This article answers common questions about Microsoft Discovery. For procedures and specifications that can change over time, the answers link to the relevant source article in the Microsoft Discovery documentation.
Overview and product choices
What is Microsoft Discovery?
Microsoft Discovery is an enterprise platform for scientific research and R&D workflows. It brings together agentic orchestration, advanced reasoning, a graph-based knowledge foundation, and high-performance computing. For an introduction, see What is Microsoft Discovery?.
Who is Microsoft Discovery designed for?
Microsoft Discovery supports scientists who generate insights, developers who build and extend capabilities, administrators and IT teams who operate the platform, and organizational leaders responsible for research outcomes. See What is Microsoft Discovery?.
What is the difference between Microsoft Discovery and the Microsoft Discovery app?
Microsoft Discovery is the Azure-hosted enterprise platform. The Microsoft Discovery app is a free, local experience that lets individual researchers begin AI-assisted discovery without deploying the Azure platform. For a comparison, see Microsoft Discovery and the Microsoft Discovery app.
What is Microsoft Discovery Studio?
Microsoft Discovery Studio is the web-based research environment for Microsoft Discovery. It is used to work with workspaces, projects, agents, shared sessions, knowledge bases, tools, and data. See What is Microsoft Discovery Studio?.
Is the Discovery Studio portal public or private?
The Discovery Studio portal UI is publicly accessible over the internet, similar to the Azure portal, and is secured with Microsoft Entra ID authentication. Backend data isn't exposed through public endpoints. See What is Microsoft Discovery Studio? and Network security overview.
What scenarios are supported by Microsoft Discovery?
Microsoft Discovery supports scientific and engineering workflows that combine organizational knowledge, data, models, computation, agents, and human review. For representative scenarios, see Key scenarios.
Is Microsoft Discovery primarily an offline optimization and prediction platform, or does it directly control physical scientific or manufacturing equipment in real time?
Microsoft Discovery is an online platform with human oversight. Researchers can review, edit, and initiate runs as needed. Direct, real-time, closed-loop control of physical equipment is not documented as a generally available platform capability. See Platform Card for Microsoft Discovery for intended uses and limitations.
Core concepts and architecture
What are the main Microsoft Discovery resources?
The main concepts include workspaces, projects, shared sessions, agents, tasks, tools, models, Bookshelves and knowledge bases, storage containers and assets, supercomputers, and node pools. Start with Service architecture overview and Microsoft Discovery projects and shared sessions.
What is the difference between a workspace and a project?
A workspace is the top-level environment that brings together platform resources. A project is an access and organizational boundary within a workspace for a specific body of research. See Microsoft Discovery projects and shared sessions.
What is a shared session?
A shared session is the collaborative research context in which people, agents, tasks, files, and execution history are organized for an investigation. See Microsoft Discovery projects and shared sessions.
What is a task?
A task is a unit of work within a shared session. Tasks can include dependencies, status, comments, execution history, and results. See Tasks and shared sessions.
What Azure resources are created when a Microsoft Discovery workspace is provisioned?
Provisioning a workspace creates a dedicated managed resource group in your subscription for backend resources. Other system-managed resource groups can also appear. Bookshelves and supercomputers are provisioned separately and have their own managed resource groups. See Service architecture overview.
In whose subscription are the resources deployed?
Microsoft Discovery resources and their supporting Azure infrastructure are deployed in the customer's Azure subscription, not a Microsoft-managed subscription. See Service architecture overview.
Is Azure Container Registry required?
No. Azure Container Registry is optional and is needed when you bring container images for custom tools that run on the supercomputer. See Microsoft Discovery tools and model integration.
Getting access and preparing Azure
What must be prepared before deploying Microsoft Discovery?
Prepare an Azure subscription, register the required resource providers, assign the required roles, reserve required service and model quota, and validate regional and networking requirements. Start with Quickstart: Deploy Microsoft Discovery infrastructure, Resource provider registration, and Quota reservations.
How do I register the Azure resource providers that Microsoft Discovery needs?
Use the supported registration methods described in Microsoft Discovery resource provider registration. The article covers the required permissions and registration through the Azure portal, Azure CLI, Azure PowerShell, and REST API.
What quota does Microsoft Discovery require?
Microsoft Discovery requires quota across multiple Azure services, including the models and compute used by the deployment. Requirements depend on the resources you deploy. Review the current requirements in Quota reservations for Microsoft Discovery.
Which regions support Microsoft Discovery?
Region availability and cross-region deployment guidance can change. Review the current region information in Service architecture overview and the deployment guidance in Quickstart: Deploy Microsoft Discovery infrastructure.
What is the Discovery Toolbox, and how do I get it?
The Discovery Toolbox is a Visual Studio Code extension for setting up, validating, managing, monitoring, and removing Microsoft Discovery infrastructure. Installation and release information is maintained in the public Discovery Toolbox directory.
How do I grant users the roles they need for Microsoft Discovery?
Assign the platform, project, and supporting Azure resource roles required for each user persona. Follow Role assignments in Microsoft Discovery and Configure project-level access.
Deployment and lifecycle management
How do I deploy Microsoft Discovery?
Follow Quickstart: Deploy Microsoft Discovery infrastructure. It is the source of truth for current prerequisites, deployment inputs, and validation steps.
Can I manage a workspace after it is created?
Yes. Workspace operations and currently supported configuration changes are documented in Manage workspaces in Microsoft Discovery.
How do I manage supercomputers and node pools?
Use the Azure portal procedures in Manage supercomputers and node pools.
Does Microsoft Discovery provide built-in backup or disaster recovery?
Microsoft Discovery doesn't currently provide built-in automatic failover or disaster recovery. If your organization requires additional resiliency, design replication and backup for the applicable stateful resources and consider a multi-region deployment. See Business continuity and disaster recovery.
How do I delete a Microsoft Discovery deployment?
Discovery resources have parent-child and cross-resource dependencies and must be removed in the correct order. Use the current procedure and supported deletion utility described in Delete Microsoft Discovery resources instead of relying on a copied deletion sequence in this FAQ.
Why can't I delete a Discovery resource or resource group?
A dependent child resource, resource link, subnet delegation, private endpoint connection, or workspace-to-supercomputer association can block deletion. Follow Delete Microsoft Discovery resources for the current dependency order and remediation guidance.
Identity and access management
How does Microsoft Discovery control access?
Microsoft Discovery uses Azure role-based access control for platform, project, and resource access. See Role assignments in Microsoft Discovery.
Can I grant access to one project without granting access to the entire workspace?
Yes. Project-level access control makes a project an access boundary within a shared workspace. See Project-level access control and Configure project-level access.
How do project roles interact with shared resources?
Project access and access to shared resources are evaluated separately. A user can require supplementary permissions to use shared tools, models, storage, or knowledge resources. See Project-level access control.
Why does Microsoft Discovery use managed identities?
Microsoft Discovery uses user-assigned managed identities to authenticate to Azure resources on behalf of workspaces, supercomputers, and Bookshelves. See Configure managed identities for Microsoft Discovery.
How do I configure managed identities?
Follow Configure managed identities for Microsoft Discovery for current creation, association, and Azure role-assignment requirements.
Networking, security, and compliance
Are Discovery services exposed publicly or privately?
Most backend resources use private connectivity controls such as private endpoints, Network Security Perimeter, and virtual network integration. Review the current design in Network security overview.
How do I turn on network isolation for a workspace?
Configure network isolation during workspace deployment or update by following Configure network security. The article is the source of truth for current properties and prerequisites.
What subnets does a hardened deployment need?
A hardened deployment uses dedicated networking for applicable workspace, agent, private endpoint, Bookshelf, and supercomputer components. Required subnet roles and sizing can change, so use End-to-end network-hardened deployment and Configure network security.
Do I need to allow inbound access from Microsoft into my environment?
No explicit customer-configured inbound access from Microsoft is required for the architecture described in the Discovery network-security guidance. See Network security overview.
Does Microsoft Discovery support customer-managed keys?
Yes, for the resources and scenarios documented by the service. See Configure customer-managed keys for current scope and prerequisites.
Where can I find compliance guidance?
Start with Security and compliance overview, Compliance guidance, and the SIG-based security and compliance FAQ.
Agents, models, and Discovery Engine
How are Discovery agents implemented?
Discovery agents are configured as Microsoft Discovery resources and use the platform's agent and workflow infrastructure to perform assigned work. For the current architecture and supported agent types, see Discovery Agent concepts and Discovery Agent types.
How do I create an agent?
Follow Create agents for current prerequisites, configuration fields, and supported creation paths.
Which models does Microsoft Discovery require?
Required models and quota can vary across Discovery Engine, agents, and Bookshelf. Review the current model and quota requirements in Quota reservations for Microsoft Discovery.
Can I choose which models my agents use?
Yes, within the models and configurations supported by Microsoft Discovery. See Select models for agents.
Where can custom models run?
Documented integration patterns include models hosted in Azure Machine Learning and containerized models run through supercomputer-backed tools. See Microsoft Discovery tools and model integration.
What is the Discovery Engine?
The Discovery Engine is the cognitive orchestration layer that plans research work, delegates tasks, monitors progress, and adapts execution. See Discovery Engine overview.
What is cognition?
Cognition is the Discovery Engine capability that reasons over the research objective, available agents, tasks, evidence, validation requirements, and execution state. See Cognition overview.
How much autonomy does the Discovery Engine have?
The level of autonomy depends on how the shared session, tasks, and validation requirements are configured. Researchers remain responsible for setting direction and reviewing results. See Trust relationship and basic shared session patterns.
How do I run my first agent and investigation?
How do I build a shared session that uses cognition?
Tools, compute, and integrations
What is the supercomputer in Microsoft Discovery?
The supercomputer is the compute abstraction used by Microsoft Discovery for applicable tool, model, and knowledge-processing workloads. Its current implementation and management model are described in Manage supercomputers and node pools.
Which VM families can I use for supercomputer node pools?
Supported VM families and node-pool requirements can change by region and release. Use Manage supercomputers and node pools and Quota reservations as the current sources of truth.
What kinds of tools can agents use?
Microsoft Discovery supports the tool types and execution patterns described in Microsoft Discovery tools and model integration.
How should I plan a custom tool?
Define the tool's function, compute requirements, dependencies, packaging, and integration pattern before publishing it. See Plan tool requirements.
How do tools authenticate to Azure resources?
Use the identity and role-assignment patterns supported for the applicable Discovery resource and tool configuration. See Configure managed identities and Microsoft Discovery tools and model integration.
How do I submit jobs directly to the supercomputer?
Use the supported supercomputer command-line utility described in the public Supercomputer CLI directory. Refer to its current README for prerequisites and command syntax.
Where can I find utilities to operate a Microsoft Discovery deployment?
The public Microsoft Discovery utilities directory contains the currently published operator utilities and their documentation, including setup and lifecycle tools.
Data, knowledge, and research artifacts
What are storage containers and storage assets?
A storage container represents a connection to storage used by a workspace, while storage assets identify data that agents can use. See Storage containers and storage assets.
What is a Bookshelf knowledge base?
A Bookshelf knowledge base organizes and indexes a curated corpus so that researchers and agents can query connected scientific knowledge. See Bookshelf and Knowledge Bases.
What is the difference between Bookshelf and Azure AI Search?
Bookshelf is designed for graph-based, thematic, and multi-hop reasoning across a curated corpus. Azure AI Search is suited to scalable indexing and precise retrieval. Use the comparison and limits in Bookshelf and Knowledge Bases when selecting an approach.
Which retrieval approach should I use?
Use the retrieval approach that matches the question and corpus. Bookshelf is intended for connected, discovery-oriented reasoning, while search indexing is appropriate for targeted retrieval at larger scale. See Bookshelf and Knowledge Bases.
Can Bookshelf handle a 100-GB corpus directly?
Bookshelf is intended for a curated corpus rather than indiscriminate ingestion of a very large repository. Review the current size guidance and limitations in Bookshelf and Knowledge Bases.
What document types does Bookshelf support?
Supported document types and ingestion limitations can change. Use Create and index a Bookshelf Knowledge Base as the source of truth.
Do scientists always need an agent to query a Bookshelf?
No. A Bookshelf can be queried directly, and it can also be used as grounding within agent-driven research. See Bookshelf and Knowledge Bases.
Where are files generated by agents stored?
Files produced during shared-session tasks are represented as research artifacts and can be made available to related tasks according to the documented file and inheritance model. See Files and storage assets.
How does data move between tools and agents?
Data handling depends on the configured storage assets, tools, task artifacts, and agent instructions. See Data handling with tools and agents and Files and storage assets.
Studio and research workflows
How do I access Discovery Studio?
Access and browser requirements are documented in What is Microsoft Discovery Studio?. Your account must also have the required Discovery and Azure permissions.
How do researchers collaborate in Discovery?
Researchers collaborate through projects and shared sessions that organize tasks, comments, files, execution history, agents, and results. See Microsoft Discovery projects and shared sessions.
How are task results and execution history preserved?
Tasks maintain status, comments, execution attempts, and results within the shared-session model. See Tasks and shared sessions.
Can I use GitHub Copilot in Discovery Studio?
Yes, where the documented preview and workspace configuration requirements are met. See Use GitHub Copilot in Microsoft Discovery.
How should I write prompts for Discovery agents?
Use clear objectives, constraints, expected outputs, relevant context, and validation criteria. See Write effective prompts for agents.
Operations, monitoring, and troubleshooting
What observability does Microsoft Discovery provide?
Microsoft Discovery integrates with Azure Monitor and provides platform-specific logging for applicable workspace, supercomputer, and control-plane operations. See Observability in Microsoft Discovery.
How do I view Azure activity logs for Discovery resources?
How do I query workspace logs?
Use the Log Analytics workspace associated with the workspace's managed resource group and follow Query workspace logs.
How do I query supercomputer logs?
Use the applicable Log Analytics workspace and the queries documented in Query supercomputer logs.
How do I enable audit logging?
Configure Azure Monitor diagnostic settings as described in Enable audit logging for Microsoft Discovery resources.
How do I trace a failed request?
Use the correlation ID and the relevant workspace, supercomputer, or activity logs. See Observability in Microsoft Discovery, Query workspace logs, and Query supercomputer logs.
What information should I collect before requesting support?
Collect the affected resource IDs, operation time, correlation ID, error text, deployment region, and relevant activity or application logs. Use Observability in Microsoft Discovery to locate the available diagnostics.
Billing and cost management
Who pays for the Azure infrastructure?
The customer pays for the supporting Azure resources deployed in the customer's subscription. See Microsoft Discovery billing overview.
How is Microsoft Discovery billed?
The Discovery app and Azure-hosted Microsoft Discovery services have different pricing models. Billable service operations and supporting Azure infrastructure charges are described in Microsoft Discovery billing overview.
Is the Microsoft Discovery app free?
The Discovery app is described as a free local edition. Check Microsoft Discovery and the Microsoft Discovery app and Microsoft Discovery billing overview for current terms.
Which resources contribute to cost?
Costs can include the Microsoft Discovery service and the Azure resources used for models, storage, databases, networking, monitoring, and compute. See Microsoft Discovery billing overview.
How can I monitor deployment cost?
Use Azure Cost Management for resources in the customer subscription and the cost capabilities documented for the Discovery tooling. See Microsoft Discovery billing overview and the public Microsoft Discovery GitHub repository.
Responsible AI and supported use
Is Microsoft Discovery intended to replace scientific review?
No. Microsoft Discovery is designed to support scientific work with human oversight. Researchers remain responsible for reviewing evidence, validating outputs, and deciding how results are used. See Platform Card for Microsoft Discovery.
What are the intended uses and limitations of Microsoft Discovery?
The intended uses, limitations, evaluations, safety components, and responsible-use guidance are documented in Platform Card for Microsoft Discovery.
How should users validate agent-generated results?
Define validation requirements, review evidence and artifacts, and calibrate autonomy to the risk and purpose of the research. See Trust relationship and basic shared session patterns and Platform Card for Microsoft Discovery.
Does Microsoft Discovery guarantee that generated conclusions are correct?
No. Agent and model outputs require human review and validation. See Platform Card for Microsoft Discovery for documented limitations and responsible-use guidance.
APIs, automation, and developer reference
Can Microsoft Discovery resources be managed programmatically?
Yes. Microsoft Discovery provides REST API reference and Azure CLI coverage for supported resource-management operations. See the Microsoft Discovery REST API reference and Microsoft Discovery Azure CLI reference.
Can I automate deployment with infrastructure as code?
Use the deployment templates and automation maintained in the supported Microsoft Discovery documentation and public repository. Start with Quickstart: Deploy Microsoft Discovery infrastructure and the Microsoft Discovery GitHub repository.
Where are Microsoft Discovery API versions documented?
Use the Microsoft Discovery REST API reference for supported operations, schemas, and API versions.
Where can I find command-line reference?
See the Microsoft Discovery Azure CLI reference and the operator utilities in the Microsoft Discovery GitHub repository.