicit, typed, graph-based orchestration for structured business processes where you decide the path.

  • Patterns: sequential · concurrent · handoff · magentic (dynamic manager)
  • Checkpointing & durable long-running state
  • Human-in-the-loop pauses & approvals
  • Type-safe routing with conditional edges

Get running in three lines

# pip install agent-framework
from agent_framework.azure import AzureAIAgentClient

agent = AzureAIAgentClient().create_agent(
    instructions="You are a helpful support agent.",
    tools=[lookup_order, issue_refund],
)
result = await agent.run("Where is order #4821?")

Enterprise-ready by default

  • OpenTelemetry tracing baked in — every step observable in Foundry & Azure Monitor
  • Entra ID auth; no keys in code
  • Open standards: MCP · A2A · OpenAPI for cross-vendor interop
  • Deploy anywhere: local dev → containers → Foundry hosted agents with no rewrite
  • .NET: dotnet add package Microsoft.Agents.AI
🔁 Local-to-cloud continuum: the same MAF agent you debug on your laptop deploys unchanged as a Foundry hosted agent — that pairing (open SDK + managed runtime) is the core developer story of Microsoft's agentic stack.
Pillar 03 · Knowledge

Foundry IQ — the shared knowledge layer

Foundry IQ (powered by Azure AI Search) turns retrieval into a reusable platform capability: knowledge bases that any agent can ground on, instead of bespoke RAG pipelines per app.

📚

Knowledge Bases

Point at SharePoint, Fabric/OneLake, Azure Blob, websites, and databases; IQ handles ingestion, chunking, indexing, and freshness automatically.

🎯

Agentic Retrieval

Query planning, iterative multi-source search, and answer synthesis with citations — delivering up to ~40% better relevance than naive vector RAG.

🔒

Document-Level Security

Results are trimmed to the caller's permissions — the agent only sees what the user is entitled to see. Compliance travels with the data.

Pillar 04 · Operate & Govern

Day 2 Operations — where trust is earned

Shipping an agent is Day 1. Foundry's Control Plane makes Day 2 — evaluating, securing, watching, and governing AI in production — a continuous, automated discipline rather than an afterthought.

🧪

Evaluations GA

  • Built-in evaluators: groundedness, relevance, coherence, fluency, safety, plus agentic evaluators — intent resolution, tool-call accuracy, task adherence
  • Custom & LLM-as-judge evaluators for domain rules
  • Run in dev, in CI/CD gates (GitHub Actions / Azure DevOps), and continuously on live traffic
  • Batch evals over datasets curated straight from production traces
🛡️

AI Red Teaming Agent

  • Automated adversarial testing built on Microsoft's open-source PyRIT framework
  • Simulates prompt injection, jailbreaks, data exfiltration & harmful-content attacks across risk categories (violence, hate, self-harm, sexual)
  • Reports Attack Success Rate (ASR) with per-attack logs for remediation
  • Run pre-deployment or on a schedule against live agents for continuous regression checks
🔭

Observability & Tracing

  • End-to-end OpenTelemetry traces of every agent step: model calls, tool invocations, retrieval, handoffs
  • Unified dashboards in the Control Plane's Operate view — quality, cost, latency, token burn per agent
  • Deep Azure Monitor / Application Insights integration and alerting on drift or quality regressions
  • Full audit trails for incident investigation and compliance
🚧

Guardrails & Content Safety

  • Azure AI Content Safety filters on inputs & outputs: hate, violence, sexual, self-harm
  • Prompt Shields against injection & jailbreak attempts; groundedness detection against hallucination
  • Protected-material detection and PII redaction
  • Policy-enforced defaults across the fleet — not per-app opt-ins
🏛️

Governance & Fleet Management

  • Foundry Control Plane: one registry of every agent, model & tool with ownership and lifecycle state
  • Entra Agent ID for identity & least-privilege permissions; conditional access for agents
  • Integration with Microsoft Purview (data compliance) & Defender (threat protection)
  • Maps to NIST AI RMF (Map · Measure · Manage) and supports EU AI Act readiness
💰

Cost & Quota Control

  • Token quotas and rate limits per project, per deployment, per agent
  • Spend dashboards spanning serverless, PTU, and Marketplace (e.g., Claude CCU) billing
  • Model Router + prompt caching + batch API as built-in cost levers

The GenAIOps loop

🧱

Build

Pick models, define agents, ground with IQ

→
🧪

Evaluate

Quality + safety evals, red team scans

→
🚀

Deploy

CI/CD gates → hosted agents / endpoints

→
🔭

Observe

Traces, dashboards, alerts, cost

→
🔁

Improve

Trace-driven datasets, prompt optimization, re-eval

Pillar 05 · Edge & Ecosystem

Foundry Local & the broader ecosystem

💻

Foundry Local

Run optimized open models (Phi, Mistral, Qwen, DeepSeek…) on-device on Windows & macOS via ONNX Runtime — OpenAI-compatible local API, zero cloud cost, full privacy, offline-capable. Same catalog & tooling mindset, at the edge.

🧑‍💼

Microsoft 365 & Copilot Studio

Publish Foundry agents into Teams, Microsoft 365 Copilot, and BizChat; Copilot Studio makers and Foundry developers share models (including Claude), connectors, and governance.

👩‍💻

Developer surface

Foundry portal, Foundry CLI, VS Code extension, GitHub integration, and REST/SDKs (Python, C#, TS, Java). Templates and quickstarts take you from idea → deployed agent in minutes.

Cheat sheet — what to reach for, when

You want to…Use
Call a frontier model (GPT-5, Claude, Grok…) with enterprise authFoundry Models — serverless deployment or PTU
Auto-pick the cheapest model that's good enoughModel Router
Stand up a tool-using agent without writing infraFoundry Agent Service (prompt agents)
Run your LangGraph / CrewAI / MAF agent as a managed serviceHosted agents (container → Foundry)
Write agent code with orchestration, memory & HITLMicrosoft Agent Framework (Python / .NET)
Ground agents on enterprise data with permission trimmingFoundry IQ knowledge bases
Prove quality before & after shippingEvaluations (+ CI/CD gates)
Attack your own app before adversaries doAI Red Teaming Agent (PyRIT)
Watch, alert, and audit production agentsControl Plane observability + Azure Monitor
Block harmful content & prompt injectionContent Safety guardrails + Prompt Shields
Give every agent an identity & least privilegeEntra Agent ID
Run models offline / on-deviceFoundry Local
Live from Sweden Central

See it running — live in Foundry

This section is wired to a real Foundry project — resource swedenfoundry93, project foundry-showcase — with three sample agents deployed on models from this subscription. Paste a short-lived access token below to list the agents and chat with them directly from this page.

🏗️

The live project

  • Resource: swedenfoundry93 (AIServices, S0)
  • Project: foundry-showcase
  • Region: Sweden Central 🇸🇪
  • Model deployments: gpt-5.4 · gpt-5.4-mini · gpt-5.4-nano · Phi-4-reasoning · gpt-image-2 · text-embedding-3-large

Open project in the Foundry portal ↗

🤖

Sample agents deployed New Agent Service

foundry-concierge — explains Foundry capabilitiesgpt-5.4-mini · v1
deep-research-analyst — structured analysis & trade-offsgpt-5.4 · v1
triage-bot — high-volume request classificationgpt-5.4-nano · v1

Built as versioned prompt agents in the new Foundry Agent Service and invoked through the OpenAI-compatible Responses API — three archetypes on three model tiers, exactly the multi-model routing story Foundry is built for.

🎮 Interactive console

Calls go from your browser straight to swedenfoundry93.services.ai.azure.com over HTTPS using the new Agent Service Responses API (POST /openai/v1/responses with an agent_reference), with multi-turn continuity via previous_response_id. Tokens expire in ~1 hour and never leave this page.