Copilot Studio explained

Microsoft Copilot Studio is the low-code platform for building custom Copilots and agents: chat-based AI experiences that can answer questions, run actions, and orchestrate workflows. It's the successor to Power Virtual Agents and the maker surface for extending Microsoft 365 Copilot with your own data and capabilities.

For building well, see Copilot Studio agent design patterns; for distribution and control, Copilot agents and Agent Store governance; for the data layer, Dataverse explained; and for how Studio agents relate to plugins and connectors, Copilot extensions and plugins.

What you can build

Copilot Studio supports several output types:

• Standalone Copilots published as web chatbots, in Microsoft Teams, on a website, on Facebook, Telegram, or via custom channels.
• Microsoft 365 Copilot agents — agents that show up inside the Microsoft 365 Copilot Chat experience, accessible to users in your tenant.
• Embedded agents in Teams apps, Power Apps, or custom apps via SDK.

Each Copilot or agent can:

• Answer questions from grounding sources you provide.
• Take actions through connectors and Power Automate flows.
• Hand off to humans when needed.
• Use AI orchestration to call multiple actions in sequence based on user intent.

Grounding sources

A Copilot grounded only on its own training data isn't very useful for your business. Copilot Studio lets you connect:

• SharePoint sites — files, pages, lists.
• Public websites — crawled at index time or on demand.
• Graph connectors — third-party systems indexed into Microsoft Search.
• Custom data via the knowledge base feature.
• Dataverse tables.
• OneDrive folders.
• Enterprise data through dedicated connectors (Salesforce, ServiceNow, etc.).

Each source's content is retrieval-augmented at query time: the system finds the relevant chunks and feeds them to the model alongside the user's question. Responses cite the source.

Topics and actions

Beyond freeform Q&A, you build topics — defined conversation flows triggered by user intent. A topic has:

• Trigger phrases (or now, natural language understanding that triggers on intent).
• Nodes for messages, questions, conditions, variable assignments.
• Actions that call Power Automate flows, Power Apps, custom Plugins, or HTTP endpoints.

Topics are useful when you need predictable behaviour — a specific workflow rather than an open question.

Generative orchestration

Newer Copilot Studio agents use generative orchestration — instead of defining every topic, you give the agent a set of available actions and let the AI decide which to call based on user intent. This dramatically reduces authoring effort for agents that span many tools.

Microsoft 365 Copilot extension

Agents published as Microsoft 365 Copilot extensions appear in the Copilot Chat experience, branded to your tenant. Users invoke them with @AgentName from inside Copilot. Permissions and data access flow through standard Microsoft 365 identity.

Licensing

Copilot Studio is licensed by message consumption (Copilot Studio Messages, billed in packs) and per-tenant. Microsoft 365 Copilot includes some Copilot Studio capacity for tenants on that licence. Standalone Copilot Studio licences exist for tenants not on Microsoft 365 Copilot.

For organisations rolling out Copilot, Copilot Studio is how you extend the experience from "generic AI assistant" to "AI assistant that knows your business."