Microsoft documentation added new Work IQ and MCP guidance for Microsoft 365 Copilot and Copilot Studio, alongside Azure OpenAI Responses API and Foundry updates.

This new documentation introduces the concept of 'skills' for the Azure OpenAI Responses API, detailing how users can package, upload, manage, and securely use reusable skill bundles to extend and control model behavior in shell environments. It covers skill creation, versioning, usage in both hosted and local shell modes, inline skills, and important security considerations.

• Introduces 'skills' as reusable, versioned bundles for the Responses API shell tool
• Explains skill packaging, manifest requirements, and compatibility with Agent Skills standard
• Details multiple methods for mounting skills: uploaded, local, and inline (base64-encoded zip)

This new documentation introduces the shell tool for the Azure OpenAI Responses API, enabling users to execute shell commands in either a hosted or local environment as part of API calls. It details setup, usage patterns, security considerations, and output handling, representing a major new capability for programmatic automation and scripting via Copilot-integrated models.

• Introduces the shell tool for executing shell commands via the Responses API
• Supports both hosted (sandboxed container) and local execution modes
• Explains container reuse for iterative workflows and artifact download

This new page documents model-specific licensing and usage terms for Grok, Llama, Black Forest Labs, and Mistral models available through Microsoft Foundry. It details legal, attribution, indemnification, and usage restrictions that customers must follow, which is critical for compliance and risk management.

• Introduces enforceable acceptable use and licensing terms for Grok, Llama, Black Forest Labs, and Mistral models in Microsoft Foundry
• Specifies unique restrictions for Llama models, including user limits, EU multimodal access bans, and mandatory attribution
• Details Black Forest Labs requirements for end user agreements, output responsibility, and termination conditions

This new documentation introduces a capability in Microsoft Foundry to convert agent production traces into curated, versioned evaluation datasets using intelligent sampling. It details both portal and SDK workflows, supported regions, prerequisites, and best practices, enabling users to efficiently generate high-quality evaluation or fine-tuning datasets from real user interactions.

• Describes how to generate evaluation datasets from production agent traces using intelligent sampling.
• Explains both portal-based and SDK-based workflows for dataset creation.
• Lists supported Azure regions for this feature.

This new documentation introduces the Trace Replay feature (in preview) for Microsoft Foundry, enabling users to inspect, analyze, and replay agent conversation traces. It details how to access, visualize, filter, and troubleshoot traces, providing significant new capabilities for debugging, optimization, and auditing of Copilot agent interactions.

• Introduction of Trace Replay feature for inspecting and replaying agent conversation traces
• Two visualization modes: Trajectories view (hierarchical, token/duration analysis) and User view (chat-based, end-user perspective)
• Interactive replay controls including playthrough, skipping, speed adjustment, and timeline scrubbing

This new documentation introduces the ability to log end user feedback (such as thumbs up/down or rating scales) from AI applications using OpenTelemetry semantics, enabling correlation of user sentiment with trace data for continuous improvement. It details supported evaluation types, event attributes, and how to view feedback in Application Insights.

• Introduces logging of end user feedback (binary and Likert scale) via OpenTelemetry events
• Describes how feedback events are correlated with traces and exported to Application Insights
• Provides rules and attributes for binary and ordinal (Likert) evaluations

This new documentation introduces the ability to generate synthetic evaluation datasets for Copilot agents using the Microsoft Foundry data generation service. It details how to create question-and-answer datasets from agent definitions, prompts, or reference files, enabling evaluation before production traffic is available.

• Describes synthetic dataset generation for Copilot agents using Foundry service
• Supports three input types: agent definition, inline prompt, and reference file
• Provides SDK and portal instructions for dataset creation

This new documentation introduces benchmark evaluation capabilities in Microsoft Foundry (preview), allowing users to evaluate models or agents against built-in benchmark datasets and evaluation logic. It details how to create, configure, and interpret benchmark evaluations, including judge model selection and troubleshooting.

• Describes how to run benchmark evaluations on models and agents using built-in datasets and evaluators
• Explains selection of targets, benchmarks, and judge models within the Foundry portal
• Provides API examples for programmatically creating evaluation groups and runs

This new documentation introduces and details the process for running agent evaluations using the Azure Developer CLI (azd) for Microsoft Foundry agents. It covers prerequisites, step-by-step workflows for both hosted and prompt-based agents, command usage, asset management, and integration with agent optimization, providing a comprehensive guide for users to measure and improve agent quality.

• Introduces azd CLI commands for initializing, running, and updating agent evaluations
• Explains workflows for both hosted and prompt-based agents
• Details management of evaluation assets (datasets, evaluators, eval.yaml)

This new page documents the Foundry tracing capability in Microsoft Copilot, detailing how it captures and manages customer data for observability, debugging, and optimization of AI agents. It explains how tracing is enabled/disabled, data storage, privacy considerations, and customer responsibilities, providing essential guidance for secure and compliant use.

• Introduces Foundry tracing as an observability feature for AI agents in Copilot
• Details what data is captured (inputs, outputs, metadata, etc.) and how it is stored
• Explains enablement and disablement procedures for tracing via Azure Monitor Application Insights

This new documentation introduces the ability to govern Microsoft Foundry model router deployments using Azure Policy. It details how administrators can enforce model selection policies at deployment time across all deployment surfaces, ensuring compliance and providing remediation guidance.

• Model router deployments now honor Azure Policy restrictions on approved models.
• Policy enforcement applies across portal, API, CLI, and ARM templates.
• Unapproved models are visible but cannot be selected; policy violation blocks deployment.

This new documentation introduces the public preview of managed compute deployment for open-source models in Microsoft Foundry. It details how users can select, deploy, scale, monitor, and manage open-source models (such as those from Hugging Face) on dedicated GPU infrastructure managed by Microsoft, including quota management and troubleshooting.

• Describes end-to-end workflow for deploying open-source models on managed compute in Foundry
• Explains selection of models and deployment templates from the Hugging Face Collection
• Covers deployment via portal and Python SDK, including inferencing with OpenAI SDK

This new documentation introduces a guided guardrail setup feature for agents in Microsoft Foundry, allowing users to configure safety and security controls through an interactive questionnaire. The process tailors guardrails based on agent use case, data handling, and tool integrations, enhancing both security and usability.

• Guided guardrail setup experience for configuring agent safety in Microsoft Foundry
• Questionnaire-based approach maps responses to targeted risk controls
• Controls are applied at specific intervention points: user input, tool calls, tool responses, and output

This new documentation introduces 'Managed compute' in Microsoft Foundry (Preview), a deployment type that allows users to host open-source models on dedicated GPU capacity without managing infrastructure. It details how managed compute works, its integration with Foundry, supported runtimes, scaling, billing, access control, and current limitations.

• Announces public preview of managed compute for open-source models on dedicated GPUs in Foundry
• Describes deployment without VM or Kubernetes management—Microsoft handles runtime, security, and patching
• Explains model instance, deployment template, accelerator family, and runtime concepts

This new documentation introduces 'instant models' in Microsoft Foundry (preview), allowing users to access supported AI models immediately by name without deployment. It details prerequisites, usage, supported models, quota management, enterprise controls, and current limitations, representing a significant new capability for rapid prototyping and experimentation with Copilot-related models.

• Instant models can be used immediately by specifying the model name—no deployment required
• Works with existing APIs, SDKs, and dev tools; no code or configuration changes needed except for the model parameter
• Instant models are currently available only in West US 3 region during preview

This new documentation introduces 'rubric evaluators' (in preview), a feature that lets users define custom, weighted evaluation criteria for agent or model responses using LLMs as judges. It details how to generate, configure, and use rubric evaluators for domain-specific quality assessment, including model recommendations and integration with continuous monitoring.

• Introduces rubric evaluators for custom, weighted LLM-based scoring of agent/model responses
• Explains auto-generation and manual creation of rubrics with specific fields and weights
• Provides guidance on selecting LLM judge models and warns against using less reliable models

This new page introduces and documents how to build ephemeral agents using the Foundry Responses API, enabling developers to define agents in application code and access Foundry models and tools without persisting agent resources. It provides step-by-step guidance, code samples for Python and .NET, and details on advanced capabilities like tool integration, streaming responses, and enterprise observability.

• Explains the ephemeral agent pattern using the Foundry Responses API, allowing agent definitions to live in application code
• Describes differences and advantages over hosted agents and prompt agents
• Details project endpoint capabilities: project-scoped data, Foundry models, platform tools, OBO authentication, and observability

This new page introduces the Agent Optimizer feature (in limited preview), providing step-by-step instructions for optimizing hosted AI agents using Azure CLI tools. It covers installation, project setup, optimization workflow, deployment of improved agents, and monitoring, enabling users to automate and enhance agent performance.

• Introduces Agent Optimizer for hosted agents (limited preview, sign-up required)
• Details CLI extension installation and project initialization
• Describes optimization workflow: dataset generation, running optimizer, deploying best candidate

This new documentation introduces the 'routines' feature (in public preview) for Microsoft Foundry, enabling users to automate agent invocations on a schedule or at a specific time. It details supported trigger and action types, prerequisites, creation and management steps, and known limitations, providing comprehensive guidance for automating Copilot agents.

• Introduces routines as named automation rules to trigger agents on schedules or timers
• Details two trigger types: recurring schedule (cron) and one-shot timer
• Supports two action types: invoking agents via Responses API or Invocations API

This new documentation introduces the ability to add managed MCP servers powered by connector namespaces in Microsoft Foundry, enabling agents to use pre-built connectors to integrate with over 1,000 SaaS, data, and line-of-business systems. It details the configuration process, authentication types, publisher tiers, data handling, and step-by-step instructions for connecting and using these tools in agents. This significantly expands Copilot's extensibility and integration capabilities.

• Explains how to add managed MCP servers via connector namespaces in Foundry
• Details support for OAuth2 and API key (CustomKeys) authentication types
• Describes publisher tiers and data handling responsibilities for connectors

This new documentation introduces the ability to register and observe external agents (running outside Foundry) in Microsoft Foundry using OpenTelemetry and Application Insights. It enables trace-based evaluation and observability for agents not hosted within Foundry, expanding monitoring and evaluation capabilities to agents on any cloud or on-premises environment.

• Explains how to instrument external agents to emit OpenTelemetry spans to Application Insights for Foundry integration
• Describes the process to register external agents in Foundry and link their telemetry for trace view and evaluation
• Details requirements for preview feature usage, including SDK setup and required headers/flags

This new documentation introduces the Agent Optimizer, a preview feature for optimizing Copilot agents in Azure AI Foundry. It details how to automatically improve agent instructions, skills, tool definitions, and model selection using CLI commands and configuration files, enabling users to enhance agent performance and efficiency.

• Describes four optimization targets: instruction tuning, skill improvement, tool optimization, and model selection
• Explains how to configure and run optimizations using the Azure CLI and YAML files
• Provides guidance on interpreting optimization results and deploying improved agent configurations

This new page introduces and documents how to make a hosted agent compatible with the Agent Optimizer (currently in limited preview). It provides step-by-step instructions for integrating the optimization package, configuring agent files, and applying optimized configurations, enabling users to improve agent prompts, tools, and skills automatically.

• Explains installation and setup of the azure-ai-agentserver-optimization package
• Details required configuration directory and file formats for baseline and optimized agent configs
• Describes how the optimizer can improve prompts, tool descriptions, and discover new skills

This new page introduces detailed documentation for creating evaluation datasets for the Agent Optimizer (in limited preview), including both automated and manual dataset creation, dataset formats, and best practices. It enables users to systematically evaluate and optimize Copilot agents using custom or generated datasets and evaluators.

• Explains how to generate evaluation datasets and evaluators using the azd CLI
• Provides guidance for manually creating custom JSONL datasets with specific evaluation criteria
• Details dataset structure, field requirements, and examples for both customer support and coding assistant agents

This new documentation introduces the ability to build real-time voice agents using hosted agents in Foundry Agent Service via the invocations_ws WebSocket protocol, currently in public preview. It details how to implement, deploy, and connect voice agents, including protocol specifics, client integration, observability, and telephony bridging.

• Introduces invocations_ws WebSocket protocol for real-time, bidirectional audio streaming with hosted agents
• Describes endpoint structure, authentication, frame semantics, and connection lifecycle for voice workloads
• Provides implementation guidance for both direct audio streaming and WebRTC signaling scenarios

This new documentation introduces 'Routines' in Foundry Agent Service (preview), enabling users to automate agent execution based on time-based triggers directly within Foundry projects. This feature reduces the need for external scheduling or orchestration tools for simple, recurring, or one-time agent automation.

• Introduces routines as a project-native way to trigger agents on schedules or timers
• Describes supported trigger types: one-time timer and recurring (cron-style) schedules
• Explains that routines manage trigger, action, permissions, connections, and run history within Foundry

This new page introduces the Agent Optimizer for Foundry Agent Service, a tool that automates the evaluation and improvement of AI agent configurations, including prompt instructions, skills, tool descriptions, and model selection. It details how the optimizer works, its optimization targets, configuration, supported models, result interpretation, and current preview limitations. This is a major new capability for users building and refining AI agents in Azure.

• Describes the Agent Optimizer's closed-loop process for improving agent configurations automatically
• Explains optimization targets: instruction tuning, skill improvement, tool optimization, and model selection
• Details configuration requirements, including supported models and how to set up eval and optimization models

The documentation now supports importing LoRA adapters and draft models for speculative decoding (both in preview) in addition to full-weight custom models. The supported model architectures and versions have been updated and expanded, and the import and deployment workflows have been revised to accommodate these new model types and preview features.

• Added support for importing LoRA adapters (preview) and draft models for speculative decoding (preview)
• Expanded and updated the list of supported model architectures and versions
• Clarified required files for each model type (full-weight, LoRA adapter, draft model)

The page has been updated for May 2026 with a large number of new and updated articles, especially around Azure OpenAI, Foundry Agent Service, evaluations, observability, and agent optimization. The previous summary of agent governance and observability features has been removed, but the new and updated articles indicate significant expansion of capabilities and documentation.

• Major expansion of new articles covering Azure OpenAI, agent routines, voice agents, optimizer, and evaluation tools
• Significant updates to existing articles, especially around provisioned throughput, model routing, and agent deployment
• Removal of the previous summary section on agent governance and observability, replaced by more granular documentation

The documentation introduces several new Fireworks models available in the Microsoft Foundry model catalog, including models from Google, Meta, Mistral AI, and updated Qwen and DeepSeek versions. The preview disclaimer was removed, indicating a possible shift from public preview to general availability. These changes expand the range of models users can deploy and may signal increased stability and support.

• Added new models: DeepSeek V4 Pro, Google Gemma 4 series, Meta Llama 3.1, Mistral AI Ministral 3, and additional Qwen variants
• Removed previous preview disclaimer from the introduction
• Updated the list of supported catalog models, removing some older models and adding new ones

The documentation now introduces new MAI image models (MAI-Image-2.5 and MAI-Image-2.5-Flash) with both text-to-image and image-to-image editing capabilities, expands deployment region support, and updates API quotas to include these new models. It also adds instructions for using the image edits API, reflecting significant new functionality for users.

• Added MAI-Image-2.5 and MAI-Image-2.5-Flash models with image-to-image editing
• Expanded deployment regions to include UAE North
• Updated API quotas and limits to cover new models and free tier

The page has been extensively expanded with detailed, actionable guidance for rolling out Microsoft Foundry, including new sections on isolation boundaries, capability requirements, resource topology, security, automation, governance, and sample deployment patterns. This provides organizations with a comprehensive framework for planning, deploying, and governing Foundry at scale, replacing a previously high-level checklist.

• Added detailed guidance on defining isolation boundaries and workload segmentation
• Introduced a capability matrix mapping Foundry features to access and isolation requirements
• Outlined resource topology choices for co-located vs. isolated workloads

New evaluator types have been introduced, including Customer Satisfaction, Quality Grader, and Rubric evaluators, along with a new section explaining evaluation levels (turn vs. conversation). These additions expand the ways Copilot agent and conversation quality can be measured and allow for more flexible, holistic, and customizable evaluation strategies.

• Added Customer Satisfaction (preview) evaluator for holistic user satisfaction measurement
• Added Quality Grader (preview) evaluator for multi-dimensional quality assessment in a single run
• Introduced Rubric evaluator (preview) for custom, weighted LLM-based scoring

The documentation now describes a versioned Skills API, introducing skill versioning, new endpoints, stricter naming rules, and the ability to attach skills to toolboxes. This enables more robust skill lifecycle management and integration options for Copilot agents in Foundry.

• Skills are now versioned; each update creates an immutable version and default version management is explicit.
• New API endpoints and workflows for creating, listing, and managing skill versions.
• Skill names must follow a strict pattern and are used as URL path keys.

The documentation introduces significant new capabilities for memory management in Microsoft Foundry Agent Service, including item-level CRUD operations, store-level retention controls, and a new procedural memory type. These enhancements provide greater flexibility, control, and personalization for Copilot agents.

• Added support for item-level create, read, update, list, and delete (CRUD) operations on memory records
• Introduced store-level default retention controls, including default TTL for memory stores
• Added procedural memory as a new memory type, alongside user profile and chat summary memory

The 'What's new' section now lists several major new Copilot-related capabilities in Microsoft Foundry, including model routing, agent-to-agent endpoints, instant model access, voice agents, routines, and integrations with Microsoft Fabric and Microsoft 365. These additions significantly expand Foundry's agent development and integration features.

• Added model routing with the Responses API for flexible model selection
• Introduced A2A agent endpoints for agent-to-agent communication
• Enabled instant model access without prior deployment

Page removed or relocated. The documentation page for the Agent Optimizer feature in Azure AI Foundry has been removed. This page detailed how to optimize agent instructions, skills, tools, and model selection using the optimizer, which is currently in limited preview. The removal means users lose official guidance on configuring, running, and interpreting results for agent optimization.

• Loss of documentation for optimizing agent instructions, skills, tools, and model selection in Azure AI Foundry
• No longer available: step-by-step CLI usage, configuration examples, and troubleshooting for the Agent Optimizer preview
• Missing guidance on interpreting optimization results and deploying optimized agent configurations

Page removed or relocated. The documentation page detailing how to make a hosted agent compatible with Agent Optimizer (preview) has been removed. This eliminates official guidance on integrating, configuring, and applying optimization workflows for Copilot agents using the Agent Optimizer preview feature.

• Loss of step-by-step instructions for installing and configuring the Agent Optimizer package
• Removal of detailed examples for setting up baseline and candidate configurations
• No longer documents how to use optimization outputs (instructions, tools, skills) in agent code

Page removed or relocated. The documentation page for the Agent Optimizer feature in Foundry Agent Service has been removed. This page described a preview capability for automatically improving hosted AI agents via configuration optimization, including instruction tuning, skill improvement, tool optimization, and model selection. The removal means detailed guidance on using and understanding this feature is no longer available, which is significant for users interested in agent optimization workflows.

• Loss of documentation for the Agent Optimizer preview feature in Foundry Agent Service
• Missing guidance on automated agent evaluation and configuration improvement
• No longer available: instructions for optimization targets, model selection, and interpreting results

This new documentation introduces a method to disable or re-enable the creation and updating of classic agents and assistants in Azure OpenAI accounts using a specific account tag. It details the operational impact and provides CLI instructions, which is important for administrators managing access to these features.

• Introduces the 'MS-AOAI-Feature-Assistants' tag to control creation/updating of classic agents and assistants
• Provides CLI commands for disabling and re-enabling the feature at the account level
• Lists specific Assistants API operations that are blocked when disabled

The deployment parameters and throughput values table was updated to add new model variants (including gpt-5.4-mini) and to provide version-specific details for each model. Some throughput and scaling values were also updated, notably for gpt-5.4-mini and related models.

• Added new model variants such as gpt-5.4-mini with specific deployment and throughput values
• Tables now include version dates for each model, clarifying which release the parameters apply to
• Updated throughput and scaling values for certain models (e.g., gpt-5.4-mini has higher Input TPM per PTU than previous entries)

The region availability tables for both Global and Data Zone deployments were updated, notably expanding the Data Zone support for the gpt-5.5 model to all listed US regions. Some Global deployment models (gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-4o, gpt-4o-mini, gpt-5-mini, gpt-5-nano) now show reduced regional availability, with several regions marked as unavailable.

• gpt-5.5 now available in all US Data Zone regions (previously only some regions)
• gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-4o, gpt-4o-mini, gpt-5-mini, and gpt-5-nano show reduced regional availability for Global deployments
• No other new models or regions added

The documentation now adds support for using Work IQ with the Python and C# SDKs, and introduces a new, more streamlined method for connecting agents to Work IQ using the Foundry Toolkit for Visual Studio Code. This makes it easier for developers to integrate Work IQ and expands the range of supported SDKs.

• Python and C# SDKs are now supported for Work IQ integration
• Instructions added for using Foundry Toolkit for Visual Studio Code to add Work IQ as a tool
• Clarified toolbox creation workflow and connection options

The documentation now reflects expanded support for several tools (including Fabric IQ, Tool Search, and Work IQ) in the Foundry Toolkit and SDKs, and introduces new guidance and code samples for using these tools, especially with Power BI semantic models. There are also updates to prerequisites, including a new unified Foundry CLI extension and required headers for toolbox MCP endpoints.

• Fabric IQ, Tool Search, and Work IQ tools now supported in more SDKs and the Foundry Toolkit
• New guidance and code samples for using Fabric IQ with Power BI semantic models, including recommended allowed_tools configuration
• Prerequisites updated to require the unified Foundry CLI extension bundle instead of individual extensions

The documentation now clarifies SDK support for Fabric IQ, adds a regional availability limitation, requires the Foundry Toolkit, and introduces recommended tool restrictions for Power BI semantic models to improve agent reasoning and query execution.

• Python and JavaScript SDKs are now listed as supported for Fabric IQ.
• Fabric IQ is not available in regions where Power BI is the only Fabric workload.
• Installation of Visual Studio Code and Foundry Toolkit is now a prerequisite.

The documentation now clarifies that only validated connectors provide the new managed MCP server experience in the Foundry portal, and introduces a new section listing all supported regions for managed MCP servers powered by connector namespaces. This helps users understand availability and connector eligibility for the new experience.

• Clarified that only validated connectors surface the managed MCP server experience in the Foundry portal; unvalidated connectors use the previous experience.
• Explained that code-first configuration (API, SDK, azd) is not gated by connector validation.
• Added a new section listing all Azure regions where managed MCP servers powered by connector namespaces are available.

The quickstart now highlights support for 'instant models' (preview), including guidance to use the West US 3 region and the option to skip manual deployment for instant models. The default deployment example is updated from gpt-4.1-mini to gpt-5.1-mini, reflecting access to a newer model.

• Added tip to use West US 3 region for instant model (preview) support
• Noted that model deployment can be skipped if using instant models
• Updated deployment example from gpt-4.1-mini to gpt-5.1-mini

A new option to use an 'instant model' (preview) instead of deploying your own model in Microsoft Foundry has been added, allowing users to quickly try out the quickstart without the full deployment process.

• Introduced the ability to use an instant model (preview) in West US 3 region
• Users can substitute any instant model for 'gpt-5.1-mini' in code examples
• Reduces setup time for trying out the quickstart

The deployment parameters and throughput values table for the latest Azure OpenAI models was updated, notably removing 'gpt-5.4-mini' and adjusting several minimum deployment and scale increment values for models such as gpt-5.3-codex and gpt-5.2. These changes affect how customers plan and size their deployments for specific models.

• Removed 'gpt-5.4-mini' from the list of latest models
• Adjusted regional provisioned minimum deployment and scale increment values for gpt-5.3-codex, gpt-5.2, and related models
• Updated input TPM per PTU and output-to-input ratio for several models

A new section describes how model router deployments can now be governed using Azure Policy, aligning with existing Foundry model deployment policies. This enables organizations to enforce deployment restrictions consistently across multiple interfaces.

• Model router deployments now honor Foundry model deployment policies via Azure Policy
• Policy enforcement applies across portal, REST API, CLI, and ARM templates
• Guidance provided for both IT admins and developers

A new model, gpt-5.5 (2026-04-24), has been added to the list of supported models for priority processing, with a higher latency target value than previous models. This expands the available options for users requiring faster processing.

• Added gpt-5.5 (2026-04-24) to the latency target table
• Set a new latency target for gpt-5.5: 99% > 100 Tokens Per Second

A new section was added describing how model router deployments can now be governed using Azure Policy, aligning with existing controls for standard Foundry model deployments. This allows organizations to enforce deployment policies on the subset of models included in model router deployments.

• Model router deployments now support governance via Azure Policy
• Policy enforcement applies to the subset of models included in a deployment
• Consistent enforcement across Foundry portal, REST API, Azure CLI, and ARM templates

The article now documents support for tracing with the Microsoft OpenTelemetry distro for LangChain and LangGraph, replacing the previous langchain-azure-ai package. It also updates environment variable and package requirements, and clarifies which agent types are in preview. These changes are important for users implementing tracing in AI agent frameworks, as they affect setup and instrumentation steps.

• Replaces langchain-azure-ai package with microsoft-opentelemetry distro for LangChain/LangGraph tracing
• Updates sample code and installation instructions to use Microsoft OpenTelemetry distro
• Clarifies preview status for hosted, workflow, and external agents

A new section was added describing the quota system for instant models (preview), introducing the concept of a global quota pool distinct from regional quotas. This informs users about how instant model usage is managed and how to request increases.

• Introduces global quota pool for instant models, separate from regional quotas
• Explains automatic sharing of global quota across instant model usage
• Provides guidance on what to do if instant model requests are throttled

The documentation now provides expanded guidance on evaluation approaches, troubleshooting, and best practices for running evaluations in the Microsoft Foundry portal. It introduces new options for evaluation targets, scopes, and instant models, and clarifies prerequisites and workflow steps.

• Added detailed tables and guidance for choosing evaluation approaches based on target, scope, and data source
• Introduced troubleshooting section for common evaluation issues (timeouts, field mapping, quota)
• Added best practices section for simulation, conversation, model, and dataset evaluations

The region availability tables for several GPT and related models have been updated, expanding support for more Azure regions, particularly in Canada, and correcting availability for certain models and versions. This change increases deployment flexibility for customers in those regions.

• gpt-4.1, gpt-4.1-mini, and gpt-4.1-nano now available in canadacentral and canadaeast in Global Standard and Provisioned Managed tables
• gpt-4o (2024-08-06) and gpt-4o-mini (2024-07-18) now available in canadacentral and canadaeast
• gpt-5.4-mini now available in all listed regions (was previously only in westus3 for Provisioned Managed and Data Zone)

The documentation now introduces 'instant models (preview)', a new deployment type that allows users to run inference on supported models without creating a deployment, streamlining quick starts and prototyping. This addition is reflected in both the deployment comparison table and the guidance for choosing a deployment type.

• Added 'instant models (preview)' as a deployment type requiring no deployment setup
• Updated deployment comparison table to include instant models with details on billing and use cases
• Guidance now recommends instant models for quick start, prototyping, and trying new models

The documentation now explains how to use AI Gateway with multiple projects, including per-project token limits and quotas, and introduces clear procedures for disabling or deleting an AI Gateway. It also removes the restriction that an API Management instance cannot be associated with another AI Gateway, potentially allowing more flexible configurations.

• Added section on using AI Gateway with multiple projects, including per-project token limits and quotas
• Clarified that new projects have AI Gateway enabled by default, while existing ones must be added manually
• Explained that separate API Management instances require separate Foundry resources for strict isolation

A new section was added detailing deployment-type-specific permissions, particularly for managed compute deployments (Preview) in Microsoft Foundry. This clarifies the required Azure resource provider operations and their mapping to built-in roles, which is important for administrators planning secure and effective role assignments.

• Added section on permissions for specific deployment types, focusing on managed compute deployments (Preview)
• Lists five required Azure resource provider operations for managing managed compute deployments
• Explains differences in role-to-permission mapping compared to standard deployments

A new section describes conversation-level evaluation for coherence, allowing assessment of logical flow across entire conversations rather than just individual responses. This expands the evaluation capabilities for Copilot agents.

• Introduces conversation-level coherence evaluation using evaluation_level="conversation"
• Evaluator now assesses logical flow and topic consistency across multiple turns
• Requires the messages field with the full conversation in OpenAI message format

A new capability has been added allowing custom evaluators to score entire conversations, not just individual turns. This enables more comprehensive evaluation metrics, such as compliance checks across a session or conversation length scoring.

• Introduced conversation-level evaluation by setting evaluation_level="conversation"
• grade() function now receives the full conversation as item["messages"]
• Provided examples for session-level compliance checks and conversation length scoring

A new 'Customer Satisfaction' agent evaluator has been introduced, allowing measurement of user satisfaction with agent performance. Additionally, a new section on 'Quality evaluation' describes the Quality Grader evaluator, which assesses multiple dimensions of response quality and aligns with Copilot Studio's evaluation tools.

• Added 'Customer Satisfaction' as a built-in agent evaluator for system evaluation
• Documented required inputs and parameters for the new evaluator
• Introduced a new section on 'Quality evaluation (preview)' describing the Quality Grader evaluator

The documentation introduces 'instant models (preview)' that allow inference without deployment, and significantly updates the managed compute deployment option, now in public preview with new capabilities, billing, and authentication details. The managed compute deployment section is replaced with a more detailed preview section, reflecting changes in supported models, deployment process, and feature availability.

• Added 'instant models (preview)' allowing inference without deployment
• Managed compute deployment is now in public preview, with registration required
• Managed compute supports open-source, partner, industry, and custom models via a unified endpoint

The quickstart now requires the 'azd microsoft.foundry' extension instead of 'azd ai agent', adds optional GitHub Copilot for Azure plugin guidance, introduces non-interactive initialization, and updates the agent scaffolding process. These changes streamline setup and enable automation scenarios.

• Replaces 'azd ai agent' extension with 'azd microsoft.foundry' as a prerequisite
• Adds optional recommendation for GitHub Copilot for Azure plugin
• Changes agent project initialization to use a manifest URL and provides non-interactive setup instructions

The documentation introduces the Responses API as a unified entry point for accessing Foundry models and tools, expands supported agent frameworks (including OpenAI Agents SDK, Anthropic Agent SDK, GitHub Copilot SDK), and clarifies agent types by removing workflow agents. These changes broaden integration options and clarify how to use and deploy agents in Foundry.

• Introduces the Responses API as a single entry point for all agent types and external processes
• Expands supported frameworks for Hosted agents to include OpenAI Agents SDK, Anthropic Agent SDK, and GitHub Copilot SDK
• Clarifies agent types by removing workflow agents and focusing on prompt and Hosted agents

The documentation now describes new preview capabilities for memory in Foundry Agent Service, including granular memory item operations, store-level retention controls (default TTL), and procedural memory. The SDK usage table and code samples have been updated to reflect these enhancements.

• Announces support for create, read, update, list, and delete operations on individual memory items.
• Introduces store-level default retention controls, including default TTL for memory entries.
• Adds procedural memory as an option when creating memory stores.

The documentation now references and demonstrates the use of the newer 2026-05-01-preview REST API version for connecting a Foundry IQ knowledge base to Foundry Agent Service, replacing the previous preview version. It also adds important compliance, security, and responsible AI usage guidance for the new preview features.

• Updated all API references and code samples to use 2026-05-01-preview instead of 2025-11-01-preview
• Added explicit compliance and data boundary warnings for connections to third-party and Microsoft services
• Introduced guidance on responsible AI mitigations and risk management for MCP implementations

The documentation now adds support for A2A protocol version 1.0 in addition to 0.3, clarifies which transports are supported for each version, and updates guidance for agent card URLs and SDK usage. This enables new integrations to target the latest protocol version and clarifies compatibility and authentication requirements.

• A2A protocol version 1.0 is now supported and recommended for new integrations
• Agent card URLs for both v1.0 and v0.3 are documented, with guidance on which to use
• Transport support table added, showing v1.0 only supports JSONRPC, not HTTP+JSON

A new WebSocket-based protocol ('invocations_ws') for hosted agents is now documented, enabling bidirectional streaming scenarios such as real-time voice agents. This protocol is in preview and currently limited to the North Central US region. Minor clarifications were also made to permissions terminology.

• Added documentation for 'invocations_ws' WebSocket protocol for hosted agents
• WebSocket protocol supports bidirectional streaming for real-time voice and interactive media scenarios
• Clarified that the WebSocket protocol is in preview and region-limited

The documentation now introduces the Invocations (WebSocket) protocol for hosted agents, currently in preview and available only in North Central US. Additionally, a new resource for optimizing agent instructions automatically has been added to the next steps.

• Added Invocations (WebSocket) protocol as a new preview feature for hosted agents
• Clarified protocol availability by region and protocol type
• Added 'Agent optimizer overview' to next steps for automatic instruction optimization

A new step, 'Optimize hosted agents (preview)', has been added to the agent development lifecycle, introducing the agent optimizer for automatic improvement of hosted agent instructions and skill discovery. This expands the lifecycle with a new optimization capability.

• Added 'Optimize hosted agents (preview)' as a lifecycle step
• Introduces agent optimizer for automatic instruction improvement and skill discovery

A new section was added detailing how to grant OpenTelemetry read/write permissions to a hosted agent for Agent 365 Observability, including step-by-step Azure CLI instructions. Additionally, terminology was updated from 'digital worker' to 'autopilot' throughout the integration section.

• Added instructions for assigning Agent365.Observability.OtelWrite app role to agent service principals
• Provided Azure CLI commands for permission assignment and verification
• Changed terminology from 'digital worker' to 'autopilot' for published agents

Page removed or relocated. The quickstart documentation for optimizing a hosted agent using the Agent Optimizer (preview) has been removed. This page provided step-by-step instructions for installing tools, deploying agents, running optimizations, and managing results, which are now unavailable.

• Loss of detailed setup and usage instructions for Agent Optimizer CLI extension
• No longer documents how to provision resources, deploy agents, or run optimizations
• Guidance for troubleshooting and resource cleanup is missing

Page removed or relocated. The documentation page detailing how to create evaluation datasets for Agent Optimizer (in preview) has been removed. This eliminates official guidance on generating, customizing, and troubleshooting evaluation datasets for Copilot agent optimization workflows.

• Loss of instructions for generating evaluation datasets via CLI and manual methods
• No longer documents dataset formats, field requirements, and best practices for evaluation
• Removes troubleshooting guidance for dataset-related errors in agent optimization

The documentation now clarifies that Private Class A IP address ranges (10.x.x.x) for virtual networks are only supported in specific Azure regions, listing those regions explicitly. This is important for users planning network configurations for Foundry Agent Service deployments.

• Added explicit list of Azure regions where Private Class A (10.x.x.x) IP ranges are supported
• Guidance to use Class B or C IP ranges in other regions

The documentation now clarifies recommended deployment paths for hosted agents, adds explicit SDK prerequisites, and provides a comparison table for deployment methods. Packaging instructions are reorganized for clarity, especially for language-specific workflows.

• Added explicit requirement for azure-ai-projects and azure-identity Python packages
• Clarified when to use Azure Developer CLI, VS Code, SDKs, or REST API for deployment
• Introduced a comparison table outlining best use cases for each deployment method

A new note clarifies that organizations using Azure Policy for model deployment must allow the 'Microsoft' publisher (for model router) and any other relevant publishers, or deployments will be blocked. This is important for administrators to ensure successful deployments.

• Added guidance about Azure Policy requirements for allowed publishers when deploying model router and routed models

The introduction now clarifies that to deploy and use the model router in Microsoft Foundry portal, 'Microsoft' must be included in the allowed publishers list, along with any other publishers for models you wish to route. This prevents accidental policy blocks on model router deployments.

• Added guidance to include 'Microsoft' in allowed publishers for model router deployment
• Clarified need to add each routed model's publisher to avoid policy blocks

A new limitation is clarified: Virtual network (VNet) integration is not supported for memory stores in Foundry Agent Service. This informs users about a key networking constraint.

• Added explicit statement that VNet integration isn't supported for memory stores

The documentation now clarifies and expands on how to enable tool search in a toolbox using the Foundry Toolkit for Visual Studio Code, and provides clearer instructions for verifying and fine-tuning tool discovery, including SDK-specific guidance. This helps users more easily configure and validate tool search capabilities in Copilot toolboxes.

• Added explicit prerequisite to install Visual Studio Code and Foundry Toolkit.
• Described enabling tool search via a checkbox in the Foundry Toolkit UI.
• Clarified verification steps and referenced SDK/language-specific tabs for validation.

The section clarifies how Azure Reservations apply to different deployment types, specifying that Global reservations can cover deployments across multiple regions, while Data Zone and Regional reservations must match the deployment's region.

• Clarifies that Global reservations can cover Global PTU deployments across multiple regions
• Specifies that for Data Zone and Regional deployments, the reservation region must match the deployment's region

The documentation now clarifies that for Global provisioned deployments, the reservation region does not need to match the deployment region, as a single Global reservation can cover deployments across multiple regions. This helps users avoid unnecessary constraints or confusion when purchasing reservations.

• Clarifies that Global reservations can cover deployments in multiple regions
• Specifies that region matching is only required for Data Zone and Regional deployments

The section now clarifies that Global Provisioned reservations are not region-specific and can apply to deployments across multiple regions, and that Global reservations can be consolidated. This provides important guidance for customers managing reservations for Copilot model deployments.

• Clarifies that Global reservations are not tied to a specific region
• Explains that a single Global reservation can cover deployments in multiple regions
• Notes the option to consolidate Global reservations or maintain one-to-one mapping per region

The documentation now clarifies the recommended approach for enabling tracing, emphasizing server-side tracing as the default and easiest method, and specifies a required Azure role for querying telemetry data.

• Added guidance that server-side tracing is enabled automatically with Application Insights and requires no code changes.
• Clarified that server-side tracing works for any agent hosted in Foundry.
• Explained when to use client-side instrumentation for additional visibility.

The built-in policy for model deployment in Microsoft Foundry portal has been renamed, and the documentation now clarifies that the policy's definition ID remains unchanged, so existing assignments are unaffected. The example Azure CLI commands and display names have been updated to reflect the new policy name.

• Policy name changed from 'Cognitive Services Deployments should only use approved Registry Models' to 'Foundry model deployments should only use approved models'
• Added note clarifying that the policy definition ID is unchanged and existing assignments require no action
• Updated Azure CLI example queries and assignment names/display names to match the new policy name

The documentation clarifies the roles and relationships of the Foundry SDK, Agent Framework, and OpenAI SDK, emphasizing the Foundry SDK as a foundational thin-client and specifying which features and APIs are accessible through each SDK and endpoint. It also clarifies that the OpenAI SDK endpoint offers best latency and maximum compatibility, but does not provide access to Foundry-exclusive features.

• Describes the Foundry SDK as a thin-client foundational SDK for all Foundry project APIs
• Clarifies that Agent Framework depends on the Foundry SDK and is for hosted/multi-agent systems
• Specifies that the OpenAI SDK endpoint offers best latency and maximum OpenAI compatibility, but lacks Foundry-exclusive features

The documentation was updated to remove preview disclaimers, clarify evaluation scenarios, and improve instructions for setting up and running cloud evaluations with the Foundry SDK. The changes provide clearer guidance on supported data sources, evaluation flows, and SDK usage, but do not introduce new features.

• Removed public preview disclaimer from the introduction, indicating general availability.
• Expanded and clarified the 'How cloud evaluation works' section to better explain evaluation scenarios and data source options.
• Updated SDK installation instructions to require version 2.2.0 or higher.

The documentation clarifies the scope of GA support in the new Microsoft Foundry portal, specifying which project and resource types are supported, the authentication requirements for different features, and the readiness status of individual features. It also provides more precise guidance on unsupported features and migration paths from classic to new portal experiences.

• Clarifies that standalone Azure OpenAI resources and hub-based projects are not supported in the new portal and require migration or continued use of the classic portal.
• Specifies that API key authentication is not supported for evaluations, dataset tab, Content Understanding, agents, and workflows, which require Microsoft Entra ID.
• Updates the feature readiness table with more granular status (e.g., partial GA for some features, preview for others).

The documentation now provides a more detailed and structured breakdown of regional support for various Copilot evaluation features, replacing previous lists of unsupported regions with explicit tables of supported regions for agent playground, batch evaluations, and risk/safety evaluators. This makes it clearer which regions are eligible for each feature.

• Added explicit tables listing supported regions for agent playground and batch evaluations.
• Clarified and expanded regional support details for risk and safety evaluators and AI red teaming.
• Removed previous lists of unsupported regions and replaced with positive lists of supported regions.

The documentation clarifies that the 'context' input for the Groundedness evaluator is now optional (though recommended), and introduces 'tool_definitions' as an input for agent response mode. This provides more precise guidance on required and optional inputs for RAG evaluation.

• 'context' input for Groundedness evaluator is now marked as optional and recommended
• 'tool_definitions' added as an input for Groundedness evaluator in agent response mode
• Clarifies input requirements for different evaluator modes

The 'Feature support' section was expanded to include new supported features such as Guardrail (RAI policy) and Skill references, indicating broader capabilities for toolbox management in Foundry. No major changes were made to the introductory or step-by-step sections.

• Added Guardrail (RAI policy) as a supported feature in the feature matrix
• Added Skill references as a supported feature in the feature matrix

The documentation now clarifies that the Azure Speech MCP tool in Foundry requires the Agent Service Enterprise tier and provides more detailed setup, troubleshooting, and security instructions. It also adds explicit steps for generating SAS URLs and selecting models that support tool calling.

• Explicit requirement for Agent Service Enterprise tier to use MCP tools
• Added troubleshooting guidance for 'Invalid tool value(s): mcp' error
• Clarified steps for assigning Storage Blob Data Contributor role and generating SAS URLs

A new clarification was added about managed OAuth with Microsoft Entra: tokens scoped to a Microsoft audience cannot be sent to custom or third-party MCP servers, and an explicit error message is returned if attempted. This helps users understand security boundaries and expected errors.

• Added warning that Microsoft Entra tokens via managed OAuth cannot be sent to custom/third-party MCP servers
• Documented the specific error message returned in this scenario

This new documentation introduces and explains how Microsoft 365 Copilot can use Power BI data, including reports and semantic models, to answer user questions. It details how admins can manage and enable access to Fabric data for Copilot, outlining both user access controls and necessary admin settings.

• Microsoft 365 Copilot can now ground responses in Power BI reports and semantic models.
• Admins can control which users or groups have access to Fabric data in Copilot.
• Fabric data must be enabled and shared via the Fabric admin portal for Copilot integration.

This new documentation introduces 'agent orchestration' in Microsoft 365 Copilot, allowing admins to configure rules that route specific user prompts to designated agents for authoritative responses on specialized topics. It details setup, rule management, user experience, feedback mechanisms, best practices, and current limitations.

• Admins can define prompt-based rules to route users from Copilot to specific agents (e.g., HR, policy).
• Rules can be created, edited, activated, disabled, or deleted individually or in bulk.
• Users see clear indications when responses come from agents and can exit back to Copilot.

This new documentation introduces dynamic tool discovery for MCP plugins in Microsoft 365 Copilot, allowing agents to fetch and update their tool lists at runtime from the MCP server. This enables near real-time access to new or updated tools without requiring agent repackaging or republishing, and supports user-specific tool surfaces. The page also details admin controls, audit logging, and trust/safety validations for this new capability.

• Dynamic tool discovery allows agents to resolve available tools at runtime from the MCP server, not just from a static manifest.
• Users gain near real-time access to new, updated, or removed tools without agent republishing.
• Supports user-specific tool lists based on SKU, role, or entitlement.

This new documentation introduces the 'copilotConversation' resource type for the Work IQ Chat API, detailing its methods, properties, and relationships. It enables developers to programmatically create and manage Copilot conversations, including synchronous and streaming chat interactions.

• Defines the 'copilotConversation' resource type for Work IQ Copilot conversations
• Documents methods for creating conversations and sending chat messages (synchronous and streaming)
• Lists properties such as agentId, state, and turnCount

This new page introduces the Work IQ REST API (preview), which allows developers to integrate Microsoft 365 Copilot's multiturn conversational capabilities, grounded in enterprise and web search, into custom applications. It details the API's capabilities, compliance features, licensing requirements, and known limitations.

• Announces the preview availability of the Work IQ REST API for Microsoft 365 Copilot integration
• Describes support for multiturn conversations with enterprise and web search grounding
• Explains compliance, security, and data access controls inherited from Microsoft 365

This new documentation introduces the Work IQ Chat API's streamed conversation endpoint, enabling developers to continue multi-turn, streamed conversations with Microsoft 365 Copilot. It details required permissions, request/response structure, and provides concrete usage examples, including how to ground Copilot responses with additional context and files.

• Introduces POST /copilot/conversations/{conversationId}/chatOverStream endpoint for streamed Copilot chats
• Documents required delegated permission (WorkIQAgent.Ask) and unsupported scenarios
• Explains request parameters for sending messages, context, location, and contextual resources

This new documentation introduces the Work IQ Chat API, enabling developers to create and continue synchronous multi-turn conversations with Microsoft 365 Copilot, including advanced context handling and resource grounding. It details permissions, request/response structure, and provides practical examples for integrating Copilot chat into custom applications.

• Introduces Work IQ Chat API for synchronous Copilot conversations
• Supports sending chat messages with additional context and location hints
• Allows use of OneDrive/SharePoint files and toggling web search grounding as context

This new documentation introduces the Work IQ MCP toolset, detailing 10 tools that enable CRUD operations, schema discovery, and conversational Copilot interactions over Microsoft 365 resources via the Model Context Protocol. It provides comprehensive reference information, including parameters, responses, and usage examples for each tool.

• Describes entity tools for CRUD operations on Microsoft 365 resources via Graph endpoints
• Documents conversational tools for interacting with Microsoft 365 Copilot and listing available agents
• Introduces schema tools for runtime discovery of API paths and OpenAPI schemas

This new page introduces the Work IQ MCP server, which exposes Microsoft 365 intelligence and Copilot capabilities to AI agents via a unified protocol. It details the design principles, tool categories, and operational model, highlighting a fixed set of generic tools for interacting with Microsoft 365 resources and invoking Copilot for natural-language reasoning.

• Describes the Model Context Protocol (MCP) as a unified endpoint for AI agents to access Microsoft 365 intelligence and Copilot
• Explains a fixed set of 10 generic tools for CRUD operations, Copilot invocation, and schema discovery
• Outlines design principles prioritizing extensibility via resource paths, runtime schema introspection, and policy-based access control

This new documentation introduces Microsoft Work IQ CLI, a tool that connects AI assistants (including Copilot) to Microsoft 365 data via a command-line interface and Model Context Protocol server. It details installation, supported platforms, usage scenarios, security, and integration with GitHub Copilot, enabling natural language queries over workplace data.

• Introduces Work IQ CLI and MCP server for accessing Microsoft 365 Copilot data
• Describes integration with AI assistants like GitHub Copilot in VS Code
• Lists supported data types: emails, meetings, documents, Teams messages, people, and more

This new page introduces the Work IQ API (preview), enabling developers to build agentic and AI-powered applications that securely reason over Microsoft 365 data using multiple protocols. It details supported protocols, data types, authentication, and provides concrete usage examples, marking a significant expansion in Copilot extensibility and integration capabilities.

• Introduction of Work IQ API for secure, permission-trimmed access to Microsoft 365 data
• Support for multiple interaction protocols: Agent-to-Agent (A2A), Model Context Protocol (MCP), and REST
• Detailed examples for A2A and MCP protocols, including multi-turn conversations and tool-based context access

This new page introduces Work IQ, a workplace intelligence layer for Microsoft 365, and provides a comprehensive quickstart for developers to use the Agent-to-Agent (A2A) protocol to build agents and workflows that interact with Microsoft 365 data. It details setup, authentication, protocol usage, and troubleshooting, enabling secure, permission-aware agent development.

• Introduces Work IQ as a semantic intelligence layer for Microsoft 365 and connected systems
• Documents the A2A protocol (v1.0 and v0.3) for agent communication, including wire formats and multi-turn conversation support
• Provides step-by-step setup for enabling Work IQ API, including service principal and app registration

This new page documents the Agent-to-Agent (A2A) protocol for Work IQ, enabling autonomous AI agents to communicate and collaborate with Microsoft Copilot using structured tasks. It introduces protocol versions, authentication, agent discovery, message exchange (including multi-turn conversations), and task management capabilities.

• Introduces A2A protocol for agent-to-agent communication with Copilot
• Supports both protocol versions 1.0 and 0.3 with version negotiation
• Details delegated authentication via Microsoft Entra ID

This new page introduces Work IQ, a comprehensive workplace intelligence platform that integrates chat, context, tools, and workspaces to enhance agent capabilities within Microsoft 365 and beyond. It details how Work IQ securely unifies data from Microsoft and external sources, provides semantic understanding, and enables extensibility for building intelligent agents and workflows.

• Introduction of Work IQ as a semantic intelligence layer for organizational data and agentic systems
• Integration of Microsoft 365, Dynamics 365, Power Apps, and external data via Copilot connectors and federated connectors
• Description of context features such as Copilot memory, semantic indexing, and Dataverse intelligence

Page removed or relocated. The documentation for Microsoft Work IQ CLI, a tool enabling AI assistants to access Microsoft 365 Copilot data via CLI and MCP server, has been removed. This leaves a gap in official guidance for installing, configuring, and using Work IQ to connect AI assistants with workplace data.

• Loss of installation and setup instructions for Work IQ CLI
• Removal of usage scenarios and command references for integrating AI assistants with Microsoft 365 data
• No longer available: security, privacy, and platform support details for Work IQ

Page removed or relocated. The documentation for the Work IQ API quickstart (preview) has been removed. This page provided essential guidance for enabling, authenticating, and using the Work IQ API and A2A protocol to build applications that query Microsoft 365 data using Copilot capabilities.

• Loss of step-by-step instructions for enabling and configuring the Work IQ API in an organization
• Removal of detailed quickstart guides for using the Agent-to-Agent (A2A) protocol with Copilot data
• No longer documents authentication flows, sample code usage, and troubleshooting for Work IQ API integration

Page removed or relocated. The documentation page for the Work IQ API overview (preview) has been removed. This eliminates official guidance on how to use Work IQ to build agentic and AI-powered applications that reason over Microsoft 365 data, including supported protocols, authentication, and migration from Copilot Chat API.

• Loss of overview and conceptual explanation of Work IQ's role in Microsoft 365 Copilot and agentic applications
• Removal of protocol descriptions (A2A, MCP, REST) and guidance on choosing between them
• No longer provides API usage examples, including authentication and security details

Page removed or relocated. The documentation page providing an overview of Work IQ, the intelligence layer that personalizes Microsoft 365 Copilot, has been removed. This eliminates detailed guidance on Work IQ's architecture, extensibility, data integration, and developer tools, which are critical for understanding and leveraging advanced Copilot capabilities.

• Loss of explanation of Work IQ's three-layer architecture (data, context, skills/tools)
• No longer documents how Work IQ integrates Microsoft 365, Dynamics 365, Power Apps, and external data
• Removal of guidance on Copilot extensibility via connectors, MCP tools, and business skills

Page removed or relocated. The documentation for the Work IQ REST API (preview), which enabled programmatic multiturn conversations with Microsoft 365 Copilot using enterprise and web search grounding, has been removed. This leaves a gap in official guidance for integrating Copilot via REST API, including its capabilities, limitations, and licensing requirements.

• Loss of documentation for the Work IQ REST API's integration with Microsoft 365 Copilot
• Missing details on API capabilities such as enterprise and web search grounding
• No longer available: official list of known limitations and licensing information

Page removed or relocated. The documentation for creating Copilot conversations via the Work IQ Chat API has been removed. This means guidance on starting new conversations with Microsoft 365 Copilot using this API is no longer available, which could indicate deprecation or relocation of this capability.

• Removed documentation for POST /copilot/conversations endpoint
• No longer provides required permissions, request/response details, or usage examples
• Uncertainty if the feature is deprecated or the documentation was moved

Page removed or relocated. The documentation for the 'chatOverStream' API endpoint, which enabled developers to continue streamed multi-turn conversations with Microsoft 365 Copilot via the Work IQ Chat API, has been removed. This leaves a gap in official guidance for implementing streaming chat interactions with Copilot, including permissions, request/response structure, and usage examples.

• Removal of documentation for the POST /copilot/conversations/{conversationId}/chatOverStream endpoint
• Loss of detailed API usage instructions, including permissions, request/response schema, and streaming behavior
• Examples for sending messages and providing additional context to Copilot via streaming are no longer available

Page removed or relocated. The documentation for the Work IQ Chat API, which enables developers to create and continue multi-turn conversations with Microsoft 365 Copilot via a REST endpoint, has been removed. This leaves a gap in guidance for integrating synchronous Copilot chat capabilities, including using context files, toggling web search, and providing additional context.

• Removal of API reference for synchronous Copilot chat conversations
• Loss of permission requirements and usage examples for the Chat API
• No longer documents how to send context files, toggle web search grounding, or provide extra context in Copilot chats

This new page documents Microsoft's use of third-party AI subprocessors in Microsoft 365 Copilot and related services, detailing which partners are involved, how they are onboarded, and how customers can manage and monitor their use. It introduces specific examples such as Anthropic's Claude models and Fireworks AI, and clarifies governance, security, and notification processes.

• Introduces the concept of AI subprocessors for Microsoft 365 Copilot and related services
• Details onboarding, security, and compliance requirements for third-party AI subprocessors
• Specifies integration of Anthropic's Claude models and Fireworks AI in Copilot experiences

This new documentation introduces the copilotWebContext resource type for the Work IQ Chat API, detailing how web search grounding can be enabled or disabled in Copilot conversations. This is important for developers integrating or customizing Copilot's web search behavior.

• Describes the copilotWebContext resource type for controlling web search grounding in Copilot conversations
• Documents the isWebEnabled property, allowing toggling of web search grounding per chat message
• Provides JSON representation for API integration

This new page documents the 'copilotConversationResponseMessage' resource type for the Work IQ Chat API, detailing its properties and JSON structure. It enables developers to understand and integrate with Copilot conversation message handling in Work IQ scenarios.

• Introduces the 'copilotConversationResponseMessage' resource type for Copilot conversations
• Details properties such as adaptiveCards, createdDateTime, id, references, sensitivityLabel, and text
• Provides a sample JSON representation for API integration

This new documentation page introduces the 'copilotConversationReference' resource type for the Work IQ Copilot API, detailing its properties and JSON structure. This enables developers to understand and integrate conversation reference data in their applications.

• Introduces 'copilotConversationReference' resource type
• Documents properties: isCitedInResponse, sensitivityLabel, targetLink
• Provides JSON representation for API integration

This new documentation introduces the copilotConversationLocation resource type, detailing how user location information is represented during Copilot conversations via the Work IQ Chat API. It specifies the properties available, including country, region, coordinates, time zone, and confidence level.

• Defines a new resource type for capturing user location in Copilot conversations
• Lists properties such as latitude, longitude, time zone, country/region, and confidence
• Provides a JSON schema for API integration

This new documentation introduces the 'copilotContextualResources' resource type for the Work IQ Chat API, detailing how to provide optional contextual resources such as files and web context to Copilot conversations. This enables richer, more informed responses from Copilot in chat scenarios.

• Defines the 'copilotContextualResources' resource type for Copilot conversations
• Describes support for including OneDrive and SharePoint file URIs as context
• Introduces the ability to specify web search grounding via 'webContext'

This new documentation introduces the copilotContextMessage resource type for the Work IQ Chat API, detailing how to provide extra context to Copilot conversations. It specifies the properties and JSON structure required for integration.

• Defines the copilotContextMessage resource type for Work IQ Copilot conversations
• Lists 'description' and 'text' as properties for providing additional context
• Provides a sample JSON representation for API usage

This new documentation introduces the Work IQ Chat API endpoint for creating Copilot conversations, enabling integration of Microsoft 365 Copilot into generative AI solutions. It details required permissions, request/response structure, and provides example usage.

• Introduces POST endpoint for creating Copilot conversations via Work IQ Chat API
• Specifies required delegated permission (WorkIQAgent.Ask) for work/school accounts
• Describes request and response formats, including example payloads

A new section details the general availability of usage-based targeting for Organizational Messages in the Microsoft 365 admin center, allowing admins to target Copilot adoption campaigns based on real user behavior. Additionally, Organizational Messages can now be delivered via email, expanding communication channels for Copilot-related guidance.

• Admins can now target Organizational Messages based on dynamic usage behaviors, not just group membership.
• Initial usage segments include users inactive with Copilot or Copilot in Teams.
• Organizational Messages can now be sent via email, in addition to existing surfaces.

Page removed or relocated. Documentation for the 'copilotWebContext' resource type, which controls web search grounding in Copilot conversations via the Work IQ Chat API, has been removed. It is unclear if this indicates deprecation or a documentation restructure, but details about enabling/disabling web search grounding are now missing.

• Loss of documentation for the 'copilotWebContext' resource type
• No longer describes how to enable or disable web search grounding in Copilot chat via API
• JSON schema and property details for this resource are no longer available

Page removed or relocated. Documentation for the 'copilotFile' resource type, which describes how to reference OneDrive or SharePoint files as context in Copilot conversations via the Work IQ Chat API, has been removed. It is unclear if this indicates deprecation or relocation, but details on this integration are now missing.

• Removal of documentation for referencing files as context in Copilot via Work IQ Chat API
• Loss of property and JSON schema details for 'copilotFile' resource type

Page removed or relocated. The documentation for the 'copilotConversationResponseMessage' resource type has been removed. This means that details about the structure and properties of messages in Copilot conversations via the Work IQ Chat API are no longer available at this URL, which may impact developers relying on this reference.

• Loss of property descriptions for copilotConversationResponseMessage
• JSON schema for the resource type is no longer documented
• No guidance on how to interpret or construct conversation response messages

Page removed or relocated. Documentation for the copilotConversationLocation resource type has been removed. This means details about how user location is represented in Copilot conversations via the Work IQ Chat API are no longer available, which may impact developers relying on this information.

• Removal of property definitions for user location (latitude, longitude, timeZone, countryOrRegion, confidence)
• Loss of JSON schema example for copilotConversationLocation resource
• No longer documents how location is represented in Copilot conversations via Work IQ Chat API

Page removed or relocated. The documentation for the 'copilotConversation' resource type, which describes how to create and manage Copilot conversations via the Work IQ Chat API, has been removed. This makes it harder for developers to understand or implement conversation management features using this API.

• Removal of API method documentation for creating and interacting with Copilot conversations
• Loss of property and relationship definitions for the copilotConversation resource
• Missing JSON schema and enumeration details for conversation state

Page removed or relocated. Documentation for the 'copilotContextualResources' resource type, which describes how to provide files and web context to Copilot conversations via the Work IQ Chat API, has been removed. It is unclear if this reflects deprecation, restructuring, or relocation, but details on this API capability are now missing.

• Loss of documentation for specifying contextual files and web context in Copilot chat APIs
• JSON schema and property descriptions for 'copilotContextualResources' are no longer available

Page removed or relocated. Documentation for the 'copilotContextMessage' resource type has been removed, which means information about how to provide extra context to Copilot conversations via the Work IQ Chat API is no longer available. It is unclear if this reflects deprecation or a documentation restructure.

• Removal of property and JSON schema details for 'copilotContextMessage'
• Loss of guidance on adding extra context to Copilot conversations via API

Page removed or relocated. The documentation page detailing API permissions for Work IQ (including setup, authentication, and permission scopes) has been removed. This leaves a gap in official guidance for configuring and securing applications that integrate with Work IQ APIs.

• Loss of instructions for enabling Work IQ API via service principal
• Missing reference for the application ID URI and OAuth scope usage
• Removal of detailed permission descriptions, especially for WorkIQAgent.Ask

This new page documents the 'searchSensitivityLabelInfo' resource type, detailing its properties and JSON structure for use with Microsoft Graph beta APIs. It enables developers to understand how sensitivity label information is represented and can be used in UI scenarios.

• Introduces the 'searchSensitivityLabelInfo' resource type for information protection labels
• Lists properties such as color, displayName, priority, sensitivityLabelId, and tooltip
• Provides a sample JSON representation for implementation reference

This new page documents the 'copilotFile' resource type, which represents a OneDrive or SharePoint file sent as context into a Copilot conversation via the Work IQ Chat API. It specifies the structure and properties of this resource.

• Introduces the 'copilotFile' resource type for Work IQ Chat API
• Documents the 'uri' property for referencing OneDrive or SharePoint files
• Provides a JSON representation for API integration

This new documentation page introduces the 'copilotConversationRequestMessageParameter' resource type for the Work IQ Chat API, detailing its structure and required properties for sending chat messages to Copilot conversations.

• Defines the 'copilotConversationRequestMessageParameter' resource type
• Specifies the required 'text' property for chat messages
• Provides a JSON example for API integration

This new documentation introduces the copilotConversationReferenceMap resource type, describing its structure and purpose as a keyed map of copilotConversationReference objects. This is relevant for developers working with Copilot-related APIs in the Microsoft Graph beta endpoint.

• Introduces copilotConversationReferenceMap resource type
• Provides JSON structure for the resource
• Clarifies that this is a beta API and not supported for production use

This new page clarifies the distinctions between different types of AI models and providers (Microsoft-hosted, AI Subprocessors, and AI Independent Processors) used in Microsoft Online Services, including Copilot. It explains data handling, compliance, and admin controls relevant to each type, which is important for understanding privacy, security, and governance implications.

• Defines and contrasts Microsoft-hosted AI, AI Subprocessors, and AI Independent Processors
• Explains contractual and data protection differences for each AI provider type
• Describes admin controls for enabling/disabling third-party AI functionalities

The documentation now clarifies that Anthropic is a subprocessor for Microsoft 365 Copilot and that Microsoft may deploy additional AI models, which are hosted and operated by Microsoft under existing data protection commitments. This provides greater transparency about data handling and model usage.

• Explicitly states Anthropic is a subprocessor for Copilot
• Notes possibility of additional Microsoft-hosted AI models being used
• Affirms all models are governed by existing data protection commitments

Page removed or relocated. The documentation for the 'searchSensitivityLabelInfo' resource type has been removed. It is unclear if the resource type is deprecated or if the documentation has been relocated, but details about its properties and usage are now missing.

• Loss of documentation describing the 'searchSensitivityLabelInfo' resource type
• Missing property definitions and JSON representation for this resource

Page removed or relocated. The documentation for the 'copilotConversationRequestMessageParameter' resource type has been removed. It is unclear if this indicates deprecation, relocation, or restructuring, but information about how to structure chat messages for the Work IQ Chat API is now missing.

• Documentation for the chat message parameter resource type is no longer available
• JSON schema and property details for sending messages to Copilot via Work IQ Chat API are missing

Page removed or relocated. The documentation for the copilotConversationReferenceMap resource type has been removed. It is unclear if this indicates deprecation, relocation, or restructuring, but information about this resource type is now missing.

• copilotConversationReferenceMap resource type documentation no longer available
• JSON representation of the resource type is missing

Page removed or relocated. Documentation for the 'copilotConversationReference' resource type has been removed. It is unclear if this indicates deprecation or relocation, but information about this resource's properties and JSON structure is now missing.

• Loss of documentation for 'copilotConversationReference' resource type
• Properties and JSON representation for this resource are no longer available

This new documentation introduces the Fabric IQ Ontology MCP (preview), detailing how to connect Microsoft Copilot Studio agents to Fabric Ontology data using new MCP server endpoints and tools. It enables Copilot agents to query and reason over business ontologies via natural language, expanding Copilot's data integration and AI capabilities.

• Describes new MCP server endpoints for Fabric Ontology integration
• Introduces two new tools: list_ontology_entity_types and search_ontology for schema exploration and natural language querying
• Provides step-by-step guidance for connecting Fabric IQ Ontology to Copilot Studio agents

This new documentation introduces the Work IQ Word reference (preview), detailing a set of Copilot tools for programmatically creating, retrieving, and commenting on Word documents via OneDrive and SharePoint. It outlines available APIs for document creation, content retrieval, and comment management, enabling automation and integration scenarios.

• New API to create Word documents in OneDrive with HTML or plain text content
• Ability to fetch DOCX content and metadata from SharePoint/OneDrive URLs
• APIs to add and reply to comments in Word documents

This new documentation introduces the Windows 365 for Agents MCP server (preview), detailing its capabilities for full operational control of Windows 365 Cloud PCs via desktop, browser, and accessibility automation tools. It provides comprehensive reference for available tools, usage patterns, and common automation scenarios, enabling advanced automation and integration for Copilot and agent-based workflows.

• Introduces Windows 365 for Agents MCP server as a new automation endpoint for Cloud PCs
• Documents a wide range of desktop interaction tools (mouse, keyboard, window management, screenshots, OCR, etc.)
• Details browser automation capabilities, primarily for Microsoft Edge, including DOM interaction and tab management

This new documentation introduces the Work IQ User reference (preview), detailing a set of Copilot MCP server tools for accessing user profiles, managers, direct reports, and organizational hierarchy within an organization. It outlines available API endpoints, parameters, and key features for user and organizational data discovery and navigation.

• Introduces preview MCP server for user and organizational awareness in Copilot
• Documents endpoints for retrieving user profiles, managers, direct reports, and user lists
• Describes OData query capabilities including $select, $filter, $expand, $orderby, and $search

This new documentation introduces the Work IQ Teams MCP server (preview), detailing a comprehensive set of API endpoints for managing Microsoft Teams chats, channels, messages, members, and teams. It outlines supported operations, required parameters, error handling, and key features, enabling advanced automation and integration scenarios for Teams via Copilot.

• Introduces MCP server endpoints for Teams chat, channel, and team management
• Details CRUD operations for chats, messages, channels, and teams
• Describes support for private, shared, and standard channels

This new documentation introduces the Work IQ SharePoint reference (preview), detailing a comprehensive set of Copilot capabilities for managing SharePoint sites, document libraries, files, folders, lists, and columns. It outlines available tools, supported operations, and key limitations, enabling early adopters to understand and test Copilot's SharePoint integration.

• Provides a full reference for Copilot's SharePoint integration via the Work IQ server
• Details tools for site, document library, file, folder, list, and column management
• Describes file size limits (≤5 MB) and asynchronous operation handling

This new documentation introduces the Work IQ OneDrive reference for Copilot, detailing a preview feature that enables Copilot to manage files and folders in a user's personal OneDrive. It outlines a comprehensive set of file, folder, sharing, and compliance operations available via the MCP server, with clear limitations and parameters.

• Introduces Copilot's ability to manage personal OneDrive files and folders (≤5MB)
• Details available tools for file/folder CRUD, sharing, sensitivity labeling, and asynchronous operations
• Specifies security, compliance, and concurrency controls (e.g., eTags, sensitivity labels)

This new documentation introduces the Microsoft MCP management MCP server reference (preview), detailing a comprehensive set of management and tool creation APIs for MCP servers. It outlines capabilities for managing MCP servers, tools, connectors, custom APIs, SDK messages, remote APIs, Graph APIs, and Dataverse environments, enabling advanced extensibility and integration scenarios for Copilot and related services.

• Introduces APIs for creating, updating, deleting, and blocking MCP servers
• Documents management of tools within MCP servers, including creation via custom APIs, SDK messages, remote APIs, connectors, and Graph APIs
• Details connector and Graph API integration for tool creation

This new documentation introduces the Work IQ Mail (preview) Copilot capability, detailing a set of tools for programmatic email management via Microsoft Graph. It covers creating, sending, searching, updating, and deleting emails, as well as advanced features like HTML formatting, KQL-style search, and concurrency control.

• New Copilot server 'Work IQ Mail' for Microsoft Graph mail operations
• Supports creating, sending, updating, deleting, and searching emails
• Enables HTML and plain text email composition with rich formatting

This new documentation introduces the Dataverse MCP Server (preview), detailing a suite of tools that enable Copilot and AI agents to create, read, update, delete, and search Dataverse data and schemas programmatically. It outlines capabilities for data management, schema manipulation, and integration with Copilot scenarios, which is significant for developers and organizations leveraging Dataverse with AI.

• Introduces a new MCP Server for Dataverse with a comprehensive set of data and schema management tools
• Details tools for creating, updating, deleting, and querying records and tables in Dataverse
• Describes integration points for Copilot and AI agents to interact with Dataverse data securely

This new page documents the Work IQ Copilot (preview), a tool for searching and discovering information across Microsoft 365 content when workload-specific tools are unavailable. It details capabilities, integration points, parameters, and use cases for leveraging Copilot to retrieve organizational knowledge.

• Introduces Work IQ Copilot as a fallback and primary search tool for Microsoft 365 content
• Describes Copilot Chat tool for querying documents, emails, Teams chats, and more
• Lists required and optional parameters for invoking the tool

This new page introduces and documents the certification process for Model Context Protocol (MCP) servers, which enable integration of external tools and actions with Microsoft 365 Copilot and other AI experiences. It details prerequisites, packaging requirements, the certification workflow, and post-certification responsibilities for publishers.

• Defines MCP servers as services providing tools/actions for Copilot and AI experiences
• Explains that MCP server certification is handled via the Microsoft connector certification program
• Lists prerequisites including publisher eligibility, authentication support, and packaging requirements

This new documentation introduces the Work IQ Calendar MCP server and its preview API tools, enabling Copilot to perform advanced calendar operations such as event creation, management, scheduling, attendee handling, and online meeting integration. It details endpoints, parameters, and supported features, providing early access for feedback.

• New MCP server for calendar operations (Work IQ Calendar)
• APIs for creating, updating, deleting, accepting, declining, and canceling events
• Support for recurring events and online meetings (Teams, Skype)

This new page introduces Copilot Credits, a metered currency for Microsoft Copilot consumption, and details what they are, what they cover, and the available purchase options. It provides essential information for organizations planning to use Copilot workloads without traditional licensing.

• Defines Copilot Credits as a unified consumption unit for various Copilot workloads
• Explains coverage for Copilot Chat, Copilot Studio agents, Dynamics 365 agents, and Work IQ APIs
• Describes three purchase models: Pay-as-you-go, Capacity Packs, and Pre-Purchase Plan

This new page documents Copilot Studio's drill-down analytics features, enabling users to explore detailed user questions, session data, and feedback for deeper insights into agent performance and user satisfaction. It outlines how to access, filter, and analyze granular data on questions, sessions, and user reactions, which is critical for identifying knowledge gaps and improving agent quality.

• Introduces drill-down analytics for user questions, sessions, and feedback in Copilot Studio
• Details requirements and steps to access question and session lists, including security roles
• Explains available filters and downloadable data for granular analysis

The May 2026 update introduces several major new Copilot Studio capabilities, including general availability of computer use automation, new workflow nodes for AI and Microsoft 365 Copilot integration, consent-based voice recording, agent inventory auditing, asynchronous agent flows, and new AI model options. These features significantly expand automation, compliance, and integration scenarios for Copilot Studio users.

• Computer use automation for web and desktop apps is now generally available
• New prompt and Microsoft 365 Copilot nodes enable advanced workflow automation
• Consent-based recording for voice agents with compliance controls

A new prerequisite was added stating that, starting June 16, 2026, access to the Work IQ API will require consumption-based billing using Copilot Credits. This introduces a new licensing and billing model for using Work IQ capabilities in Copilot Studio.

• Announcement of Copilot Credits as required currency for Work IQ API usage starting June 16, 2026
• Introduction of pay-as-you-go and prepurchase plans for Copilot Credits

The documentation now includes support for DOCX files and experimental support for XLSX and PPTX file types for file uploads in Copilot agents. File size and character limits have also been clarified, and the list of supported file types has been updated.

• DOCX files are now officially supported for upload.
• Experimental support for XLSX and PPTX files is available upon request.
• PDF is no longer listed as a supported file type; DOCX is added.

The documentation now specifies that real-time voice agents can use either the GPT-Realtime model (hosted in North America) or the GPT-Realtime-Mini model (hosted in Australia), expanding regional support and clarifying cross-geo processing requirements. This update is significant for customers in Australia and those outside North America, as it introduces a new hosting region and model option.

• Added support for GPT-Realtime-Mini model hosted in Australia
• Clarified regional prerequisites and cross-geo processing rules
• Updated timeline for regional hosting to June 2026

The documentation now specifies that users can select between two real-time voice models, GPT-Realtime and GPT-Realtime-Mini, when configuring a real-time voice agent. Additionally, there are clarifications and minor rewordings in best practices for topic usage.

• Added option to select between GPT-Realtime and GPT-Realtime-Mini models for real-time voice agents
• Clarified that model selection is a one-time setting
• Minor clarifications in best practices for Conversation Start and On Error topics

A new section on topic usage analytics has been added, providing detailed insights into how individual topics perform within Copilot Studio agents. This enables users to analyze topic-level outcomes, usage trends, and customer satisfaction, which can help prioritize improvements.

• Added 'Topic usage analytics' section describing how to analyze individual topic performance
• Explains metrics such as topic outcomes (Escalated, Resolved, Abandoned), usage trends, and satisfaction scores
• Clarifies that topic analytics is only available for agents in classic mode, not for those using generative orchestration

The Analytics documentation for Copilot Studio has been restructured to introduce new sections for Custom metrics and Use, clarify the core analytics areas, and reorganize existing content. Notably, users can now define up to three custom business-specific metrics, and the Use section provides detailed insights into answer quality and operational performance.

• Added 'Custom metrics' section allowing up to three user-defined business metrics for agent analytics.
• Introduced 'Use' section to analyze answer rate, quality, tool use, and knowledge source reliability.
• Clarified and consolidated core analytics areas from seven to four, with updated descriptions.

Page removed or relocated. The documentation page explaining how to analyze active user metrics in Copilot Studio Analytics has been removed. This leaves a gap in guidance for understanding and interpreting DAU/MAU metrics, their calculation, and prerequisites for usage.

• Loss of explanation for DAU and MAU definitions and calculation
• Missing instructions on viewing and interpreting active user trends
• No longer documents the authentication requirement for active user metrics

Page removed or relocated. The documentation page explaining how to configure a custom privacy statement URL for Copilot Studio agents has been removed. This leaves a gap in guidance for admins on how to ensure users see their organization's privacy or legal statement when interacting with Copilot agents across various channels.

• Loss of step-by-step instructions for configuring a privacy statement URL in Power Platform admin center
• No longer documents the automatic disclosure message behavior across Copilot Studio, Teams, Microsoft 365, and SharePoint
• Missing details on scope, exemption from URL-blocking, and compliance requirements for privacy disclosures

The documentation now clarifies that standalone computer use tools can only be added to agent flows, not to workflows. This distinction is important for users planning automation architectures in Copilot Studio.

• Explicit statement that standalone computer use tools are not yet supported in workflows
• Section heading updated for clarity

The documentation clarifies preview feature limitations, warns that tool names and parameters may change, and updates the server endpoint path for the Windows 365 for Agents MCP server. These changes help users avoid hard-coded dependencies and ensure they use the correct server endpoint.

• Added warning that preview MCP tool names and parameters may change; users should avoid hard-coded dependencies.
• Updated the server endpoint path from 'servers/mcp_W365AServer' to 'servers/mcp_W365ComputerUse'.
• Clarified preview feature terms and usage restrictions.

Page removed or relocated. The documentation page explaining how to analyze topic usage in Copilot Studio has been removed. It is unclear if this reflects a feature deprecation or a documentation restructure, but users now lack official guidance on accessing and interpreting topic analytics in classic mode.

• Loss of instructions for accessing the topic analytics pane
• No longer documents the metrics and charts available for topic analysis
• Uncertainty whether topic analytics for classic agents is still supported

This new page introduces Agent 365, a centralized control plane for managing AI agents in enterprise environments. It details how Agent 365 provides identity, observability, governance, security, and lifecycle management for agents, addressing challenges like agent sprawl and fragmented oversight.

• Introduction of Agent 365 as a unified management solution for enterprise AI agents
• Explanation of agent identity management via Microsoft Entra Agent ID
• Description of observability features for tracking agent actions and access

This new page introduces Microsoft Entra's support for Agent 365, detailing how agent identities are managed, governed, and secured using Entra Agent ID. It outlines foundational identity constructs, governance, and developer integration, enabling organizations to manage agents at scale with the same rigor as users and applications.

• Introduction of Microsoft Entra Agent ID for agent identity management
• Agent identity blueprints for scalable governance and policy enforcement
• Support for both user-context and autonomous agent identity models

This new page introduces and details how Microsoft Defender integrates with Agent 365 to provide real-time security monitoring, proactive protection, and incident response for agents. It highlights that agents are now observable and manageable security entities within the Microsoft security ecosystem, aligning their protection with existing enterprise practices.

• Describes Microsoft Defender's real-time monitoring and threat detection for Agent 365 agents
• Explains integration of agent activity into Microsoft security tools (Entra, Purview, Intune)
• Details proactive and reactive security measures, including runtime protection and incident investigation for agents

Several Agent 365 CLI commands for MCP server management (approve, block, package-mcp-server) have been removed. Approval and blocking are now handled in the Microsoft admin center, and server packaging for Admin Center submission is no longer supported via CLI.

• Removed CLI commands: develop-mcp approve, develop-mcp block, develop-mcp package-mcp-server
• Server approval and blocking now performed in Microsoft admin center, not CLI
• CLI no longer produces MCP server packages for Admin Center submission

Page removed or relocated. The documentation for the Windows 365 for Agents MCP server (preview) has been removed. This page described a wide range of automation and control capabilities for Windows 365 Cloud PCs, including desktop, browser, and accessibility tools, as well as usage guidance and tool reference. The removal means detailed information about these preview features is no longer available, which may impact users and developers relying on this documentation.

• Loss of detailed reference for all MCP server tools (desktop, browser, accessibility, shell, Python execution, etc.)
• No longer documents how to automate Windows 365 Cloud PCs via MCP server
• Missing guidance on common automation use cases and tool parameters

Page removed or relocated. The documentation page detailing the Work IQ Teams reference (preview) has been removed. This page described the MCP server endpoints and operations for managing Teams chats, channels, messages, and members via Copilot, including supported API operations and parameters. Its removal means that detailed reference information for these Copilot Teams integration capabilities is no longer available, which could impact developers and admins relying on this preview feature.

• Loss of endpoint documentation for Teams chat, channel, and team management via Copilot MCP server
• Missing details on required/optional parameters, error codes, and supported operations for Teams automation
• No longer provides reference for preview features enabling Copilot to manage Teams resources

Page removed or relocated. The documentation for the deprecated SharePoint lists tools MCP server has been removed. This page detailed all available actions, parameters, and features for managing SharePoint lists, columns, items, and site discovery via the MCP server, which is deprecated as of March 13, 2026. Its removal means users lose reference material for existing connections and migration guidance.

• Loss of detailed reference for all SharePoint lists MCP server tools and their parameters
• No longer documents how to create, edit, delete, and list SharePoint lists, columns, and items via MCP
• Removes guidance on OData query support and site discovery/search operations

Page removed or relocated. The documentation page detailing the Work IQ SharePoint reference (preview) has been removed. This page described the Copilot integration for managing SharePoint sites, document libraries, files, folders, lists, columns, and permissions, including supported operations and parameters. Its removal means there is no longer official documentation for these Copilot SharePoint capabilities, which may impact preview users and developers.

• Loss of documentation for Copilot's SharePoint management capabilities (sites, files, folders, lists, columns, permissions)
• No longer available: reference for tool names, parameters, and usage examples for SharePoint operations via Copilot
• Uncertainty whether the feature is deprecated or documentation was relocated

Page removed or relocated. The documentation page detailing the Work IQ OneDrive reference (preview) has been removed. This page described a set of Copilot-integrated tools for managing, searching, and sharing files and folders in a user's personal OneDrive, including file operations, folder management, sharing, and security features. The removal means that detailed guidance on these preview capabilities is no longer available, and it is unclear whether the feature is deprecated or the documentation has been relocated.

• Loss of documentation for Copilot's OneDrive management tools (file/folder CRUD, sharing, sensitivity labels, etc.)
• No longer provides reference for tool parameters, usage, and operational limits (e.g., 5MB file size cap, eTag concurrency)
• Uncertainty about the current status or location of these preview features

Page removed or relocated. The documentation for the deprecated Microsoft SharePoint and OneDrive MCP server has been removed. This page described the capabilities, tools, and usage details for the legacy MCP server, which will be unsupported for new connections after March 13, 2026. Users now lack official reference for this server's features and migration guidance.

• Loss of detailed reference for all available tools (e.g., createFolder, findSite, shareFileOrFolder)
• No longer documents folder and file management, sharing, and compliance features for the deprecated MCP server
• Deprecation timeline and migration instructions are now missing

Page removed or relocated. The documentation page detailing the Microsoft MCP management MCP server reference (preview) has been removed. This page described a wide range of management and tool creation capabilities for MCP servers, including APIs for server and tool lifecycle management, integration with custom APIs, connectors, Graph, and Dataverse environments. The removal means users no longer have official documentation for these preview features, which could impact development and integration efforts.

• Loss of reference documentation for managing MCP servers and tools via API
• No longer documents how to create, update, delete, or block MCP servers
• Removal of guidance for integrating custom APIs, connectors, Graph, and Dataverse tools

Page removed or relocated. The documentation for the Work IQ Calendar reference (preview) has been removed. This page described the MCP server and API tools for Copilot calendar operations, including event management, scheduling, attendee management, recurrence, and online meeting integration. The removal means detailed reference information for these Copilot calendar capabilities is no longer available, which may impact developers and admins relying on this preview feature.

• Loss of documentation for all MCP calendar tools APIs (create, update, delete, accept, decline, cancel, list, get events, scheduling, etc.)
• No longer available: details on parameters, usage, and supported features for Copilot calendar operations
• Missing reference for integration with Teams, Skype, and advanced scheduling/recurrence features

This new page introduces and explains how Microsoft Purview supports Agent 365 by extending existing data security, compliance, and observability capabilities to autonomous agents. It details how sensitivity labels, DLP, audit, and compliance policies apply to agent activity, ensuring consistent governance for AI-driven operations.

• Describes how Purview's sensitivity labels, DLP, and retention rules apply to agents as they do to users
• Explains that agent activity is captured in audit and eDiscovery logs for compliance
• Clarifies that Agent 365 leverages Purview to provide consistent data security and compliance for agents

This new page introduces and explains how Microsoft Intune supports Agent 365, enabling unified governance of AI agents by integrating device compliance, conditional access, and app protection policies. It highlights how Intune helps organizations manage and secure agent operations across various device types, reducing risks from unsanctioned AI tools and ensuring endpoint health is part of agent governance.

• Describes unified governance for agents and devices using Intune
• Explains how Intune policies and conditional access apply to Agent 365
• Details how Intune helps prevent shadow AI risks by controlling agent access and data flow

This new page introduces foundational governance capabilities for AI agents using Agent 365, including agent registry, ownership assignment, and observability. It outlines how these features enable organizations to scale agent adoption securely and efficiently without hindering innovation.

• Introduces agent registry with Microsoft Entra-backed Agent IDs for enterprise-level management
• Describes explicit ownership assignment for accountability of agent behavior and lifecycle
• Details built-in observability and telemetry via Agent 365 and OpenTelemetry conventions

The catalog of available Work IQ MCP servers has been expanded to include Windows 365 agents, Fabric IQ Ontology, and the Microsoft MCP Management server, enabling new management and knowledge graph capabilities. The documentation now also highlights the ability to build custom MCP servers. These additions broaden the scope of what agents can automate and manage within Microsoft 365 and beyond.

• Added Windows 365 agents for Cloud PC management
• Added Fabric IQ Ontology for querying organizational knowledge graphs
• Added Microsoft MCP Management server for building custom MCP servers

The documentation adds significant new browser automation capabilities for the Windows 365 for Agents MCP server, including configurable navigation waits, consolidated page state retrieval, advanced DOM and accessibility-based interactions, and multi-tab management enhancements. These changes expand the automation and control options available to Copilot integrations.

• Browser automation now supports configurable wait conditions on navigation (load, networkidle0, networkidle2)
• Consolidated page state retrieval allows fetching URL, title, DOM, screenshot, and tab list in a single call
• DOM-level automation expanded to include form fill, drag, and <select> option selection by CSS selector

Page removed or relocated. The documentation page for the Work IQ Word reference (preview) has been removed, resulting in the loss of detailed information about preview capabilities for programmatically creating, retrieving, and commenting on Word documents via the Work IQ MCP server. It is unclear if the feature is deprecated or if the documentation has been relocated.

• Loss of documentation for creating new Word documents in OneDrive via API
• Loss of details on retrieving Word document content and metadata using sharing URLs
• Loss of guidance for adding and replying to comments in Word documents programmatically

Page removed or relocated. The documentation page detailing the Work IQ Copilot (preview) feature has been removed. This eliminates reference material for a tool that enables broad search and discovery across Microsoft 365 content, including its integration, parameters, and use cases.

• Loss of overview and usage guidance for Work IQ Copilot's search capabilities across Microsoft 365
• No longer documents Copilot Chat tool parameters and conversation management features
• Removes examples and scenarios for using Work IQ Copilot as a fallback or primary search tool

Page removed or relocated. The documentation for the Work IQ User reference (preview) has been removed, resulting in the loss of detailed guidance on using MCP server endpoints for user and organizational profile queries within Copilot. It is unclear if the feature itself is deprecated or if the documentation was relocated, but users now lack official reference material for these preview capabilities.

• Loss of endpoint documentation for user profile, manager, and direct report queries
• No longer provides guidance on OData query options and organizational hierarchy navigation
• Missing details on preview feature limitations and usage notes

Page removed or relocated. The documentation page detailing the Work IQ Mail Copilot preview feature and its mail management capabilities has been removed. This eliminates reference information for a set of Copilot-enabled email tools, but it is unclear if the feature itself is deprecated or if the documentation has been relocated.

• Loss of documentation for Copilot's Work IQ Mail tools, including draft creation, sending, replying, searching, and managing emails via Microsoft Graph.
• Missing reference for required and optional parameters for each mail operation.
• No longer provides details on preview feature limitations, supported operations, and API usage notes.

Page removed or relocated. The documentation for the Fabric IQ Ontology MCP (preview) endpoints and integration with Copilot Studio has been removed. This eliminates official guidance on connecting Copilot agents to Fabric Ontology data using MCP, which may impact users seeking to leverage this preview capability.

• Loss of documentation for MCP server endpoints enabling Copilot and AI assistants to query Fabric Ontology.
• Removal of tool definitions for 'list_ontology_entity_types' and 'search_ontology' within the MCP server.
• No longer available: step-by-step instructions for integrating Fabric IQ Ontology with Copilot Studio agents.

Page removed or relocated. The documentation for the Dataverse MCP Server (preview), which detailed Copilot-accessible tools for Dataverse data operations, has been removed. This leaves a gap in guidance for using Copilot or AI agents to interact programmatically with Dataverse tables and records.

• Loss of reference for tools enabling Copilot to create, read, update, delete, and search Dataverse data
• No longer documents how Copilot or AI agents can perform schema management or data lifecycle operations in Dataverse
• Uncertainty whether the feature is deprecated or documentation was relocated

Page removed or relocated. The documentation detailing the certification process for Model Context Protocol (MCP) servers, which enable integration with Microsoft 365 Copilot and other AI experiences, has been removed. This leaves a gap in official guidance for third-party developers seeking to certify MCP servers for Copilot integration, though it is unclear if the process itself is deprecated or if the documentation has been relocated.

• Loss of step-by-step guidance for certifying MCP servers for Copilot and Microsoft 365 integration
• Missing prerequisites, packaging, and submission requirements for MCP server certification
• No longer available: details on post-certification responsibilities and compliance monitoring

The command reference table now includes the 'develop-mcp list-environments' command, which allows users to list all Dataverse environments available for MCP server management. This expands the documented capabilities of the Agent 365 CLI.

• Added 'develop-mcp list-environments' command to the documented command set

The documentation now clarifies that both registering and deleting agent messaging endpoints are performed via Teams Graph for all agents, and introduces a new option to specify the messaging endpoint URL directly on the command line without editing the config file.

• Removed distinction between M365 and non-M365 agents for endpoint registration and deletion; all use Teams Graph now
• Added support for specifying the messaging endpoint URL directly with --messaging-endpoint in the setup command

The prerequisites for Microsoft Agent 365 have been clarified, now recommending Microsoft E5 as the best pre-requisite instead of Entra P1/P2/Suite and Purview Data Loss Prevention. This may affect deployment planning for organizations.

• Changed recommended prerequisite from Entra P1/P2/Suite and Purview DLP to Microsoft E5
• Removed mention of Entra and Purview DLP as recommended for full benefits

This new page documents the ability to discover, view, add, and share agents created in SharePoint directly from the Teams app store. It details how users can find personalized lists of agents, view their details, and integrate them into chats or channels, significantly enhancing integration between SharePoint agents and Teams.

• Agents created in SharePoint are now discoverable in the Teams app store
• Users see a personalized list of recently used agents
• Agents can be added directly to chats, channels, or meetings from Teams

The documentation now clarifies that users can choose from multiple AI image generation models (Microsoft AI, Anthropic, OpenAI, Flux.2) when creating images with Copilot in PowerPoint, expanding the flexibility and potential quality of generated images.

• Users can now select the image generation model (Microsoft AI, Anthropic, OpenAI, Flux.2) when creating images with Copilot
• Previously, only DALL-E 3 was mentioned as the image generator

The documentation now clarifies that users can select from a broader range of AI models—including Microsoft AI and Black Forest Labs' Flux.2—in addition to Anthropic's Claude and OpenAI's ChatGPT, to power Copilot in PowerPoint. This expands the customization and capability options for users. Minor clarifications and rewording were also made regarding Brand kit setup and the availability of Copilot skills.

• Expanded list of selectable AI models for Copilot in PowerPoint, now including Microsoft AI and Flux.2
• Clarified that users can choose specific AI reasoning and image models for different tasks
• Minor rewording and clarification in Brand kit setup instructions

The documentation adds and summarizes several significant new features and breaking changes for Copilot-related Python SDKs, including a major upgrade to the GitHub Copilot SDK, new tool exposure and skills discovery enhancements, and support for additional providers and structured output. These changes impact how developers integrate and extend Copilot capabilities in Python environments.

• GitHub Copilot SDK upgraded to v1.0.0 with breaking changes: removed SubprocessConfig, changed import paths, and updated permission handler types
• Progressive tool exposure now available via FunctionInvocationContext
• MCP-based skills discovery introduced with McpSkillsSource

A new environment variable, GITHUB_COPILOT_BASE_DIRECTORY, has been added to allow users to specify the directory for CLI session state and configuration, providing more flexibility in agent configuration.

• Added GITHUB_COPILOT_BASE_DIRECTORY environment variable for specifying CLI session state/config directory

The introduction now details specific service limits for Copilot Call Delegation during the Frontier program, including call length, monthly tenant usage, and concurrent call processing. It also clarifies that these limits may affect call handling and are subject to change before general availability.

• Added call length limit of 4 minutes per call
• Added monthly usage limit of 15,000 minutes per tenant
• Added AI model processing limits based on concurrent calls