import { MicrosoftCopilotIcon } from '@/components/icons'
import type { CompetitorProfile } from '@/lib/compare/data/types'

/** Researched and cross-verified against live vendor sources on 2026-07-02. */
export const microsoftCopilotProfile: CompetitorProfile = {
  id: 'microsoft-copilot',
  name: 'Microsoft Copilot Studio',
  website: 'https://www.microsoft.com/en-us/microsoft-copilot-studio',
  brand: {
    icon: MicrosoftCopilotIcon,
    selfFramed: true,
    colors: ['#0736c4', '#8c48ff', '#00e5cc'],
    source: 'Official brand guidelines',
    asOf: '2026-07-02',
  },
  oneLiner:
    'Microsoft Copilot Studio is a low-code Microsoft tool for building, testing, and publishing conversational and autonomous AI agents with topics or LLM-driven generative orchestration, connectors, agent flows, and Dataverse-grounded knowledge.',
  standoutFeatures: [
    {
      title: 'Broad, independently audited compliance certification list',
      description:
        'Copilot Studio is certified under HIPAA (BAA), HITRUST CSF, FedRAMP, SOC, multiple ISO standards (9001, 20000-1, 22301, 27001, 27017, 27018, 27701), PCI DSS, CSA STAR, UK G-Cloud, Singapore MTCS Level 3, Korea K-ISMS, and Spain ENS, each with an audit report on the Microsoft Service Trust Portal.',
      shortDescription:
        'HIPAA, FedRAMP, SOC, multiple ISO standards, PCI DSS, and more, each audited.',
      source: {
        url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-certification',
        label: 'Review ISO, SOC, and HIPAA compliance - Microsoft Copilot Studio | Microsoft Learn',
        asOf: '2026-07-02',
      },
    },
    {
      title: 'Generative orchestration picks topics, tools, and knowledge dynamically',
      description:
        "Generative orchestration replaces fixed decision-tree topic flows with an LLM-driven planning layer. It interprets user intent, selects from an agent's topics, tools, knowledge sources, and child agents at runtime, and executes multistep plans, instead of requiring every path hand-authored with trigger phrases in advance.",
      shortDescription:
        'An LLM planning layer selects topics/tools/knowledge at runtime, not a fixed script.',
      source: {
        url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/generative-orchestration',
        label:
          'Apply generative orchestration capabilities - Microsoft Copilot Studio | Microsoft Learn',
        asOf: '2026-07-02',
      },
    },
    {
      title: 'Automatic fallback to a default model if a selected model is disabled',
      description:
        "If an admin disables a non-default model tenant-wide, Copilot Studio automatically falls back to the default OpenAI model rather than failing the request. Admins choose among OpenAI GPT models, Anthropic Claude Sonnet 4 and Opus 4.1, any model in the Azure AI Model Catalog, or a bring-your-own-model connection to an Azure AI Foundry deployment, but Sim's own model-fallback only retries the same model with hosted keys, not a cross-model fallback like this.",
      shortDescription:
        'Falls back to the default OpenAI model automatically if a selected model is disabled.',
      source: {
        url: 'https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/anthropic-joins-the-multi-model-lineup-in-microsoft-copilot-studio/',
        label: 'Anthropic joins the multi-model lineup in Microsoft Copilot Studio',
        asOf: '2026-07-02',
      },
    },
    {
      title: 'Deep, per-session orchestration traces',
      description:
        "Session tracking in the Monitor tab (part of Copilot Studio's new agent experience, currently in preview) lists each conversation session's status, duration, message count, and which tools were used, and selecting a session opens its full transcript with tool-invocation and knowledge-source usage detail.",
      shortDescription:
        'Per-session transcripts in the Monitor tab cover status, tools, and knowledge used.',
      source: {
        url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/agents-experience/analytics-overview',
        label: 'Monitor an agent overview (preview) - Microsoft Copilot Studio | Microsoft Learn',
        asOf: '2026-07-08',
      },
    },
  ],
  limitations: [
    {
      title: 'Consumption pricing carries overage and cost-forecasting complexity',
      description:
        'Usage is billed in Copilot Credits across many separately metered feature rates (classic answer, generative answer, agent action, tenant graph grounding, agent flow actions per 100, three tiers of AI tools, three tiers of voice), and reasoning models add a second, separate premium-token charge on top of the base rate. Microsoft publishes a dedicated usage estimator tool, since forecasting cost across these dimensions manually is difficult.',
      shortDescription:
        'Costs span many separately metered credit rates, needing a dedicated estimator tool.',
      source: {
        url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/requirements-messages-management',
        label: 'Billing rates and management - Microsoft Copilot Studio | Microsoft Learn',
        asOf: '2026-07-02',
      },
    },
    {
      title:
        'Full ALM tooling (pipelines, environment variables) is gated to solution-aware agents',
      description:
        'Version history, environment promotion via pipelines, and environment variables only apply to agents built inside a Dataverse solution. Agents created outside a solution, the common default for individual makers experimenting in Copilot Studio, get none of this ALM tooling.',
      shortDescription:
        'Pipelines and environment-variable promotion require building inside a solution.',
      source: {
        url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/alm',
        label:
          'Establish an Application Lifecycle Management (ALM) strategy - Microsoft Copilot Studio | Microsoft Learn',
        asOf: '2026-07-02',
      },
    },
    {
      title: 'Session transcript detail defaults to a 29-day window',
      description:
        "An agent's per-session transcripts, showing which topic fired, tools called, and knowledge consulted, are downloadable from the Analytics area for roughly the last 29 days by default. Retaining that detail longer requires a separate export pipeline, not a built-in retention setting.",
      shortDescription:
        'Detailed session transcripts default to ~29 days; longer retention needs a manual export.',
      source: {
        url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/analytics-transcripts-studio',
        label:
          'Understand downloaded session data from Copilot Studio - Microsoft Copilot Studio | Microsoft Learn',
        asOf: '2026-07-08',
      },
    },
    {
      title: 'No self-hosting; a Microsoft-operated cloud service only',
      description:
        "Copilot Studio's authoring, orchestration, and runtime are Microsoft-operated multi-tenant (or government-cloud) services. There is no on-premises deployment option for the core agent-building and conversation-orchestration engine, unlike products that offer a self-hosted or air-gapped runtime.",
      shortDescription:
        'No on-premises option; agents run only on Microsoft-operated cloud infrastructure.',
      source: {
        url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/requirements-licensing-gcc',
        label: 'US Government customers - Microsoft Copilot Studio | Microsoft Learn',
        asOf: '2026-07-04',
      },
    },
    {
      title: 'No native white-labeling of the core builder or default chat UI',
      description:
        'Agent-level branding is limited to setting a name, icon, and accent color, or hosting a fully custom canvas web app for the chat window. There is no option to remove Microsoft/Copilot Studio branding from the authoring canvas or the default published chat surface, unlike a dedicated white-label offering.',
      shortDescription:
        'Branding is limited to name/icon/color or a custom-built chat canvas, not full white-label.',
      source: {
        url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/customize-default-canvas',
        label:
          'Customize the look and feel of an agent - Microsoft Copilot Studio | Microsoft Learn',
        asOf: '2026-07-02',
      },
    },
  ],
  facts: {
    platform: {
      builderType: {
        value:
          'Low-code conversational/agent builder with a topics-based (trigger phrase) authoring mode, an LLM-driven generative orchestration mode, natural-language agent creation, and separate agent flows for deterministic step sequences',
        detail:
          'Classic authoring uses topics triggered by phrases. Generative orchestration lets the agent dynamically select topics, tools, and knowledge based on a description of each. Agent flows, built on the Power Automate flow engine, handle deterministic, step-by-step logic invoked as agent tools.',
        shortValue: 'Topics or generative orchestration, plus deterministic agent flows',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/fundamentals-what-is-copilot-studio',
            label: 'Overview - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/generative-orchestration',
            label:
              'Apply generative orchestration capabilities - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/flows-overview',
            label: 'Agent flows overview - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      learningCurve: {
        value:
          'Low for a single-topic agent answering from one knowledge source; steep for generative orchestration instruction design, solution-based ALM, and Dataverse security modeling at production scale',
        detail:
          'Basic agent creation is accessible to non-developer makers via natural language and templates, but generative-orchestration instructions and solution/environment/pipeline structuring both require deliberate design work, per dedicated Microsoft guidance on each.',
        shortValue: 'Easy for a single simple agent, steep for orchestration and ALM at scale',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/generative-mode-guidance',
            label:
              'Configure high-quality instructions for generative orchestration - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/alm',
            label:
              'Establish an Application Lifecycle Management (ALM) strategy - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      selfHostOption: {
        value:
          'No: Copilot Studio has no self-hosted deployment of its authoring or orchestration/runtime engine; it runs only as a Microsoft-operated cloud service (commercial or government cloud)',
        detail:
          "Copilot Studio is an Online Service, defined as such in the Online Services Terms, running only across Commercial, GCC, GCC High, and DoD environments, all Microsoft-operated. The US Government plans documentation describes the service as running 'in a manner consistent with a multitenant, public cloud deployment model,' with no on-premises or customer-hosted runtime offered.",
        shortValue: 'No, Microsoft-operated cloud service only',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-certification',
            label:
              'Review ISO, SOC, and HIPAA compliance - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-04',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/requirements-licensing-gcc',
            label: 'US Government customers - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-04',
          },
        ],
      },
      deploymentOptions: {
        value:
          'Commercial multi-tenant cloud, plus Office 365 GCC, GCC High, and DoD sovereign/government cloud environments',
        detail:
          "Copilot Studio's US Government customers documentation describes GCC as compliant with FedRAMP High and available since December 2019, and GCC High as available to eligible customers since February 2022 for DISA SRG IL4-aligned workloads, all still running on Microsoft-operated (not customer-operated) infrastructure.",
        shortValue: 'Commercial cloud plus GCC/GCC High/DoD government clouds',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/requirements-licensing-gcc',
            label: 'US Government customers - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-04',
          },
        ],
      },
      templates: {
        value:
          'Yes: an Agent Library of ready-to-use, pre-built agent templates (e.g. My Company Policy, Request Tracker, Know Your Customer, Plan My Day, Executive Brief) with preconfigured instructions, actions, topics, and starter knowledge, deployable into an environment and then customized',
        detail:
          'Templates are distributed as a visual, guided-deployment catalog on Microsoft Marketplace and as raw solution files on GitHub.',
        shortValue: 'Yes, Agent Library of pre-built, customizable templates',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/agent-library-overview',
            label:
              'Configure and deploy agents from the Agent Library - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/template-fundamentals',
            label:
              'Create a custom agent from a template - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      license: {
        value: 'Proprietary, commercial SaaS, not open source',
        detail:
          'Copilot Studio is a licensed Microsoft commercial cloud product billed via Copilot Credits (prepaid packs, pay-as-you-go, or annual commitment); no open-source licensing model exists for the product.',
        shortValue: 'Proprietary commercial SaaS',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/requirements-messages-management',
            label: 'Billing rates and management - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      environmentPromotion: {
        value:
          'Yes: agents built inside a Dataverse solution promote dev to test to production via exported/imported managed and unmanaged solutions, Power Platform Pipelines for automated promotion, and environment variables for per-environment config, the same ALM model Power Automate uses',
        detail:
          'Solutions are the transport container across environments, with pipelines or Azure DevOps/GitHub Actions automating the import chain and Git integration available for source control.',
        shortValue: 'Dataverse solutions plus Pipelines for dev/test/prod',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-solutions-overview',
            label:
              'Create and manage custom solutions - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/alm',
            label:
              'Establish an Application Lifecycle Management (ALM) strategy - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      versionControlDepth: {
        value:
          'Solution-based promotion with Git integration and pipelines for solution-aware agents; no visual diff/compare view between two agent versions, and this ALM tooling does not apply to agents built outside a solution',
        detail:
          'Git integration lets a solution connect to a repository for version control and collaboration, and pipelines automate deployment between environments, but there is no dedicated side-by-side version-diff UI for an individual agent.',
        shortValue: 'Git-backed solution promotion, no visual diff view found',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/alm',
            label:
              'Establish an Application Lifecycle Management (ALM) strategy - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      realtimeCollaboration: {
        value:
          "No: Copilot Studio agents can be shared with co-owners and accessed by multiple makers, but there is no live, concurrent multi-user editing of the same agent with visible cursors and synced changes. Microsoft's live coauthoring feature (visible cursors, simultaneous editing) exists for Power Apps Studio canvas apps, a separate product, not for the Copilot Studio agent designer.",
        detail:
          'Sharing an agent with other users grants asynchronous edit access, not simultaneous live co-editing. Cursor-level real-time collaboration does not exist in the Copilot Studio designer itself.',
        shortValue: 'No true live co-editing; only async sharing of an agent',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-share-bots',
            label: 'Share agents with other users - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      nativeFileStorage: {
        value:
          'No: Copilot Studio has no file storage system of its own (no folder hierarchy, link-based sharing, or recycle bin built into the product). Files used as knowledge sources or by connectors are stored in and managed through external services such as SharePoint, OneDrive, or Dataverse.',
        detail:
          'Knowledge sources reference SharePoint sites/document libraries or uploaded files stored in Dataverse. There is no first-party, user-facing file manager inside Copilot Studio itself.',
        shortValue: 'No native file store; relies on SharePoint/OneDrive/Dataverse',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-copilot-studio',
            label: 'Knowledge sources summary - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      dataTables: {
        value:
          'No: Copilot Studio does not offer a lightweight, spreadsheet-like data table with arrow-key navigation and copy-paste. Its native structured-data store is Microsoft Dataverse, a full relational database with tables, relationships, and business rules, used both as an agent knowledge source and, via dedicated bot/botcomponent/conversationtranscript tables, for storing conversation transcripts and custom analytics.',
        detail:
          "Dataverse is the same underlying data platform Power Automate uses, offering rich metadata/relationships plus business rules and workflows for data validation, not a simple spreadsheet-grid UI. Copilot Studio's custom-analytics guidance documents the ConversationTranscript, Copilot (bot), and Copilot component (botcomponent) tables it writes usage data into.",
        shortValue: 'Dataverse tables are a full DB, not a spreadsheet grid',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/power-apps/maker/data-platform/data-platform-intro',
            label: 'What is Microsoft Dataverse? - Power Apps | Microsoft Learn',
            asOf: '2026-07-08',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/custom-analytics-strategy',
            label:
              'Develop a custom analytics strategy - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      richTextEditor: {
        value:
          'No: Copilot Studio does not provide an inline rich-text/WYSIWYG document editor. Agent instructions, topic content, and Skill files (Markdown with YAML front matter) are authored as plain text or Markdown source, either inside the Copilot Studio UI or an external text editor, not through a WYSIWYG surface.',
        detail:
          'Skills are portable Markdown files a maker can author in Copilot Studio or any text editor, so authoring is raw-text/Markdown-based rather than WYSIWYG.',
        shortValue:
          'No WYSIWYG editor; instructions and Skills are authored as Markdown/plain text',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/agents-experience/skills-overview',
            label:
              'Skills overview for agents (preview) - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      subWorkflows: {
        value:
          'Yes: agent flows built in the flow designer include a built-in child-flow action, alongside loop and branch control structures, that calls another published flow as a step, waits for it to finish, and passes/receives data, distinct from an agent merely calling a flow as a tool at the conversation level',
        detail:
          'The agent flow designer lists control structures for looping and branching, data operations, date and time functions, and child flows as built-in action categories, the same child-flow composition model Power Automate uses for its underlying flow engine.',
        shortValue: 'Yes, a built-in child-flow action calls and awaits another flow',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/flows-overview',
            label: 'Agent flows overview - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      customBlocks: {
        value:
          "Partial: a published agent flow (with a 'When an agent calls the flow' trigger and a 'Respond to the agent' action) can be added by another maker as a tool, and a published Copilot Studio agent can be added by another maker as a connected-agent tool that always resolves to the source's latest published version — but discovery for both is a per-item share-and-pick model within one environment, not an admin-governed entry that automatically appears in a block toolbar for the whole org, and there is no publisher step to hand-pick and rename a curated subset of outputs the way Sim's Custom Blocks do.",
        detail:
          "Adding a flow as an agent-level tool lists qualifying, published flows in the 'Add tool > Flow' picker; Microsoft's docs don't specify how the tool's inputs are derived from the flow's trigger or confirm it always tracks the flow's latest published version. Connected agents are the better-documented half: a maker can turn on 'Let other agents connect to and use this one' on a published agent so any maker who owns it or has it shared with them can add it as a connected-agent tool in their own, separate agent, always using the latest published version after republish. Neither mechanism auto-populates a single, organization-wide block palette; each is discovered per-item through sharing/run permissions, and there's no dedicated step for the publisher to select and rename specific outputs to expose, unlike Sim's Custom Blocks.",
        shortValue:
          'Partial: published flows/agents become tools, but via per-item sharing, not an org-wide toolbar',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/flow-agent',
            label:
              'Add an agent flow as a tool to an agent - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/add-agent-copilot-studio-agent',
            label:
              'Connect to an existing Copilot Studio agent - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-add-other-agents',
            label: 'Add other agents overview - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
    },
    aiCapabilities: {
      multiLlmSupport: {
        value:
          'Yes: agents can use OpenAI GPT models, Anthropic Claude Sonnet 4 and Opus 4.1, any model in the Azure AI Model Catalog, or a bring-your-own Azure AI Foundry model deployment for individual prompts',
        detail:
          'Admins enable or restrict non-default models tenant-wide in the Microsoft 365 Admin Center. If a selected alternate model is disabled or unavailable, agents fall back automatically to the default OpenAI model. Google Gemini is not supported.',
        shortValue: 'OpenAI, Anthropic Claude, Azure AI Model Catalog, or bring-your-own model',
        confidence: 'verified',
        sources: [
          {
            url: 'https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/anthropic-joins-the-multi-model-lineup-in-microsoft-copilot-studio/',
            label: 'Anthropic joins the multi-model lineup in Microsoft Copilot Studio',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/bring-your-own-model-prompts',
            label:
              'Bring your own model for your prompts - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      agentReasoningBlocks: {
        value:
          "Yes: generative orchestration is an LLM-driven planning layer that selects among an agent's topics, tools, knowledge sources, and child agents at runtime, and an optional deep reasoning model (the Azure OpenAI o3 model) can be enabled for tasks requiring step-by-step logical analysis, distinct from a fixed decision-tree topic flow",
        detail:
          'Deep reasoning is regionally limited to the United States and the EU (excluding the UK) and trades response speed for accuracy on complex tasks. The agent decides when to apply it, or a maker can force it via a "reason" keyword in instructions.',
        shortValue: 'Generative orchestration planning plus an optional deep reasoning model',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/generative-orchestration',
            label:
              'Apply generative orchestration capabilities - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-reasoning-models',
            label:
              'Add a deep reasoning model for complex tasks (preview) - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      naturalLanguageBuilding: {
        value:
          "Yes: makers can describe an agent in plain language and Copilot Studio's AI-based authoring drafts topics, instructions, and structure from that description, distinct from manually assembling a topic tree",
        detail:
          "This is the product's core AI-based agent authoring capability, letting non-developers describe the desired behavior instead of hand-building every topic and trigger phrase.",
        shortValue: 'Describe an agent in text; Copilot Studio drafts the topics/instructions',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/nlu-gpt-overview',
            label: 'AI-based agent authoring overview - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      knowledgeBaseRag: {
        value:
          'Yes: agents can be grounded via retrieval-augmented generation over Dataverse tables, SharePoint/Office files, connectors to systems like Salesforce or ServiceNow, and a tenant-wide semantic search index (tenant graph grounding) over Microsoft Graph data including connector-synced external content',
        detail:
          'Tenant graph grounding is a separately billed, optional per-agent feature (10 Copilot Credits per message) layered on top of ordinary knowledge-source retrieval, for higher-quality, up-to-date grounding.',
        shortValue: 'RAG over Dataverse, SharePoint, connectors, plus tenant graph grounding',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-copilot-studio',
            label: 'Knowledge sources summary - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://www.microsoft.com/en-us/power-platform/blog/2026/05/05/dataverse-agent-data-platform/',
            label: 'Dataverse Is Your Agent Data Platform - Microsoft Power Platform Blog',
            asOf: '2026-07-02',
          },
        ],
      },
      mcpSupport: {
        value:
          'Yes: agents connect to external MCP servers as tools directly from the Tools page (Add Tool > New Tool > MCP, specifying just a URL), and tools/resources the server publishes are automatically added as actions that inherit their name, description, inputs, and outputs, staying in sync as the server changes',
        detail:
          'MCP servers are wired in via connector infrastructure under the hood, so they inherit enterprise controls like Data Loss Prevention policies, virtual network integration, and multiple authentication methods.',
        shortValue: 'Yes, connects to external MCP servers as auto-syncing tools',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/mcp-add-existing-server-to-agent',
            label:
              'Connect your agent to an existing MCP server - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/mcp-add-components-to-agent',
            label:
              'Add tools and resources from a Model Context Protocol (MCP) server to your agent - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      evaluationGuardrails: {
        value:
          "Yes: a dedicated test-set evaluation feature runs up to 100 test cases (authored manually, imported from a spreadsheet, or AI-generated from the agent's design/knowledge/topics) against an agent, alongside mandatory content moderation that screens both user input and agent responses for harmful content, jailbreaking, prompt injection, and prompt exfiltration",
        detail:
          'Content is evaluated twice per turn (at input and again before the response is returned); a detected violation blocks the response and shows an error rather than letting it through.',
        shortValue: 'Test-set evaluations (up to 100 cases) plus dual-stage content moderation',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/analytics-agent-evaluation-create',
            label: 'Create a single response test set - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/troubleshoot/power-platform/copilot-studio/generative-answers/agent-response-filtered-by-responsible-ai',
            label:
              'Resolve responsible AI content filter errors - Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      humanInTheLoop: {
        value:
          'Yes: a dedicated Request for Information action in agent flows pauses an automated run to collect structured input from a designated human reviewer via email, and a separate hand-off feature lets a conversational agent transfer an in-progress chat, with full history and variables, to a live human agent in a connected engagement hub',
        detail:
          'Request for Information targets agent-flow-style automation pausing on a decision point, and hand-off targets live conversational escalation to a human, covering both the automation-approval and conversational-escalation senses of human-in-the-loop.',
        shortValue: 'Request-for-Information flow action, plus live-agent conversation hand-off',
        confidence: 'verified',
        sources: [
          {
            url: 'https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/introducing-request-for-information-in-copilot-studio-agent-flows/',
            label:
              'Introducing request for information in Copilot Studio agent flows | Microsoft Copilot Blog',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/advanced-hand-off',
            label: 'Hand off to a live agent - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      generativeMedia: {
        value:
          'Partial: AI Builder, the shared Power Platform model catalog Copilot Studio prompts draw from, ships an image-description (captioning) model and GPT-based text generation, but no native image-generation, video-generation, or text-to-speech/speech-to-text model. Generating images or audio requires an external connector, such as Azure OpenAI DALL-E or Azure AI Speech.',
        detail:
          'Copilot Studio prompts and agent flows both draw from the same AI Builder catalog Power Automate uses, so the gap in native generative-media models is shared across both products.',
        shortValue:
          'Captioning and text gen only; image/audio generation needs an external connector',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/ai-builder/use-in-flow-overview',
            label: 'AI Builder in Power Automate overview - Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      dynamicToolUse: {
        value:
          "Yes: with generative orchestration turned on, an agent's LLM-driven planner dynamically selects which topics, tools, and knowledge sources to invoke at runtime from the full set attached to the agent, rather than following only the specific tool-call path a maker pre-wired into a fixed topic",
        detail:
          'This is what generative orchestration adds over classic topic-based authoring, where a maker had to hand-author which tool a given trigger phrase path calls.',
        shortValue: "Yes, via generative orchestration's runtime tool/topic selection",
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/generative-orchestration',
            label:
              'Apply generative orchestration capabilities - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      modelFallback: {
        value:
          'Yes: Copilot Studio automatically falls back to the default OpenAI GPT-4o model when a selected alternate model, such as Anthropic Claude, is disabled or unavailable for the tenant',
        detail:
          'This is a multi-model fallback behavior, not a broader multi-provider retry-on-rate-limit policy beyond that single fallback path.',
        shortValue: 'Falls back to the default OpenAI GPT-4o model automatically',
        confidence: 'verified',
        sources: [
          {
            url: 'https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/anthropic-joins-the-multi-model-lineup-in-microsoft-copilot-studio/',
            label: 'Anthropic joins the multi-model lineup in Microsoft Copilot Studio',
            asOf: '2026-07-02',
          },
        ],
      },
      agentSkills: {
        value:
          "Yes: Skills are named capabilities (name, description, Markdown instructions, optional supporting files) defined once, created in Copilot Studio or a text editor, added to multiple agents, and exported as a Markdown file or ZIP package for sharing, distinct from a single agent's own instructions",
        detail:
          'This is a preview feature in the new Copilot Studio agent-building experience. The standard SKILL.md format with YAML front matter makes skills portable between agents and organizations.',
        shortValue: 'Yes, named Markdown Skills reusable and exportable across agents',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/agents-experience/skills-overview',
            label:
              'Skills overview for agents (preview) - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      nativeChatDeployment: {
        value:
          'Yes: an agent can be published directly to a live website chat widget, Microsoft Teams, Microsoft 365 Copilot Chat, SharePoint, Power Pages, mobile, and further channels via Azure Bot Service (Slack, Telegram, Twilio SMS, and more), not only a form/API/webhook',
        detail:
          'Publishing pushes the current agent version out to every connected channel at once. A channel is "the integration point where an end-user can interact with a Copilot Studio agent."',
        shortValue: 'Yes, publishes to website, Teams, M365 Copilot Chat, and more channels',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/publication-fundamentals-publish-channels',
            label:
              'Key concepts - Publish and deploy your agent - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/publication-connect-bot-to-azure-bot-service-channels',
            label:
              'Publish an agent to Azure Bot Service channels - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      kbChunkVisibility: {
        value:
          "Partial: generative answers return citation links back to the knowledge source consulted for an answer, rendered automatically in the chat surface (Copilot Studio's own chat and Microsoft Teams, unless the response is customized), letting a user trace an answer to its source, but there is no documented chunk-level metadata (such as document/page numbers) or a dedicated raw chunk-content inspector distinct from the citation link itself.",
        detail:
          "Citations returned from a knowledge source currently can't be used as inputs to other tools or actions, per Copilot Studio's own knowledge-sources documentation, and citation rendering must be explicitly re-added when a generative answer is customized before being sent to Teams.",
        shortValue: 'Citation links trace an answer to its source; no chunk-inspector documented',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-copilot-studio',
            label: 'Knowledge sources summary - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/nlu-boost-node',
            label: 'Add a generative answers node - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      parallelExecution: {
        value:
          'Yes: agent flows, built on the Power Automate flow engine, support native parallel branches so multiple actions run concurrently and the flow proceeds only once every branch completes; loop actions separately support a parallel (concurrency) mode for independent iterations',
        detail:
          'Parallel branches and concurrent loop iterations add complexity and can affect output quality when combining concurrent results, and flows must still return within a documented action-count/time budget.',
        shortValue: 'Yes, native parallel branches and concurrent loop iterations in agent flows',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://www.microsoft.com/en-us/power-platform/blog/power-automate/parallel-actions/',
            label:
              'Add parallel branches in flows and five new services - Microsoft Power Platform Blog',
            asOf: '2026-07-02',
          },
          {
            url: 'https://www.candede.com/articles/copilot-studio-workflow-engine-loop-component/',
            label: 'Mastering Copilot Studio Workflows: The Loop Component',
            asOf: '2026-07-02',
          },
        ],
      },
      a2aProtocol: {
        value:
          'Yes: Copilot Studio ships a first-party Agent2Agent (A2A) connection type, generally available since April 2026, that lets an agent delegate tasks to any external agent implementing the open A2A protocol via an endpoint URL, with automatic Agent Card discovery and API key or OAuth 2.0 authentication.',
        detail:
          'This covers connecting a Copilot Studio agent out to an external A2A agent (first-party, second-party, or third-party). There is no documented feature to publish a Copilot Studio agent itself as a callable A2A server for other systems to reach, the same one-directional pattern as its MCP support.',
        shortValue: 'Yes, native A2A connections to external agents, GA since April 2026',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/add-agent-agent-to-agent',
            label:
              'Connect to an agent over the Agent2Agent (A2A) protocol - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-04',
          },
          {
            url: 'https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/new-and-improved-multi-agent-orchestration-connected-experiences-and-faster-prompt-iteration/',
            label:
              "What's new in Copilot Studio: Updates to multi-agent systems | Microsoft Copilot Blog",
            asOf: '2026-07-04',
          },
        ],
      },
      loopIteration: {
        value:
          'Yes: agent flows include a built-in Loop action supporting both For each (iterating a known collection) and Do until (repeating until a condition becomes true) patterns, running sequentially by default with an optional parallel/concurrency mode (a configurable degree-of-parallelism setting) for iterations that are fully independent',
        detail:
          'Operations accumulating a value or depending on order must stay Sequential, since parallel mode risks race conditions on shared variables and API throttling. Do until loops also require a maximum iteration count and timeout to prevent runaway execution.',
        shortValue: 'Yes, a Loop action with For each/Do until, sequential by default',
        confidence: 'verified',
        sources: [
          {
            url: 'https://www.candede.com/articles/copilot-studio-workflow-engine-loop-component/',
            label: 'Mastering Copilot Studio Workflows: The Loop Component',
            asOf: '2026-07-02',
          },
          {
            url: 'https://www.microsoft.com/en-us/power-platform/blog/power-automate/parallel-actions/',
            label:
              'Add parallel branches in flows and five new services - Microsoft Power Platform Blog',
            asOf: '2026-07-02',
          },
        ],
      },
    },
    integrations: {
      integrationCount: {
        value: 'Many pre-built connectors from the shared Power Platform connector catalog',
        detail:
          'Copilot Studio shares the same underlying Power Platform connector catalog that Power Automate and Power Apps use, split into standard connectors (included with all plans) and premium connectors (available on select plans), plus custom connectors for any other API. Microsoft does not publish a single current total connector count on this catalog documentation.',
        shortValue:
          'Many connectors from the shared Power Platform catalog, standard/premium/custom',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/advanced-connectors',
            label:
              'Use connectors in Copilot Studio agents - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      triggerTypes: {
        value:
          'Conversational trigger phrases (topics) that fire only when a user types a matching phrase, plus event triggers that let an agent act autonomously without user input in response to a connector-sourced event, such as a new SharePoint item, an added OneDrive file, a completed Planner task, or a changed Dataverse row, including a built-in, schedule-based Recurrence trigger',
        detail:
          "Event triggers require generative orchestration and deliver a JSON or plain-text payload describing the event to the agent, whose instructions then choose which action or topic to call in response. The Recurrence trigger is Copilot Studio's schedule-based event-trigger type, firing after a configured amount of time passes.",
        shortValue:
          'Topic trigger phrases, plus connector-sourced and schedule-based event triggers',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-triggers-about',
            label: 'Event triggers overview - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      customCodeSteps: {
        value:
          'No generic inline script step in the conversational designer; custom logic is reached via agent flows composed of Power Automate flow actions, or custom connectors wrapping an Azure Function or REST/SOAP API',
        detail:
          'Agent flows are built with the same low-code flow-action model as Power Automate, not an arbitrary code editor. A custom connector can front an Azure Function to run bespoke code as a callable action.',
        shortValue:
          'No inline script step; custom logic via agent flows or Azure Function connectors',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/advanced-flow-create',
            label: 'Create an agent flow as a tool - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      apiPublishing: {
        value:
          'Yes: a published agent is reachable over the Azure Bot Service Direct Line API/channel, giving external code a callable HTTP interface into the conversation, alongside the standard chat channels',
        detail:
          'Direct Line is the same Bot Framework mechanism used to embed a custom canvas or drive an agent from an external application, not a distinct, separately branded API-publishing feature.',
        shortValue: 'Yes, via the Azure Bot Service Direct Line channel/API',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/publication-connect-bot-to-azure-bot-service-channels',
            label:
              'Publish an agent to Azure Bot Service channels - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      extensibilitySdk: {
        value:
          'Microsoft Bot Framework SDK and custom-canvas hosting for programmatic chat-UI control, the Power Platform CLI (pac CLI) for solution/pipeline scripting, and the shared, open-source microsoft/PowerPlatformConnectors GitHub repo for community connector submissions',
        detail:
          'No separate, independently monetized third-party marketplace exists beyond the shared certified-connector catalog. Custom canvas hosting lets a developer take full programmatic control of the chat surface via Bot Framework.',
        shortValue: 'Bot Framework SDK, custom canvas, pac CLI, open-source connector repo',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/customize-default-canvas',
            label:
              'Customize the look and feel of an agent - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://github.com/microsoft/PowerPlatformConnectors',
            label: 'microsoft/PowerPlatformConnectors GitHub repo',
            asOf: '2026-07-02',
          },
        ],
      },
      mcpPublishing: {
        value:
          "No: Copilot Studio primarily consumes MCP servers, adding an existing server's tools/resources to an agent, the reverse direction. No feature lets a maker publish a Copilot Studio agent itself as a callable MCP server for external AI tools to invoke.",
        detail:
          'Exposing an MCP server for Copilot Studio to reach means hosting a custom MCP server separately (e.g. via Azure Container Apps or VS Code dev tunnels) and connecting Copilot Studio to it as a client, not publishing the agent as a server.',
        shortValue: 'Consumes MCP servers; no publish-your-own-agent-as-MCP feature found',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/mcp-add-components-to-agent',
            label:
              'Add tools and resources from a Model Context Protocol (MCP) server to your agent - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
    },
    pricing: {
      pricingModel: {
        value:
          'Consumption-based Copilot Credits, purchasable as prepaid capacity packs, pay-as-you-go against an Azure subscription, or a discounted annual commitment, billed across separately metered feature rates (classic/generative answers, agent actions, tenant graph grounding, agent flow actions, three tiers of AI tools, three tiers of voice)',
        detail:
          'Microsoft 365 Copilot licensed users get included usage at no additional Copilot Credit charge for employee-facing (business-to-employee) scenarios, up to fair-usage limits.',
        shortValue: 'Consumption-based Copilot Credits, several metered feature rates',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/requirements-messages-management',
            label: 'Billing rates and management - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      entryPaidPlan: {
        value:
          'Copilot Studio Pre-Purchase Plan: $200/month for a 25,000 Copilot Credit capacity pack (roughly 20% cheaper than pay-as-you-go), or pay-as-you-go at $0.01 per Copilot Credit with no upfront commitment',
        detail:
          'Both require an active Azure subscription for billing; a $200 free Azure credit is available for new accounts to trial the pay-as-you-go meter.',
        shortValue: '$200/month for 25,000 credits, or $0.01/credit pay-as-you-go',
        confidence: 'verified',
        sources: [
          {
            url: 'https://www.microsoft.com/en-us/microsoft-365-copilot/pricing/copilot-studio',
            label: 'Microsoft 365 Copilot Pricing - AI Agents | Copilot Studio',
            asOf: '2026-07-02',
          },
        ],
      },
      freeTier: {
        value:
          'No permanent free tier for building/running agents at production scale; a $200 Azure account credit and a Copilot Studio trial are available for new users, and Microsoft 365 Copilot licensed users get included agent usage for internal, employee-facing scenarios',
        detail:
          "The $200 Azure credit is a one-time trial allowance against the pay-as-you-go Copilot Credit meter, not an ongoing free plan. Microsoft 365 Copilot's included usage only applies to authenticated, licensed-user interactions.",
        shortValue: 'Trial credit and licensed-user included usage, no ongoing free tier',
        confidence: 'verified',
        sources: [
          {
            url: 'https://www.microsoft.com/en-us/microsoft-365-copilot/pricing/copilot-studio',
            label: 'Microsoft 365 Copilot Pricing - AI Agents | Copilot Studio',
            asOf: '2026-07-02',
          },
        ],
      },
      byok: {
        value:
          'Yes, for the underlying model rather than a raw provider API key: a maker can bring a specific Azure AI Foundry model deployment (via its endpoint URI, deployment name, and API key) into a prompt as a first-class Bring Your Own Model option, billed separately from the standard Copilot Credit meters',
        detail:
          "Currently available for individual prompts/tools. Microsoft's release plan lists using a brought-in model as the agent's primary response model, not just in prompts, as a future preview capability.",
        shortValue: 'Yes, via a bring-your-own Azure AI Foundry model deployment',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/bring-your-own-model-prompts',
            label:
              'Bring your own model for your prompts - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
    },
    security: {
      soc2: {
        value:
          'Yes: Copilot Studio (listed by its former name, "Copilot Studios") is one of the Microsoft online services explicitly in scope of the Office 365 SOC 2 Type 2 attestation report, with audit reports available from the Microsoft Service Trust Portal',
        detail:
          'Copilot Studio\'s own admin-certification page confirms SOC compliance without naming the specific report type, but Microsoft\'s dedicated SOC 2 Type 2 compliance offering page lists "Copilot Studios" by name among the in-scope Office 365 services, resolving which SOC report type applies.',
        shortValue: 'Yes, named in scope of the SOC 2 Type 2 attestation report',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-certification',
            label:
              'Review ISO, SOC, and HIPAA compliance - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-04',
          },
          {
            url: 'https://learn.microsoft.com/en-us/compliance/regulatory/offering-soc-2',
            label: 'SOC 2 Type 2 - Microsoft Compliance | Microsoft Learn',
            asOf: '2026-07-04',
          },
        ],
      },
      dataResidency: {
        value:
          'Yes: organizations can create agents in a specific environment/region so agent data resides within that geography, with Microsoft replicating only within the same geographic area for durability',
        detail:
          "Covered in Copilot Studio's dedicated geographic data residency documentation, distinct from the general Power Platform environment-region setting.",
        shortValue: 'Region-selectable environments keep agent data in-geography',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/geo-data-residency-security',
            label:
              'Geographic data residency - Security - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      rbac: {
        value:
          'Yes: Dataverse-backed security roles (System Administrator, Environment Maker, Bot Contributor, Bot Transcript Viewer, Analytics Viewer, and others) scope who can author, publish, view transcripts for, or view analytics on an agent, assignable directly or via Microsoft Entra ID group teams',
        detail:
          'Controlling agent creation requires layering tenant-level licensing/Author-group access with environment-level security roles, since neither alone is sufficient. Bot Transcript Viewer specifically scopes access to conversation transcripts.',
        shortValue:
          'Dataverse security roles (System Admin, Bot Contributor, Transcript Viewer, etc.)',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/sec-gov-phase3',
            label:
              'Secure your Copilot Studio projects - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      auditLogging: {
        value:
          'Yes: Copilot Studio provides dedicated audit logging covering admin, maker, and user activity across agents, viewable in the Microsoft Purview compliance portal',
        detail:
          'Covered in a Copilot Studio-specific audit-logging article, distinct from the general Microsoft 365 unified audit log Power Automate relies on for the same purpose.',
        shortValue: 'Yes, dedicated admin/maker/user audit logging via Microsoft Purview',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-logging-copilot-studio',
            label:
              'View audit logs for admins, makers, and users of Copilot Studio - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      additionalCompliance: {
        value:
          'HIPAA (Business Associate Agreement), HITRUST CSF, FedRAMP, multiple ISO standards (9001, 20000-1, 22301, 27001, 27017, 27018, 27701), PCI DSS, CSA STAR, UK G-Cloud, Singapore MTCS Level 3, Korea K-ISMS, and Spain ENS, each with an audit report on the Microsoft Service Trust Portal',
        detail:
          "This is the full list from Copilot Studio's admin-certification documentation. Each certification links to a corresponding audit report or certificate.",
        shortValue: 'HIPAA, HITRUST, FedRAMP, multiple ISO standards, PCI DSS, CSA STAR, and more',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-certification',
            label:
              'Review ISO, SOC, and HIPAA compliance - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      modelAndToolGovernance: {
        value:
          'Yes: admins can enable or restrict which non-default LLMs (e.g. Anthropic Claude) are available tenant-wide, and Data Loss Prevention policies classify connectors, and therefore the tools built from them, into Business/Non-Business/Blocked groups at the tenant or environment level',
        detail:
          'Model access control is a distinct admin toggle from DLP connector classification. Together they cover both which model an agent may use and which connector-backed tools it may call.',
        shortValue: 'Yes, admin-toggled model access plus DLP-based connector/tool classification',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/anthropic-joins-the-multi-model-lineup-in-microsoft-copilot-studio/',
            label: 'Anthropic joins the multi-model lineup in Microsoft Copilot Studio',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-data-loss-prevention',
            label:
              'Configure data policies for agents - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      credentialGovernance: {
        value:
          'No: Data Loss Prevention policies and Advanced Connector Policies govern access at the connector/action/endpoint level (e.g. blocking the SharePoint connector tenant-wide), not at the level of restricting which specific stored credential or connection a given role may use for the same connector',
        detail:
          'This is the same Power Platform DLP model Power Automate uses, since Copilot Studio agent flows and connectors sit on the same underlying platform.',
        shortValue: 'DLP governs connectors, not specific stored credentials by role',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/power-platform/admin/wp-data-loss-prevention',
            label: 'Data policies - Power Platform | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      whiteLabeling: {
        value:
          'No: agent-level branding is limited to a custom name, icon, and accent color in Settings, or fully hosting a custom canvas web app for the chat window (via CSS/JavaScript or the Bot Framework SDK). There is no option to remove Microsoft/Copilot Studio branding from the authoring canvas or the default published chat UI itself.',
        detail:
          'A custom canvas gives full programmatic control over the embedded chat experience, but that is a developer-built replacement UI, not a native white-labeling setting inside the product.',
        shortValue: 'No, only name/icon/color or a fully custom-built chat canvas',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/customize-default-canvas',
            label:
              'Customize the look and feel of an agent - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      dataRetention: {
        value:
          'Partial: conversation transcripts downloaded directly from Copilot Studio only cover the past 29 days, while the underlying Dataverse conversation-transcript and custom-analytics tables default to a 30-day retention period; extending retention beyond that default requires either an admin changing the Dataverse retention setting or exporting the data via Azure Synapse Link for Dataverse into Azure Data Lake Storage Gen2',
        detail:
          "Copilot Studio's own download experience is capped at the past 29 days regardless of the underlying Dataverse retention window. Microsoft's custom-analytics guidance documents the 30-day default retention on the Dataverse-side bot/botcomponent/conversationtranscript tables and recommends Synapse Link as the export path for longer-term or custom-reporting needs.",
        shortValue:
          '29-day direct download; 30-day default Dataverse retention, extendable via export',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/analytics-transcripts-studio',
            label:
              'Understand downloaded session data from Copilot Studio - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/custom-analytics-strategy',
            label:
              'Develop a custom analytics strategy - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      piiRedaction: {
        value:
          'Yes, but as a block rather than in-line redaction: Microsoft Purview inline Data Loss Prevention for Copilot Studio agents (public preview) scans prompts sent to an agent in real time for Sensitive Information Types (SSNs, credit card numbers, custom types), and blocks the prompt from being processed, with no AI response generated, if one is detected before the agent is invoked.',
        detail:
          'This stops sensitive content from reaching the agent at all rather than redacting it in-line and continuing. It is configured as a Purview DLP policy targeting the Copilot Studio location, separate from the product itself.',
        shortValue: 'Purview inline DLP blocks prompts containing detected PII before invocation',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/purview/dlp-microsoft365-copilot-location-learn-about',
            label:
              'Microsoft Purview DLP for Microsoft 365 Copilot and Copilot Chat | Microsoft Learn',
            asOf: '2026-07-08',
          },
        ],
      },
      sso: {
        value:
          'Yes: Copilot Studio and the wider Power Platform are built on Microsoft Entra ID, which natively supports SAML/OIDC single sign-on plus automatic user/app provisioning (SAML just-in-time and SCIM-based), so signing in via Entra ID grants org-level access without manual per-user account setup',
        detail:
          "SSO and provisioning are inherited from the Microsoft 365/Entra ID tenant rather than a Copilot Studio-specific setting, consistent with the rest of Microsoft's enterprise stack.",
        shortValue: 'Yes, SSO via Entra ID (SAML/OIDC) with automatic provisioning',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/entra/identity/app-provisioning/user-provisioning',
            label:
              'What is automated app user provisioning in Microsoft Entra ID | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      thirdPartyVetting: {
        value:
          'Partial: the primary connector catalog (1,000+ connectors) is Microsoft-certified, requiring code/security review, functional verification, and malware scanning before publication, including connectors submitted by third-party (independent) publishers. But any maker can also build an uncertified custom connector against an arbitrary API for their own tenant, and security researchers at Zenity have shown that custom connectors can bypass Power Platform Data Loss Prevention policies to reach connectors an admin has explicitly blocked',
        detail:
          "Microsoft's connector certification process scans all submitted connector code for malware or viruses and runs a functional/security review before a connector joins the shared, browsable catalog. Custom connectors skip that review entirely unless separately submitted for certification, and Zenity's published research demonstrates a custom connector reaching a DLP-blocked connector, a documented real-world weakness in the uncertified path.",
        shortValue: 'Certified catalog is vetted; uncertified custom connectors can bypass DLP',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/connectors/custom-connectors/submit-certification',
            label: 'Get your connector certified - Overview | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://zenity.io/blog/research/microsoft-power-platform-dlp-bypass-uncovered-finding-3-custom-connectors',
            label:
              'AI Agent Security | Microsoft Power Platform DLP Bypass Uncovered - Finding #3 - Custom Connectors | Zenity',
            asOf: '2026-07-02',
          },
        ],
      },
    },
    observability: {
      tracingDepth: {
        value:
          'Yes: per-session conversation transcripts show which topic fired, which knowledge sources were consulted, which tools were called, which child agents or MCP servers were invoked, what the orchestration plan was, and how long each step took, alongside an Analytics dashboard with conversation volume, engagement, satisfaction, and response-quality metrics over time',
        detail:
          'Telemetry can additionally be sent to Azure Monitor Application Insights, where a dedicated Copilot Studio dashboard workbook surfaces total conversations, latency, exceptions, tool usage, and topic analytics in one view for deeper analysis.',
        shortValue: 'Per-session traces (topic/tools/knowledge/timing) plus a metrics dashboard',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/agents-experience/analytics-overview',
            label:
              'Monitor an agent overview (preview) - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/advanced-bot-framework-composer-capture-telemetry',
            label:
              'Capture telemetry with Application Insights - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      durabilityModel: {
        value:
          'Yes for agent-flow steps: because agent flows are built on the Power Automate flow engine, individual flow actions can use configurable retry policies with backoff, the same durability mechanism Power Automate exposes. A documented troubleshooting pattern for tool-call timeouts explicitly recommends adding retry with backoff',
        detail:
          'This durability applies to the deterministic agent-flow layer. No separate, distinct replay/checkpoint mechanism exists for the conversational/generative-orchestration layer itself.',
        shortValue:
          'Yes, per-action retry with backoff inherited from the Power Automate flow engine',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/troubleshoot/power-platform/copilot-studio/authoring/error-codes',
            label: 'Understand Error Codes - Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      failureAlerting: {
        value:
          "Partial: proactive failure alerting is reachable through Azure Monitor alert rules (email, SMS, or webhook action groups) on Application Insights exceptions/latency once telemetry is wired up, but no Copilot Studio-specific documentation describes an automatic, built-in per-run failure-email or weekly-digest feature comparable to Power Automate's flow-failure notifications.",
        detail:
          "Copilot Studio's Analytics area is dashboard/lookup-based, so a maker must open it to see failures, while pushing a notification depends on separately configuring Azure Monitor alert rules on the Application Insights resource. Confidence is marked unknown for the negative half of this claim: no Microsoft Learn page was found that explicitly confirms or rules out a native failure-alert feature inside Copilot Studio itself.",
        shortValue:
          'Alerting requires configuring Azure Monitor; no native failure-email feature confirmed',
        confidence: 'unknown',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/azure/azure-monitor/alerts/alerts-overview',
            label: 'Overview of Azure Monitor alerts - Azure Monitor | Microsoft Learn',
            asOf: '2026-07-04',
          },
        ],
      },
      dataDrains: {
        value:
          'Yes: telemetry can be continuously streamed to Azure Monitor Application Insights, and conversation transcripts/custom analytics data stored in Dataverse can be continuously exported via Azure Synapse Link for Dataverse into Azure Data Lake Storage Gen2 in Common Data Model format',
        detail:
          'This is the same Azure-native export pattern used across Power Automate and Power Platform more broadly (Microsoft Sentinel for audit/activity logs, Application Insights for telemetry), not a generic user-configurable webhook/S3/BigQuery drain built directly into Copilot Studio.',
        shortValue: 'Yes, exports to Application Insights and Azure Data Lake via Synapse Link',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/custom-analytics-strategy',
            label:
              'Develop a custom analytics strategy - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      asyncExecution: {
        value:
          "Yes: agent flows inherit the Power Automate flow engine's asynchronous response pattern (an HTTP 202 plus a Location header the caller polls), letting a long-running flow action continue beyond a synchronous request's time limit",
        detail:
          'Agent flow actions invoked by an agent are still bound by a 100-second action limit within a conversational turn, distinct from the longer-running asynchronous pattern available to a flow triggered independently of a live conversation.',
        shortValue: 'Yes, via the same 202 + Location polling pattern as Power Automate',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/power-automate/guidance/coding-guidelines/asychronous-flow-pattern',
            label: 'Use asynchronous responses - Power Automate | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      executionLimits: {
        value:
          'Microsoft publishes concrete limits: agent flow actions invoked by an agent must respond within roughly 100-120 seconds per conversational turn; express-mode flow runs are capped at 100 actions and a 64 KB message size per action; and generative AI/Copilot Credit throttling applies per Dataverse environment when consumption exceeds capacity.',
        detail:
          'These limits are specific to the agent-flow/tool-calling layer inside a live conversation. The underlying flow engine also carries broader Power Automate execution limits (30-day max run, etc.) when a flow runs independently of an agent turn.',
        shortValue: '~100-120s per-turn action limit, 100-action/64KB express-mode cap',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/agent-flow-express-mode',
            label:
              'Speed up agent flow execution with express mode (preview) - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://learn.microsoft.com/en-us/troubleshoot/power-platform/copilot-studio/licensing/throttling-errors-agents',
            label:
              'Resolve usage limit and agent unavailable errors in Copilot Studio agents - Copilot Studio | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      partialFailureHandling: {
        value:
          'Yes for agent-flow steps: because agent flows run on the Power Automate flow engine, an individual failing action can be routed to an error-handling path via the same per-action "Configure run after" (has failed / is skipped / has timed out) setting, letting the rest of the flow continue rather than the whole run always halting',
        detail:
          'This inherits directly from the Power Automate flow-action model agent flows are built on. No separate, distinct partial-failure mechanism exists for the conversational/topic layer.',
        shortValue: "Yes, via the inherited per-action 'Configure run after' setting",
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/power-automate/guidance/coding-guidelines/error-handling',
            label: 'Employ robust error handling - Power Automate | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      unattendedExecution: {
        value:
          "Yes in Commercial environments: autonomous event-triggered agent runs and agent flows execute on Microsoft-operated cloud infrastructure, not on a maker's own device or browser session. Per Microsoft's own GCC/GCC High licensing documentation, Triggers/Autonomous Agents are currently NOT available in GCC or GCC High government cloud environments.",
        detail:
          "Autonomous triggers let an agent proactively respond to a connector event or schedule without a live conversation open, and agent flows run on the Power Automate flow engine's server-side runtime in Commercial environments. Closing the authoring browser tab or shutting down a laptop has no effect on a published agent's ability to fire on a trigger or complete a run, but Microsoft's GCC/GCC High feature-availability table lists Triggers/Autonomous Agents as unavailable in those government-cloud environments.",
        shortValue: 'Yes in Commercial cloud; not available in GCC/GCC High per Microsoft docs',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-copilot-studio/requirements-licensing-gcc',
            label: 'US Government customers - Microsoft Copilot Studio | Microsoft Learn',
            asOf: '2026-07-08',
          },
          {
            url: 'https://adoption.microsoft.com/files/copilot-studio/Autonomous-agents-with-Microsoft-Copilot-Studio.pdf',
            label: 'Autonomous Agents with Microsoft Copilot Studio',
            asOf: '2026-07-08',
          },
        ],
      },
    },
    support: {
      supportChannels: {
        value:
          'Documentation via Microsoft Learn, the Power Platform/Power Users community forum and Microsoft Q&A for general questions, and the Microsoft 365 Admin Center for business-critical, SLA-based support requests',
        detail:
          "Microsoft's guidance distinguishes routine community/Q&A support from business-critical issues, which go through the Microsoft 365 Admin Center support flow rather than the public forums.",
        shortValue: 'Docs, community/Q&A forums, and Admin Center for business-critical issues',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/feedback',
            label:
              'Microsoft 365 Copilot Developer Community Support and Feedback Channels | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      sla: {
        value:
          "Not publicly documented as a Copilot Studio-specific, financially backed SLA: Microsoft's Service Level Agreements for Online Services document lists uptime terms for Azure, Dynamics 365, Office 365, and Intune, but names neither Copilot Studio nor Power Virtual Agents, while core services like Exchange Online, SharePoint Online, and Teams carry a widely cited 99.9% financially backed uptime guarantee",
        detail:
          'Independent reporting on the June 2026 Microsoft 365 Copilot outages noted enterprise customers lack the same financially backed SLA protection for Copilot that exists for core services like Exchange Online, since many enterprise agreements do not explicitly define uptime commitments for AI components.',
        shortValue: 'No Copilot Studio-specific SLA found in the Online Services SLA document',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://www.microsoft.com/licensing/docs/view/Service-Level-Agreements-SLA-for-Online-Services?lang=1',
            label: 'Service Level Agreements (SLA) for Online Services - Microsoft Licensing',
            asOf: '2026-07-04',
          },
          {
            url: 'https://windowsnews.ai/article/microsoft-365-copilot-outage-exposes-ai-reliability-gaps-in-enterprise-slas.425641',
            label: 'Microsoft 365 Copilot Outage Exposes AI Reliability Gaps in Enterprise SLAs',
            asOf: '2026-07-04',
          },
        ],
      },
      community: {
        value:
          'Large: an active, Microsoft-hosted Power Platform/Power Users community forum with structured Q&A on building, publishing, and troubleshooting Copilot Studio agents, plus Microsoft Q&A for developer-specific questions',
        detail:
          'The same community forum infrastructure serves the broader Power Platform (Power Apps, Power Automate, Copilot Studio), not a dedicated, separately branded Copilot Studio-only forum.',
        shortValue: 'Large, shared Power Platform community forum plus Microsoft Q&A',
        confidence: 'estimated',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/feedback',
            label:
              'Microsoft 365 Copilot Developer Community Support and Feedback Channels | Microsoft Learn',
            asOf: '2026-07-02',
          },
        ],
      },
      companyMaturity: {
        value:
          'Microsoft Corporation. Founded April 4, 1975. Approximately 228,000 employees. Market capitalization approximately $2.8 trillion USD. Publicly traded (NASDAQ: MSFT) with quarterly revenue in the $80B+ range as of FY2026 SEC filings',
        detail:
          "Copilot Studio is a product within Microsoft's Power Platform/Business Applications segment, backed by Microsoft's overall corporate scale, not an independent startup.",
        shortValue: 'Microsoft Corporation. Public, ~228,000 employees',
        confidence: 'verified',
        sources: [
          {
            url: 'https://www.microsoft.com/en-us/investor/earnings/fy-2026-q3/press-release-webcast',
            label: 'FY26 Q3 - Press Releases - Investor Relations - Microsoft',
            asOf: '2026-07-04',
          },
          {
            url: 'https://stockanalysis.com/stocks/msft/market-cap/',
            label: 'Microsoft (MSFT) Market Cap - StockAnalysis.com',
            asOf: '2026-07-02',
          },
        ],
      },
      academy: {
        value:
          'Yes: Microsoft Learn provides structured, self-paced training paths, hands-on labs (Microsoft Copilot Studio Labs, Agent Academy), and official certifications spanning Copilot Studio and the broader Power Platform, including Microsoft Certified: Power Platform Fundamentals (PL-900)',
        detail:
          'Beyond formal certification paths, Microsoft publishes dedicated, scenario-based hands-on labs (mcs-labs, agent-academy) specifically for Copilot Studio, going beyond ad hoc docs or blog content.',
        shortValue: 'Microsoft Learn courses, hands-on labs, and PL-900 certification',
        confidence: 'verified',
        sources: [
          {
            url: 'https://learn.microsoft.com/en-us/credentials/certifications/power-platform-fundamentals/',
            label:
              'Microsoft Certified: Power Platform Fundamentals - Certifications | Microsoft Learn',
            asOf: '2026-07-02',
          },
          {
            url: 'https://microsoft.github.io/mcs-labs/labs/setup-for-success/',
            label:
              'Set yourself up for success & discover ALM best practices | Microsoft Copilot Studio Labs',
            asOf: '2026-07-02',
          },
        ],
      },
    },
  },
}