Microsoft Copilot Studio

User Guide — Build, Deploy & Govern AI Agents

Zoom

What this guide covers

Copilot Studio is Microsoft's low-code platform for building custom AI agents. It is the builder layer — separate from M365 Copilot Chat, which is the end-user AI assistant. You use Copilot Studio to create agents that answer questions, automate workflows, and connect to your organization's data sources.

Agents can be deployed to Teams, SharePoint, public websites, WhatsApp, and embedded directly in M365 Copilot Chat. You connect them to SharePoint, Dataverse, Salesforce, ServiceNow, and 1,400+ other systems.

Low-code builder Teams deployment SharePoint grounding Knowledge sources Credit billing DLP & security Declarative agents Custom engine agents
ⓘ Builder vs. User: Copilot Studio is where you build agents. M365 Copilot Chat is where users chat with them. You need a Copilot Studio maker license to build; end users need no special license to use agents you deploy internally (with an M365 Copilot license covering credits).
Enterprise security: Copilot Studio respects your existing Microsoft permissions, supports Customer-Managed Encryption Keys, Customer Lockbox, Purview audit logs, and Power Platform DLP policies. Data stays in your chosen Azure region.

Copilot Studio: The Agent Builder Layer

Copilot Studio is a cloud-based, graphical low-code platform. It is not the end-user chat experience — it is the tooling you use to build, configure, and deploy AI agents across your organization.

🤖 Builder vs. User — Two Different Products

Copilot Studio (This Guide)
Where makers build custom agents. Connect knowledge sources, define conversation flows, configure actions, and publish to channels. Requires a maker license. Accessible at copilotstudio.microsoft.com.
M365 Copilot Chat
Where end users interact with AI. Available in Teams, Word, Excel, Outlook, and Copilot.cloud.microsoft. Can surface agents built in Copilot Studio but is not the building environment.

🔗 Agents built in Copilot Studio can appear inside M365 Copilot Chat — that is the integration point, not a shared interface.

Two Agent Types

Choose the right type before you build. The two types have fundamentally different architectures.

Feature Declarative Agent Custom Engine Agent
Orchestration Microsoft's Copilot infrastructure Your own (Copilot Studio or custom code)
Models Copilot's foundation models only Any model: Azure AI Foundry, GPT-4o, o1, third-party
Channels M365 apps only (Teams, Word, Outlook, Copilot Chat) M365 + web + WhatsApp + Facebook + custom
Autonomous triggers Not supported Fully supported (25 credits per trigger)
DLP policies Not directly supported Full Power Platform DLP
Agent-to-agent Not supported Fully supported via A2A protocol
Compliance / RAI Inherited from M365 automatically Must implement for non-Microsoft models
Hosting Microsoft 365 (no extra hosting cost) Copilot Studio or external Azure
Best for Internal Q&A, HR policy, sales enablement Customer bots, complex workflows, multi-channel

Core Capabilities

Low-Code Authoring Canvas

  • Drag-and-drop conversation flow designer with visual topic trees
  • Condition and switch nodes, variable management, branching logic
  • Adaptive Cards, quick-reply buttons, rich message formats
  • Power Fx formula integration for dynamic logic
  • Natural-language description to auto-generate topics and agent setup
  • Generative orchestration: AI decides which topic to invoke based on user intent

Deployment Channels

  • Microsoft Teams — first-class channel, no SDK setup
  • SharePoint — published as a .agent file in Shared Documents
  • Websites — iframe embed or self-hosted WebChat
  • WhatsApp — GA as of September 2025
  • M365 Copilot Chat — agent surfaces in the Copilot sidebar
  • Facebook Messenger, Slack, Twilio SMS — via Azure Bot Service
  • Voice / Telephony — via Azure Communication Services

Multi-Turn Conversations

Agents maintain conversation context across multiple turns. Topics redirect to one another, variables carry user data through the flow, and inactivity triggers clear stale state. Generative orchestration handles unstructured user language without exhaustive trigger phrase lists.

Voice Support

  • Accepts speech and DTMF input from telephony
  • Natural text-to-speech responses
  • Barge-in control and silence detection
  • Three tiers: Basic (10 credits/min), Standard (35/min), Premium real-time (75/min)

Multi-Agent Orchestration (A2A)

Agents built in Copilot Studio can delegate tasks to other agents, including those built with Azure AI Foundry and Microsoft Fabric. The open Agent2Agent (A2A) protocol enables cross-platform agent collaboration as of May 2025.

Example Agent Build Scenarios

Real-world agent patterns you can build with Copilot Studio today.

01
HR Policy Q&A Agent
Build a declarative agent grounded in your SharePoint HR policy library. Deploy to Teams so employees can ask policy questions in natural language.
Add your HR SharePoint site as a knowledge source, write instructions scoping the agent to HR only, publish to Teams.
02
IT Helpdesk Ticket Agent
Create a custom engine agent that collects issue details from users and creates ServiceNow tickets automatically via a Power Automate flow.
Design a topic that gathers Category, Priority, and Description, then calls a Power Automate flow that posts to ServiceNow REST API.
03
Customer-Facing FAQ Bot
Deploy a public-facing agent on your website that answers product or service questions grounded in your Confluence knowledge base.
Connect Confluence as a knowledge source, configure WebChat, embed the iframe snippet on your web page. Enable generative orchestration for natural phrasing.
04
Sales Enablement Agent
Surface product specs, pricing guides, and competitive battlecards inside M365 Copilot Chat for your sales team without leaving their workflow.
Create a declarative agent with OneDrive and SharePoint knowledge sources scoped to the Sales team's document libraries. Publish to M365 Copilot.
05
Autonomous Document Processor
Build an agent that triggers automatically when a file is uploaded to a SharePoint folder, processes it with AI Tools, and routes the output to Dataverse.
Use an autonomous trigger (25 credits each) watching a SharePoint library. Run a content processing tool (8 credits/page) and write results to Dataverse via connector.
06
WhatsApp Customer Support Agent
Route customer WhatsApp messages to your Copilot Studio agent, answer common questions with generative AI, and escalate to a human when needed.
Enable the WhatsApp channel (GA Sept 2025), connect a Zendesk or ServiceNow knowledge source, add an escalation topic that hands off to a live agent queue.
07
Meeting Prep Agent
An agent that accepts a meeting title or attendee name and retrieves relevant SharePoint documents, Outlook emails, and Dataverse records to generate a briefing.
Connect Outlook, SharePoint, and Dataverse as actions. Use tenant graph grounding (10 credits). Deliver a formatted Adaptive Card briefing in Teams.
08
Procurement Approval Router
Collect purchase request details via a Teams bot, run approval logic with Power Automate, and post status updates back to the requestor automatically.
Author a topic that collects Vendor, Amount, and Justification. Trigger an approval flow (13 credits per 100 flow actions). Post approval outcome via Adaptive Card reply.

Knowledge Sources

Agents use Retrieval Augmented Generation (RAG) — they retrieve semantically relevant document chunks per query and send them to the language model for answer generation. Connect one or many sources per agent.

Knowledge SourceLimits / Notes
SharePoint sites and libraries Modern pages only; classic ASPX and SPFx pages not indexed; ~25 docs retrieved per query; 2,048-row max for lists
OneDrive files and folders Up to 1,000 files per source; sync every 4 to 6 hours
Uploaded files (PDF, DOCX, PPTX, XLSX) Up to 512 MB per file; maximum 500 files per agent
Dataverse tables Up to 2 sources and 15 tables per agent
Public URLs Agent crawls and indexes the URL; good for external documentation or FAQs
Salesforce Knowledge No article count limit; syncs every 4 to 6 hours
Confluence No article count limit; syncs every 4 to 6 hours
ServiceNow Knowledge No article count limit; syncs every 4 to 6 hours
Zendesk No article count limit; syncs every 4 to 6 hours
Azure AI Search indexes More control over chunking, ranking, and filtering; recommended for large document corpora
Copilot connectors (40+ graph connectors) Bring external data into Microsoft Graph; preview or GA depending on connector

SharePoint Grounding — Details and Known Limitations

How SharePoint Grounding Works

  • Agent performs semantic retrieval — finds the most relevant chunks for a query, not every document
  • The Copilot Retrieval API caps results at approximately 25 documents per query
  • Files larger than 7 MB require an M365 Copilot license in the same tenant to be indexed; otherwise the file is skipped or truncated
  • The Work IQ feature (requires M365 Copilot license) improves SharePoint grounding quality and supports files up to 200 MB
  • Metadata filtering by filename, owner, and last-modified date was added in November 2025
  • SharePoint list queries return a maximum of 2,048 rows
  • Up to 15 SharePoint lists can be added per session in the Add Knowledge dialog
  • Sensitivity labels are surfaced in the agent chat, showing the highest label applied to sources used in a response

Known SharePoint Grounding Limitations

  • Not supported Classic ASPX pages
  • Not supported SPFx component pages
  • Not supported Accordion navigation / custom CSS SharePoint sites
  • Not supported Queries referencing files by name ("What does file X say?")
  • Not supported Guest users (SSO-enabled apps block generative answers for guests)
  • Not supported Saved SharePoint list views — only the full list can be added
  • Not indexed Attachments column content in SharePoint lists
  • Not indexed Confidential or Highly Confidential sensitivity-labeled files
  • Limitation Multi-document comparison and policy conflict analysis are unreliable given the 25-doc cap
  • Limitation "List all documents where X" queries fail — retrieval is semantic, not exhaustive

Actions

Agents can do more than answer questions — they can take actions by calling APIs, running flows, and invoking other agents.

Power Platform Connectors (1,400+)

  • Standard connectors — SharePoint, OneDrive, Teams, Outlook (included with all plans)
  • Premium connectors — Salesforce, ServiceNow, SAP, Dynamics 365 (requires standalone subscription)
  • Custom connectors — any REST API via OpenAPI (Swagger) definition

Power Automate Flows

Triggered from agent conversations for multi-step automations. Power Automate flows as actions are the recommended pattern for orchestrating multi-system integrations. Billed at 13 credits per 100 flow actions.

HTTP Requests

Direct REST calls from within an agent topic. Can be restricted or blocked by Power Platform DLP policies. Billed at 5 credits per invocation.

Model Context Protocol (MCP)

GA as of May 2025. MCP connects agents to 1,400+ external systems using a standardized open protocol. Enables agents to call tools exposed by MCP servers without custom connectors.

AI Tools

  • Run AI prompts directly inside a topic flow
  • Extract and process document content (8 credits per page)
  • Call Azure AI Foundry models (custom fine-tuned or foundation models)
  • Text basic (GPT-4o-mini): 1 credit per 10 responses
  • Text standard (GPT-4o): 15 credits per 10 responses
  • Text premium / reasoning (o1): 100 credits per 10 responses

Multi-Agent Orchestration (A2A Protocol)

Agent2Agent (A2A) Protocol
An open protocol that allows agents from different platforms to collaborate. A Copilot Studio agent can delegate a subtask to an Azure AI Foundry agent, a Microsoft Fabric agent, or even a third-party agent that supports A2A. GA as of May 2025.
Orchestrator Pattern
Build a router agent that receives user requests and delegates to specialized atomic agents (HR agent, IT agent, Finance agent). The router uses generative orchestration to decide which sub-agent handles each request.

📚 Multi-agent orchestration is available for custom engine agents only. Declarative agents do not support agent-to-agent collaboration.

Licensing Paths

The billing unit is the Copilot Credit (renamed from "message" in September 2025; price unchanged). Three primary paths cover most organizations:

Option 1 — Standalone Capacity Pack
$200 /pack/month

25,000 credits per pack per month. Annual subscription. Effective rate $0.008/credit (20% cheaper than PAYG). Multiple packs stack additively. Unused credits do not roll over.

Best for: Predictable volume, external-facing agents, organizations not on M365 Copilot.

Option 2 — Pay-As-You-Go
$0.01 /credit

Billed in arrears via Azure. No annual commitment. Agents are never blocked at 125% — credits billed to Azure subscription. Use as primary billing or as overflow safety net alongside capacity packs.

Best for: Variable or low usage, development and testing, overflow protection.

Option 3 — Pre-Purchase Plan (CCPU)
$2,850 to $2.4M upfront

Volume discounts of 5 to 20%. CCCUs (Copilot Committed Capacity Units) expire after 1 year from purchase. MACC-eligible. 100 credits per CCCU purchased.

Best for: High-volume or seasonal usage, organizations with Azure consumption commitments.

Option 4 — M365 Copilot License
$30 /user/month

Unlimited internal (B2E) Copilot Studio agent usage at fair-use limits for M365 Copilot licensed users. Annual commitment. Covers most internal agent features except autonomous triggers, voice, and external-facing usage.

Best for: Organizations already deploying M365 Copilot. Most cost-effective for internal agents.

Credit Billing Rates

Features stack within a single interaction. Example: a generative answer using tenant graph grounding = 2 + 10 = 12 credits per response.

FeatureCreditsNotes
Classic answer (scripted topic response) 1 Base cost for any interaction using a scripted flow
Generative answer (AI-generated from knowledge source) 2 RAG-based answer from SharePoint, OneDrive, uploaded files, etc.
Agent action (connector call, HTTP request) 5 Per invocation; each connector call costs 5 credits
Tenant graph grounding 10 Querying org-wide Microsoft Graph data (SharePoint, Teams, etc.)
Agent flow actions (Power Automate-style) 13 per 100 Embedded flows inside Copilot Studio; separate enforcement from agent credits
AI Tools — Text basic (GPT-4o-mini) 1 per 10 0.1 credits per response; cost-effective for simple prompt tasks
AI Tools — Text standard (GPT-4o) 15 per 10 1.5 credits per response; good balance of capability and cost
AI Tools — Reasoning model (o1) 100 per 10 10 credits per response; very expensive at scale — use only when chain-of-thought reasoning is essential
Content processing tools 8 per page Document processing, OCR, extraction per page
Autonomous agent trigger 25 always Always billed per trigger fire — no M365 Copilot exemption. Estimate monthly trigger volume carefully.

Rate Limits and Overage Behavior

Rate Limits

ScopeLimit
Agent messages (any type), paid plan8,000 RPM per Dataverse environment
Generative AI — 1 to 10 capacity packs50 RPM / 1,000 RPH
Generative AI — 11 to 50 capacity packs80 RPM / 1,600 RPH
Generative AI — 51 to 150 capacity packs100 RPM / 2,000 RPH
Pay-as-you-go or M365 Copilot100 RPM / 2,000 RPH

Overage Enforcement

ThresholdWhat Happens
Under 100% consumedNormal operation
100% consumedOverage begins; grace period applies
125% consumedCustom agents disabled after current conversation ends
After enforcementUsers see "agent unavailable" message

Agents in pay-as-you-go environments are never blocked. Pair capacity packs with PAYG as an overflow safety net. Set per-agent credit caps in Power Platform admin center under Licensing > Copilot Studio > Manage Agents.

What M365 Copilot License Covers (Zero-Rated for Licensed Users)

Included at No Additional Credit Cost

  • Zero-rated Classic answers
  • Zero-rated Generative answers
  • Zero-rated Agent actions (connector calls)
  • Zero-rated Tenant graph grounding
  • Zero-rated Agent flow actions
  • Zero-rated All AI Tools tiers (basic, standard, premium)
  • Zero-rated Content processing tools

Applies in: Copilot Chat, Teams, SharePoint (via M365 Copilot), Word, Excel, PowerPoint, Outlook. Fair usage limits apply.

Always Billed (Not Covered by M365 Copilot)

  • Always billed Autonomous agent triggers (25 credits each)
  • Always billed External-facing (B2C) scenarios — public websites, WhatsApp, Facebook serving unlicensed users
  • Always billed Internal users without an M365 Copilot license
  • Always billed Voice interactions (billed per minute regardless of license)

Deployment Channels

Publish once in Copilot Studio; the agent is available across all selected channels simultaneously.

ChannelTypeSetup ComplexityKey Notes
Microsoft Teams Internal Low First-class channel; no SDK setup; publish directly from Studio; appears as a Teams app
SharePoint Internal Low GA May 2025; published as a .agent file in Shared Documents; appears in the Copilot menu on the site
M365 Copilot Chat Internal Low Agent surfaces in the Copilot sidebar across Word, Excel, PowerPoint, Outlook, Teams
Website (iframe / WebChat) Internal or External Medium Microsoft-hosted or self-hosted; full styling control via botframework-webchat library; paste iframe snippet into any HTML page
WhatsApp External Medium GA September 2025; requires WhatsApp Business API account
Facebook Messenger External Medium Via Azure Bot Service channel; requires Facebook developer app
Voice / Telephony External High Via Azure Communication Services; three billing tiers (10 / 35 / 75 credits per minute)
Azure Bot Service channels External High Slack, Twilio SMS, Line, Kik, and others via Azure Bot Framework
Direct Line API Custom High Public REST API for connecting any custom client app; for developers building custom UIs

Data Loss Prevention (DLP) — Two Separate Systems

Power Platform DLP

For custom engine agents built in Copilot Studio

  • Configured in the Power Platform admin center
  • Classifies connectors into Business, Non-business, and Blocked groups
  • Connectors in different groups cannot share data within the same agent
  • Enforced in real time; makers see errors immediately for violations
  • DLP enforcement is enabled by default for all tenants as of early 2025
  • Can block: auth types, knowledge sources (e.g., block SharePoint grounding), specific connectors, HTTP requests, publication to certain channels, autonomous triggers
  • Supports PowerShell cmdlets for custom error messages with governance links

Microsoft Purview DLP

For M365 Copilot and Copilot Chat interactions only

  • Scans prompts in real time for Sensitive Information Types (SITs) before processing
  • Can block Copilot from responding if a prompt contains sensitive data (credit card numbers, SSNs, custom org-defined terms)
  • Can prevent sensitive data from being used in web grounding
  • GA as of early 2026
  • Configured via the Microsoft Purview DLP portal or DSPM portal
  • Important: Purview DLP for M365 Copilot does NOT apply to custom engine agents in Copilot Studio — those are governed by Power Platform DLP

Encryption, Lockbox, and Audit

Encryption and Key Management

  • Data at rest and in transit encrypted using Microsoft-managed keys by default
  • Customer-Managed Encryption Keys (CMK) available for Copilot Studio environments — configure via Power Platform admin center
  • Customer Lockbox supported for Microsoft support access to customer data

Sensitivity Labels

  • Sensitivity labels from SharePoint sources are surfaced to users in the agent chat
  • The highest label applied to sources used in a given response is displayed
  • Confidential or Highly Confidential labeled files are excluded from grounding

Audit Logging

CapabilityTool
Maker activity (who created/edited)Microsoft Purview audit logs
Agent activity monitoring and alertsMicrosoft Sentinel integration
Agent usage analytics (credits, sessions, topics)Power Platform admin center
End-user activity auditingMicrosoft Purview (extended 2025)
AI observability across all agentsMicrosoft Purview DSPM (preview)

Admin Controls

Control AreaAdmin ToolCapability
Agent creation governance Power Platform admin center Environment routing, maker welcome messages, data policies
DLP enforcement Power Platform admin center Block connectors, channels, triggers, HTTP requests, auth types
Credit allocation and per-agent caps Power Platform admin center Assign credits to environments; set caps to prevent runaway consumption
Publishing governance (M365 Copilot agents) M365 admin center Control which agents appear in M365 Copilot; approve or reject agents
Cross-region AI data movement Power Platform admin center Enable or disable data movement for AI features per environment
Encryption Power Platform admin center CMK configuration per environment
Agent inventory Power Platform admin center Extended view of all agents across the tenant

Data Residency — 18 Supported Azure Regions

Data is stored in the Azure region closest to the tenant's home geography. If a tenant's home geography is not listed, data defaults to the United States.

RegionAzure Datacenters
United StatesEast US, South Central US, West US 2
EuropeWest Europe (NL), North Europe (IE)
United KingdomUK South (London), UK West (Cardiff)
AustraliaAustralia East (NSW), Southeast (VIC)
CanadaCanada Central (Toronto), East (Quebec)
GermanyGermany North (Berlin), West Central (Frankfurt)
FranceFrance Central (Paris), South (Marseille)
JapanJapan East (Tokyo), Japan West (Osaka)
IndiaCentral India (Pune), South India (Chennai)

Data Residency Notes

  • Cross-region exceptions: Brazil South data may replicate to South Central US for disaster recovery
  • Global services (Entra ID, MFA, admin center) may store metadata globally
  • Historical conversation logs stored in the geographic region of the end user's Exchange mailbox
  • Generative AI features may require data movement to regions with model hosting capacity — admins can disable this in Power Platform admin center (may limit AI feature availability)
  • In-country processing available for M365 Copilot in Australia, UK, India, and Japan as of November 2025

Good Habits for Copilot Studio Builders

These patterns separate agents that work reliably in production from those that disappoint users or run up bills unexpectedly.

📌 Core Design Patterns

Router + Atomic Topic Pattern
Create a routing topic that classifies user intent and redirects to small, single-purpose atomic topics. Each atomic topic handles exactly one task. This makes maintenance, updates, and debugging dramatically easier than one massive topic.
Start Declarative, Upgrade When Needed
Build declarative agents first for internal use cases — they are faster to create, inherit M365 compliance automatically, and consume no credits for M365 Copilot licensed users. Move to custom engine agents only when you need features declarative cannot provide: external channels, autonomous triggers, custom models, or multi-agent orchestration.

When to Use Declarative vs. Custom Engine

ScenarioUse DeclarativeUse Custom Engine
Internal employee Q&A (HR, IT, policies) ✓ Yes — simple, fast, zero-rated with M365 Copilot Only if you need a custom model or autonomous triggers
M365 Copilot Chat extension ✓ Yes — built for this use case If complex multi-step orchestration is required
External customer-facing website bot Not supported — declarative agents run in M365 only ✓ Yes — deploy WebChat or iframe to any site
WhatsApp or Facebook Messenger Not supported ✓ Yes — via Azure Bot Service channels
Autonomous / scheduled processing (no user present) Not supported ✓ Yes — 25 credits per autonomous trigger
Custom fine-tuned or third-party AI model Not supported ✓ Yes — bring any Azure AI Foundry model
Multi-agent orchestration (A2A) Not supported ✓ Yes — delegate to specialized sub-agents

Practical Tips

Content Hygiene First
Out-of-date, duplicated, or contradictory documents in your knowledge sources cause poor responses. Hold the business accountable for AI-ready content before connecting sources.
Test Before Publishing
Test thoroughly in the embedded panel — test chat interactions do not consume credits. Deploy to yourself with "Show to only me" in Teams before rolling out. Test on Teams mobile and desktop separately.
Set Per-Agent Credit Caps
Set per-agent credit caps in Power Platform admin center under Licensing > Copilot Studio > Manage Agents to prevent runaway consumption. Pair capacity packs with PAYG as an overflow safety net to avoid service disruption at 125%.
Modular Flows
Keep topics modular — one topic per task. Use redirect nodes to compose complex flows from atomic components. Use Power Automate or agent flows for multi-system orchestrations rather than chaining actions within a topic.
Debug with OnKnowledgeRequested
Enable the OnKnowledgeRequested trigger (YAML only as of late 2025) during development to see the actual search query the agent rewrites. Disable it before production deployment.
Configure Inactivity Triggers
Set up an inactivity trigger (e.g., 15 minutes) that clears variables and session state. In Teams, this prevents context overflow and stale connector authentication tokens from causing silent failures.

Known Limitations to Plan Around

SharePoint Grounding Failures

  • Fails "List all documents where the status is Active" — retrieval is semantic, not exhaustive iteration
  • Fails "What does the file named Policy-2024.pdf say?" — name-based queries are not supported
  • Fails Policy conflict analysis across many documents — 25-doc cap makes multi-document comparison unreliable
  • Fails Large SharePoint lists over 2,048 rows — data beyond 2,048 is silently excluded
  • Workaround For large-scale RAG over big document corpora, use Azure AI Search + Azure OpenAI instead of native SharePoint grounding

Custom Engine Agents in M365 Copilot

  • Not supported Users cannot upload files in agent chats; agents cannot return files for download
  • Not supported HTML in agent response messages
  • Not supported Custom engine agents in Outlook, Word, Excel, PowerPoint, or Microsoft Edge
  • Limitation Adaptive Cards refreshed via Action.Execute lose updated content when chat is reopened
  • Limitation The agent cannot access Copilot conversation history that occurred before the user invoked it

Autonomous Agent Cost Risk

  • Autonomous triggers cost 25 credits each, always — even for M365 Copilot licensed users
  • Estimate monthly trigger volume before deploying to avoid unexpected bills
  • Reasoning model (o1) responses cost 100 credits per 10 responses — avoid unless chain-of-thought is essential

Recommended Publishing Workflow

  1. Author and test thoroughly in the embedded test chat panel (no credits consumed)
  2. Publish to the internal demo website URL Copilot Studio generates automatically for stakeholder review
  3. For Teams: deploy with "Show to only me" first; verify connector auth, session timeout, and inactivity trigger
  4. Add a version string to the Greeting topic so users and support can confirm which version is running
  5. Test real-world scenarios: token expiry, session timeout, Teams mobile, Teams desktop
  6. Configure inactivity and error handling topics
  7. Enable "Force newest version" on every Teams publish so users get updates immediately
  8. Publish to production channel; monitor via Power Platform admin center analytics
🎓 Copilot Studio Resources

Official Microsoft documentation, learning paths, and tutorials for building agents in Copilot Studio.

Microsoft Learn -- Official 📄 Documentation · Free · Reference Microsoft Copilot Studio Official Documentation The complete official Microsoft documentation hub for Copilot Studio -- covering agent creation, knowledge sources, deployment channels, DLP policies, licensing, and administration. Read Documentation → Microsoft Learn -- Official 🎓 Video Series · Free · Self-paced Copilot Studio Agent Academy A comprehensive official Microsoft video series covering how to master building agents in Copilot Studio -- produced by the Power Platform Advocacy team. Start Academy → Microsoft Learn -- Official 🎓 Learning Module · Free · Self-paced Get Started with Microsoft Copilot Studio The official Microsoft Learn introductory module -- covers creating agents, testing, working with topics, generative AI, publishing, and analyzing performance. Start Module → Microsoft Learn -- Official 📄 Quickstart Guide · Free Quickstart: Create and Deploy Your First Agent Microsoft's official step-by-step quickstart for building your first Copilot Studio agent -- add knowledge, test in real time, and publish to a demo website you can share. View Quickstart → Microsoft Learn -- Official 📄 Guidance Docs · Free · Best Practices Copilot Studio Guidance Documentation Microsoft's official architectural guidance and best practice patterns for Copilot Studio -- covers design decisions, performance, scalability, and production deployment patterns. Read Guidance → YouTube -- Microsoft Official 📹 Video Playlist · Free · Multi-level Microsoft Copilot Studio Video Playlist Official Microsoft YouTube playlist of Copilot Studio tutorials, feature walkthroughs, and demo videos -- covers beginner to advanced topics across agent building, deployment, and governance. Watch Playlist →

Resources are provided for informational purposes. Always verify content is current before sharing.

This guide is for training and orientation only. Features, limits, and pricing may change as Microsoft updates Copilot Studio. Verify current information at learn.microsoft.com/microsoft-copilot-studio and the official pricing page. Always follow your organization's data security and acceptable use policies.