Models

Select and configure AI models from any supported provider. Switch mid-session without restarting.

Switching Models

Use the /model command in the REPL:

❯ /model                 # interactive provider/model picker
❯ /model openai          # switch to OpenAI provider
❯ /model gpt-5.5         # set specific model
❯ /model list            # list available models from all providers

Capability Tags

The /model list command fetches live model lists from provider APIs and displays capability tags:

deepseek-v4-flash      [200K ctx, tools, reason, free]
gpt-5.5                [128K ctx, vision, tools]
claude-sonnet-5        [200K ctx, vision, tools, reason]
gemini-2.0-flash       [1M ctx, vision, tools, free]

Tags include context window size, vision support, tool calling, reasoning, and free tier availability.

Reasoning Effort

For models that support configurable reasoning (DeepSeek, NVIDIA, OpenRouter, etc.):

❯ /effort low
❯ /effort medium
❯ /effort high

Local Models (Ollama)

Run models locally with Ollama:

❯ /model ollama/llama3.3
❯ /model ollama/mistral
❯ /model ollama/codellama

Set the Ollama host:

export OLLAMA_HOST=http://localhost:11434

Image Generation Models

GenerateImage auto-discovers image-capable models from provider APIs on first use:

ProviderModels DiscoveredAPI Key
OpenAIdall-e-3, future DALL-E versionsOPENAI_API_KEY
OpenRoutergemini-3.1-flash-image-preview, flux.2-pro, recraft-v3, etc.OPENROUTER_API_KEY
Googleimagen-3.0-generate-002, future Imagen versionsGOOGLE_API_KEY

To override auto-discovery, add "capabilities": { "imageGen": { "defaultModel": "...", "apiType": "..." } } to the provider in providers.json.

Provider Defaults

Each provider has a default model. The first time you switch to a provider, its default model is selected automatically. Override with:

❯ /model deepseek-v4-flash