Skip to main content
Version: Cloud

Get started with Deeploy

TL;DR Immediately get your hands dirty by following our technical quickstart or governance quickstart guides.

The documentation is organized to support both new and experienced users, whether you're interested in the technical workflow or the governance workflow. On the home page, you'll find quick links to the main jobs-to-be-done for each workflow. This article outlines the hierarchy and core concepts that underpin both.

Deeploy is organized in the following way:

  • Organizations
    • Workspaces
      • Use Cases
        • Deployments

Organization​

An organization is the highest level in Deeploy. Within an organization, you can have multiple Workspaces to separate different projects or departments. On the organization level, you'll find:

  • Governance dashboard: Govern all AI systems used within the organization
  • AI registry*: Overview of all AI Deployments with actionable insights
  • Use case registry: Overview of all use cases in the organization with actionable insights
  • Control frameworks: Define and manage control frameworks that can be applied to use cases
  • Approval rules: Define approval rules for deploying AI systems
  • Periodic reviews: Set up and manage periodic reviews for AI use cases to ensure up-to-date controls
  • Administrative settings: Manage organization settings, billing information, user roles, and integrations
  • Audit Logs: Track all activities for compliance and security purposes
  • Organization documents: Manage and store important documents related to the usage of AI

Workspace​

A Workspace is a dedicated space within an organization for teams to collaborate on specific projects or departments. Each Workspace can include multiple use cases and Deployments. Features that apply across use cases and Deployments can be configured at the Workspace level. In a Workspace, you'll find (+ relevance for use case or deployment):

ConceptUse caseDeployment
Integrations: Configure integrations and webhooks with external tools and platforms.✅✅
Documentation templates: Create and manage Q&A style templates for documentation.✅
Repositories: Manage code repositories to source and version model artefacts and metadata for Deployments.✅
Guardrails: Define and manage custom guardrails to be applied and reused across all types of Deployments.✅
Job schedules: Schedule periodic cron-style jobs for scheduled managed Deployment calls.✅
Credentials*: Manage credentials and access tokens for integrations with external tools and platforms.✅
Environment variables: Manage environment variables for use in managed Deployments. Check here which env vars are supported for the default frameworks✅

Use case​

A Use case represents a specific application of an AI model. It encompasses the entire lifecycle of the AI system. In a use case, you'll find

  • Risk classification: Evaluate the risk associated with the use case based on various factors such as data sensitivity, model complexity, and potential impact.
  • Control framework compliance: Apply predefined control frameworks to ensure compliance and governance for the use case.

Deployment​

A Deployment is the actual implementation of an AI model in a production environment. It is associated with a specific use case and can be monitored and managed independently. We distinguish between three types of deployments:

  • Registration Deployment: Register an existing deployment to monitor and govern it within Deeploy.
  • External Deployment: Use an external API with Deeploy as a API gateway to monitor and govern the Deployment.
  • Managed Deployment: Use a pre-trained model artefact, manage the AI model directly within Deeploy using pre-built images.

Important concepts for Deployments and relevance per Deployment type:

ConceptRegistration DeploymentExternal DeploymentManaged Deployment
Monitoring, tracing & alerting: Track the heatlh, performance and human evaluations of deployed AI models. Additionally, custom metrics can be added.✅✅✅
Model artefact store: Manage and version artifacts and metadata with Git-based version control systems (e.g. GitHub, GitLab) or connect specialized model stores like MLFlow, Unity Catalog, and Azure model registry to manage and version artifacts.✅✅✅
State management: Track the Deployment's state and approve new versions.✅✅✅
Event logging: Maintain a comprehensive log of all events and actions.✅✅✅
Guardrailing: Implement custom guardrails to ensure that AI models operate within predefined boundaries. This includes input validation and output constraints.✅✅
Deployment service: Manage Deployments in Kubernetes (Kserve), AzureML, and Sagemaker.✅
Explainability: Built-in global and local explainability frameworks that can be inferenced in real-time so that they can be used in human-in-the-loop scenarios.✅
Supported frameworks: Pre-built images for model and explainer frameworks that pull images from the artifact source.✅

Best way to get your hands dirty​

Looking for something else?​

Need support?​