Model Settings — User Guide

Audience: Workspace admins who configure which LLMs are available to users.

Overview

The Model Settings page (/model-settings) is where admins register provider credentials and model IDs. End users choose from configured models in AI Chat; they cannot add new providers themselves.

Supported providers

The add/edit dialog supports these provider values:

Provider Summary
Google Gemini presets and custom model IDs; optional Vertex AI (vertexai), Project ID, and Location.
OpenAI API key and temperature (and related options in the form).
Anthropic API key and temperature — Claude family presets in the model list.
DeepSeek API key and temperature.
Snowflake Cortex Snowflake account, PAT or key-pair auth, optional base URL for Cortex inference endpoints.
Open Source Base URL (Ollama, vLLM, etc.), temperature, max tokens; presets may include Gemma, Llama, Qwen, DeepSeek-class IDs depending on the UI catalog.
Custom Free-text custom provider name plus custom model ID for vendors not listed above.

Preset model IDs in the UI (for example Gemini 1.5/2.x, GPT-4o family, Claude 3.x, DeepSeek chat/coder, a long Snowflake Cortex list, and open-source presets) come from the product’s availableModels mapping—exact labels can change with product updates.

Model cards

Each saved configuration shows:

  • Display name and Default chip when is_default is set
  • Provider / model ID
  • API key status (masked or “configured”)
  • Google-only fields such as project when relevant
  • Pricing (input/output cents per 1K tokens) when usage analytics pricing is configured

Behaviour

Default model

One row can be the default (is_default). That model is pre-selected in the Chat model picker when users start a session.

Pricing and usage

Model pricing (model_pricing) is tied to billing admin configuration: per provider + model_id, input and output cost in cents per 1K tokens, feeding Usage Analytics and chargeback views. If pricing is missing, costs may show as $0.00 until configured.

Adding a model

  1. Click Add Model.
  2. Choose provider, then model from presets or enter a custom ID.
  3. Enter display name, API key (where required), and provider-specific fields (Vertex, Cortex URL, open-source base URL, etc.).
  4. Optionally set Set as default.
  5. Save.

Editing and deleting

Edit any row; use Set Default to switch the workspace default. You cannot delete the current default until another model is default.

Troubleshooting

Symptom Cause Fix
Model shows API key not configured Key never saved Edit and paste the key again
Usage Analytics shows $0.00 Per-model pricing not set Configure input/output pricing for that provider + model
Open-source errors Wrong base URL or model name Confirm vLLM/Ollama URL and model ID
Cortex auth fails PAT or key-pair mismatch Verify account identifier and secrets per Snowflake docs

Related

Last updated: February 21, 2026