Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Microsoft Discovery organizes scientific research through two key concepts: projects and shared sessions. Projects define the functional boundary for your research resources, while shared sessions are where you interact with agents and conduct research. This article explains how both concepts work, how they relate to other platform resources, and best practices for using them.
Projects
A project is the organizational unit within a Microsoft Discovery workspace where you bring together agents, tools, knowledge bases, storage containers, and shared sessions into a single, access-controlled boundary. Every research activity in Microsoft Discovery happens within the context of a project.
Resources a project contains and uses
Projects contain project-owned resources and reference shared resources:
| Resource | Description |
|---|---|
| Agents | Every project includes a default Discovery agent ready to use. You can also create custom prompt agents that execute specific scientific tasks. Agents declare their project affiliation at creation and can't be shared across projects. However, you can clone agents between projects using the add existing agents option. |
| Tools and knowledge bases | Shared resources that exist outside the project scope. Users need the corresponding companion roles to view and select them while creating or updating an agent. A Project Contributor can run a pre-created agent without those reader roles. |
| Shared sessions | Research sessions where you chat with agents, run analyses, and collect insights. |
| Storage containers | Shared Microsoft Discovery resources that reference Azure Blob Storage for project inputs and outputs. They exist outside the project scope and require separate Discovery and Azure Storage role assignments. |
Project and workspace relationship
Projects exist within a workspace, which provides the shared infrastructure layer:
- Workspace provides the networking, supercomputers, managed identities, and chat model deployments that all projects in the workspace share.
- Projects consume workspace-level resources while maintaining their own isolated set of agents, shared sessions, and data.
This separation allows platform administrators to manage infrastructure at the workspace level while scientists work independently within their projects.
Project access boundary
Each project is an Azure RBAC access boundary. A platform administrator can assign Microsoft Discovery Project Contributor (Preview) or Microsoft Discovery Project Reader (Preview) directly to a project without granting access to other projects in the workspace.
Project roles cover project-owned resources such as agents, investigations, shared sessions, conversations, tasks, and Discovery Engine operations. Tools, chat model deployments, storage containers, and bookshelves are shared resources outside the project scope and require separate least-privilege role assignments.
For role inheritance, supplementary roles, and security limitations, see Project-level access control in Microsoft Discovery. For assignment steps, see Configure project-level access.
Creating a project
You create projects in Microsoft Discovery Studio. Each project requires:
- A workspace to host it
- A storage container for input and output data
For step-by-step instructions, see Create a project.
Shared sessions
A shared session is a research session within a project where you interact with agents through natural language conversations. Shared sessions are the primary interface for conducting AI-powered scientific research and collaboration in Microsoft Discovery.
How shared sessions work
Within a shared session, you can:
- Chat with agents by selecting an agent and entering prompts in natural language. Use the agent selector in the chat input box or
@and select to route messages to specific agents. - Run computational analyses where agents invoke tools and knowledge bases, on your behalf to accomplish a given task.
- Collect data-driven insights as agents reason over your knowledge bases, run calculations, and synthesize findings.
- Orchestrate multi-agent research by using the Discovery Engine that coordinates multiple prompt agents across research steps.
Each shared session maintains its own conversation history and context, allowing you to track the progression of a research thread from initial question to final insight.
Creating a shared session
You create shared sessions from the Discovery tab within your project in Discovery Studio, or by simply typing a prompt in the chat box on the Welcome page. For step-by-step instructions, see Create a shared session.
How projects and shared sessions fit in the platform
The following hierarchy shows how projects and shared sessions relate to other Microsoft Discovery resources:
Subscription
└── Resource Group
├── Supercomputer
│ └── Node Pools
├── Bookshelf
│ └── Knowledge Bases
├── Tool
├── Storage Container
│ └── Storage Assets
└── Workspace
├── Chat Model Deployments (shared across projects)
└── Project
├── Agents (prompt and workflow)
│ └── References to approved tools and knowledge bases
└── Shared Sessions
└── Conversations with agents
Best practices
- One project per research initiative — Scope each project to a distinct research initiative or team. This keeps agents, knowledge bases, and data isolated and manageable.
- Use descriptive shared session names — Since shared sessions track the evolution of a research question, use names that reflect the research objective (for example,
Aspirin SynthesisorMolecule Screening). - Leverage workspace-level resources — Chat model deployments and supercomputers are shared at the workspace level. You don't need to create them for each project.
- Plan your agent structure first — Before creating shared sessions, set up the agents your research requires. You can create agents individually or use the default Discovery agent provided with every project.