Multi-Tenancy

Workspaces

Organize your work with complete data isolation for teams, projects, and organizations

What are Workspaces?

Workspaces provide complete isolation for your data, AI providers, and team members. Each workspace operates independently with its own:

Libraries - Document collections and semantic search
Assistants - AI chatbots configured for specific tasks
Lists - Structured data with AI-powered enrichments
AI Providers - Ollama and OpenAI configurations
Team Members - Role-based access control (Owner, Admin, Member)

Data Isolation

Zero cross-workspace access

Team Collaboration

Role-based permissions

AI Provider Scoping

Configure per workspace

Common Use Cases

Departments

Separate workspaces for HR, Engineering, Sales, Marketing - each with their own AI providers and data.

Client Projects

Dedicated workspace per client for complete data isolation, billing separation, and custom AI configurations.

Environment Separation

Development, staging, and production workspaces with different AI provider configurations and access controls.

Regional Data

Separate workspaces for different geographic regions to comply with data residency requirements.

Workspace Switcher

The workspace switcher is located in the top navigation bar, allowing you to seamlessly switch between workspaces.

When you switch workspaces, you'll see only the resources (libraries, assistants, lists) that belong to that workspace.

How to Switch Workspaces

Click the workspace switcher button

Located in the top navigation bar, shows your current workspace name

Select a workspace from the dropdown

You'll see all workspaces where you're a member

View updated content

The page automatically updates to show the selected workspace's resources

Important: Switching workspaces while on a detail page (e.g., viewing a specific library or assistant) will navigate you back to the list view to prevent access errors.

Default "Shared" Workspace

When you first install George AI, a default workspace named "Shared" is automatically created. All existing data is migrated to this workspace.

Migration Details

All existing libraries, assistants, lists, and files are moved to the "Shared" workspace
The first admin user becomes the workspace owner
Migration happens automatically on first startup after upgrade
Zero downtime - your data remains accessible throughout the process
Rollback safety - migration can be reversed if needed

The "Shared" workspace can be renamed or deleted like any other workspace. However, you must have at least one workspace to use George AI.

Workspace-Scoped AI Providers

Each workspace can configure its own AI providers (Ollama, OpenAI), giving you complete flexibility in how you use AI models across different teams or projects.

Benefits

Use Ollama for privacy-sensitive workspaces
Use OpenAI for performance-critical workspaces
Mix providers based on budget constraints
Test new models in development workspace first
Share physical infrastructure (same Ollama instance)

Configuration

Admin users can manage providers
Configure via /admin/ai-services UI
Test connections before saving
API keys never exposed to frontend
60-second cache for optimal performance

Configuring Providers for a Workspace

Switch to the target workspace

Use the workspace switcher to select the workspace you want to configure

Navigate to AI Services admin page

Open Settings menu → Admin → AI Services (requires Admin or Owner role)

Add or configure providers

Click "Add Provider" and enter connection details (Base URL, API Key)

Test the connection

Use the "Test Connection" button to verify configuration before saving

Sync models

Navigate to Admin → AI Models and click "Sync Models" to discover available models

Multiple workspaces can share the same physical AI infrastructure (e.g., same Ollama instance at http://ollama:11434) while maintaining logical separation.

For detailed provider configuration instructions, see AI Services Administration.

Automatic Model Filtering

When you select AI models for libraries, assistants, or list fields, George AI automatically shows only the models available in your current workspace.

How It Works

1. You select a model in any dropdown

Library embedding models, assistant language models, list enrichment models, etc.

2. George AI filters by workspace providers

Only models from enabled providers in the current workspace are shown

3. Model list updates automatically

When you switch workspaces, available models update in real-time

Workspace with OpenAI

Users see: GPT-4, GPT-3.5, text-embedding-3-large, etc.

Workspace with Ollama

Users see: llama3, mistral, nomic-embed-text, etc.

Zero Configuration: Model filtering happens automatically based on workspace providers. No manual configuration required!

Workspace Roles & Permissions

Each workspace member has a role that determines their permissions within that workspace.

Role	Manage Resources	Manage AI Providers	Manage Members	Delete Workspace
Owner	✓	✓	✓	✓
Admin	✓	✓	✓	✗
Member	✓	✗	✗	✗

Note: System admins (configured via environment variables) have full access to all workspaces regardless of role.

Best Practices

Naming Conventions

Use clear, descriptive workspace names that indicate their purpose (e.g., "Engineering-Dev", "Client-Acme-Corp", "Marketing-Production").

Provider Strategy

Use Ollama for sensitive data, OpenAI for performance. Consider cost implications when configuring providers for each workspace.

Access Control

Assign roles carefully. Only give Admin/Owner roles to users who need to manage providers and members. Regular users should be Members.

Testing New Models

Create a dedicated "Testing" or "Sandbox" workspace to experiment with new models and configurations before rolling them out to production workspaces.

Resource Organization

Keep related libraries, assistants, and lists in the same workspace. Avoid duplicating data across workspaces unless truly needed for isolation.

Regular Model Syncing

Run "Sync Models" periodically to discover new models added to your providers. This ensures users always see the latest available models.