Private LLM Gateway
Use Shortcut with an organization-controlled LLM gateway for AI requests.
How It Works
Shortcut separates AI requests from normal product operations. The browser or Excel add-in sends AI requests directly to your configured gateway; Shortcut servers do not forward those AI requests.
| Connection | Used for | Examples |
|---|---|---|
| Your LLM gateway | Generating AI responses | User messages, spreadsheet context, selected model ID, tool instructions and results, supported images, and response settings |
| Shortcut services | Running the Shortcut product | Sign-in, organization policy, allowed models, workspace metadata, usage status, and diagnostic metadata |
Supported Gateway APIs
| API type | Request endpoint | Model discovery |
|---|---|---|
| Anthropic Messages | POST /v1/messages | GET /v1/models |
| OpenAI Chat Completions | POST /v1/chat/completions | GET /v1/models |
| OpenAI Responses | POST /v1/responses | GET /v1/models |
Requirements
Gateway
- HTTPS gateway URL
- Model discovery through
GET /v1/models - AI request endpoint for the selected API type
- Streaming responses for interactive agent turns
- Authentication through
x-api-keyorAuthorization
Network
- DNS resolution for the gateway host
- HTTPS access from the browser or Excel add-in
- CORS access from Shortcut web and add-in origins
- Office add-in domain policy access, where applicable
- VPN or firewall access, if the gateway is private
Credentials and Models
- Organization credentials: an admin configures one gateway URL and token for the organization.
- User credentials: an admin configures the gateway policy, and each user provides their own gateway URL and token locally.
If required user credentials are missing, Shortcut does not fall back to Shortcut's default model provider.
Admins choose which gateway model IDs are allowed. Shortcut passes gateway model IDs through as-is and does not call model discovery for every request.
Capabilities
Depending on the configured API type and gateway model, Shortcut can use:
- streaming text responses
- tool use, where the model asks Shortcut to perform an action
- helper requests for features such as suggestions or summaries
- image input
- reasoning or thinking output, when supported by the gateway
- token usage, when returned by the gateway
If a workflow requires something the selected gateway model does not support, Shortcut shows an error instead of switching to Shortcut's default model provider.
Privacy and Diagnostics
Your gateway receives AI request content, including the user's message, spreadsheet or workbook context, selected model ID, tool instructions and results, supported image input, and response settings.
Shortcut services receive product and diagnostic information, including signed-in user and organization identity, the private gateway policy, allowed model metadata, workspace and thread identifiers, request type, platform, and diagnostic metadata.
Shortcut's private gateway usage log is metadata-only. It does not contain user messages, AI response text, tool inputs or outputs, screenshots, images, file contents, or workbook cell contents.
If token usage is provided, Shortcut will record it.
Troubleshooting
| Symptom | What to check |
|---|---|
| Gateway connection fails | DNS, VPN, firewall, HTTPS certificate, and Office add-in domain policy |
| CORS error | CORS headers on OPTIONS, GET, POST, and error responses |
| Authentication fails | Auth header choice, token value, and whether Authorization needs a Bearer prefix |
| No models appear | GET /v1/models response shape and gateway auth |
| Selected model fails | Refresh discovery and update the allowed model list |
| Tool or image workflow fails | Enable a model with the needed tool, image, reasoning, or streaming support |
| User is prompted for credentials | User credential mode requires local gateway credentials |