Models and Parameters

All Models

The Models view aggregates every model from every discovered service into a single searchable list. Use the search bar to filter by name.

Favorites and Aliases

Star a model to pin it in the chat dropdown. You can also create aliases — short names that map to a specific service/model pair:

@fast  →  ollama/llama3
@smart →  lmstudio/deepseek-r1

Aliases are usable in the model dropdown search.

Manual Endpoints

Add backends that are not auto-discovered. Click Add Endpoint and provide:

Field	Description
Name	Display name for the service
URL	Base URL (e.g., http://192.168.1.50:11434)
API type	openai, ollama, or anthropic

Manual endpoints appear alongside discovered services.

Settings Overlay

Click the gear icon on any model or service to open the settings overlay. Settings have two scopes:

• Global — applies to all services unless overridden
• Per-service — overrides global for a specific backend

Sampling Parameters

Parameter	Range	Default
temperature	0.0 -- 2.0	0.7
max_tokens	1 -- model max	4096
top_p	0.0 -- 1.0	1.0
top_k	1 -- 500	40
frequency_penalty	-2.0 -- 2.0	0.0
presence_penalty	-2.0 -- 2.0	0.0
seed	integer	—

Stop Sequences

Add one or more strings that cause the model to stop generating. Enter each sequence and press Enter to add.

System Prompt

Set a system prompt at global or per-service scope. The per-service prompt fully replaces the global one when set.

You are a helpful coding assistant. Always include type annotations.

Response Format

Choose between:

• Text — default free-form output
• JSON schema — paste a JSON schema to constrain model output to that structure

Ollama-Specific Parameters

When an Ollama service is selected, additional parameters are available:

Parameter	Description
keep_alive	How long to keep the model loaded after last use (e.g., 5m, 24h)
mirostat	Mirostat sampling mode (0 = disabled, 1, 2)
repeat_penalty	Penalty for repeated tokens (1.0 = none)
repeat_last_n	Lookback window for repeat penalty
min_p	Minimum probability threshold
tfs_z	Tail-free sampling parameter
typical_p	Locally typical sampling threshold
num_ctx	Context window size in tokens
num_batch	Batch size for prompt processing