Skip to main content
Manage the list of models pulled from connected LLM providers. From this tab, control individual model activation, customization, display order, and default model selection. Configure in Admin > Settings > Models.
Model management tab

Model List

Shows Workspace models (admin-created/customized) combined with Base models (system models pulled from connected APIs).
ItemDescription
Model nameModel name shown to users
IDModel identifier used in API calls
Enabled / DisabledToggle (Switch) for whether the model is available

Hide Model (Shift+Click)

Hold Shift and click the icon in a model row to toggle hide/show. Hidden models are managed as meta.hidden and don’t appear in the user’s model picker.
Hidden and Disabled are separate features. Hidden removes from the model picker; Disabled makes the model itself unavailable.

Default Model and Order

Click the settings (gear) icon at the top-right of the model list to open the model settings modal.
SettingDescription
Default modelModel auto-selected when starting a new chat
Model orderDisplay order in the model picker (drag or arrows)

Activate / Deactivate Models

Use per-model Enabled / Disabled toggle (Switch) to expose only needed models to users.
Disabled models don’t appear in the user’s model picker dropdown. Disabling a model that’s set as an agent’s base model doesn’t affect agent execution.

Model Customization

Click the edit (pencil) icon on a model row to customize that model.
SettingDescription
Name/DescriptionChange the user-facing model name and description
System promptSet the default system prompt for this model
Response format promptSet response format guidance
ParametersTemperature, Top P, Top K, Stop tokens, etc.
Resource connectionsConnect Knowledge Bases, DBs, tools, guardrails, glossaries
CapabilitiesEnable web search, image generation
Auto-evaluationAuto-monitor response quality
Access controlExpose only to specific groups/users
Customizing a model creates a Workspace setting record for that model. Unlike agents, this doesn’t directly modify the base model — it adds an override configuration.

Controlling Models Visible to Users

MechanismEffectWhere to Configure
Activate/Deactivate (toggle)Disabled → excluded from model pickerToggle on model row
Hide (Shift+click)Hidden → excluded from model pickerShift+icon on model row or menu
Default modelAuto-selected on new chatSettings modal
Display orderOrder in model picker dropdownSettings modal (drag)
Access controlOnly specific groups/usersEdit model > Access control
Disable and Hide have the same user effect (excluded from list) but are distinguished in the admin screen. Use Hide for temporary concealment; use Disable to fully block use.

Ollama Model Management

Click the download icon at the top of the model list to open the Ollama model management modal.
FeatureDescription
PullDownload a model from the Ollama model registry (up to 3 concurrent)
Update AllUpdate all Ollama models to latest versions
UploadUpload a GGUF file or download from URL to create a model
DeleteDelete an Ollama model
When multiple Ollama instances are connected, you must first pick the instance to manage.

Model Preset Import / Export

Back up and restore model settings as JSON. This feature is admin-only.
FeatureDescription
Import PresetsRestore model presets from a JSON file
Export PresetsDownload current model presets as a JSON file
Single ExportMore menu (⋯) on a model row > Export
Import overwrites existing model settings. Always Export to back up first.
Only models with null base_model_id (Workspace custom settings) are imported. Agents (which reference another model via base_model_id) aren’t imported in this tab — import them separately under Workspace > Agents.
Click Reset All Models at the bottom of the settings modal (gear) to delete all Workspace custom settings. Connected base model lists are preserved, but customizations like name/description/prompt are all reset.

Connections

Configure LLM provider API connections

Agents

AI assistants combining Knowledge Bases, tools, and guardrails on top of a model

Usage

Per-model token usage and cost analysis