Skip to main content
Admin › Settings › Connections
To use AI models, first configure connections to LLM providers. Connect multiple providers simultaneously so users can pick from various models.
Connections settings tab

Configure LLM providers, cloud accounts, and storage in Admin > Settings > Connections


AI Provider

Manage connections to all LLM providers (OpenAI, Azure OpenAI, Azure AI Foundry, Vertex AI, etc.) in a single AI Provider section. Each connection is listed under AI Provider Connection Management, and you can configure multiple connections at once to balance load or expose a variety of models.
1

Add connection

Click the + button in AI Provider Connection Management and select the provider type.
2

Enter connection info

In the connection modal, enter API Base URL and API Key.
3

Verify connection

After saving, verify the model list auto-refreshes.

Common Connection Settings

Each connection has these common features:
Adding multiple connections to the same provider auto-merges model lists. For example, adding 2 Azure OpenAI connections (US/Korea regions) shows models from both regions. Use Prefix ID to distinguish.

Ollama API Connection

Connect to a local or internal server Ollama instance. Add multiple servers to load-balance. Ollama provides complete data privacy without external network.

Direct Connections

Beyond admin-configured connections, allow users to connect directly to LLMs with personal API keys. When enabled, users can add OpenAI-compatible APIs in Personal Settings > Connections.
Direct connections support OpenAI-compatible APIs only (Ollama unsupported). The user’s API server must have CORS correctly configured.

Cloud Accounts

Configure global cloud service accounts. Service Account Keys set here serve as fallback when individual features (TTS, STT, images) lack separate keys.
Leaving Service Account Key empty uses Application Default Credentials (ADC).

Google Workspace Integration

Allow users to use Gmail sending, calendar event creation, and Drive document creation from chat. Each capability is controlled by an individual toggle, and actual use requires both a per-user Google OAuth connection and group permissions. For safety, all defaults are disabled.
Gmail, Calendar, and Drive fall under Google’s sensitive and restricted scopes. Complete App verification or CASA review before production deployment for these to work correctly.

Usage flow

Turning on the toggle is only the start — each user must also finish their connection before it works in chat.
1

Attach Google tools in chat

Once connected, Gmail · Google Calendar · Google Drive tools attach as chips in the chat input.
Gmail, Google Calendar, and Google Drive tools attached as chips in the chat input

Google tools attached in the chat input

2

Ask in natural language

Within the attached tool scopes, users ask in natural language and the agent queries or acts on the relevant Google service to respond.
The agent looks up the calendar to answer a question about today's schedule

Calendar lookup Q&A


Image Attachment Mode

Configure how images attached to chats are processed and stored.

Shared Storage (file uploads)

Configure storage location for uploaded files like documents and PDFs. Same storage provider options as Image Attachment Mode.
After changing storage settings, always verify connection state with the Test button.

Models

Activate/deactivate connected models and configure Workspace models

Documents

Embedding engine and RAG pipeline settings

Code Gateway

LLM proxy settings for coding tools