The provider and model commands manage providers and models from your crushrc.
The provider and model commands manage providers and models from your crushrc. Entity commands use add to create or update something and remove (or rm) to delete it. Booleans accept true/false/1/0/yes/no, in any case.
provider
Manage model providers.
Usage: provider [command]Available Commands: add Add or update a provider remove Remove a provider and its custom models rm Alias for removeprovider add
Add a provider, or update an existing provider with the same ID.
Usage: provider add <id> [flags]Flags: --name string display name --type string provider type (completions, responses, messages, ollama, …) --api-key string API key --base-url string API base URL --disable bool disable without removing --flat-rate bool use flat-rate billing --discover-models bool auto-discover and merge provider models --system-prompt-prefix string text prepended to the system prompt --extra-header key value add an HTTP header (repeatable) --extra-body JSON merge a JSON object into request bodies --provider-options JSON merge a provider-specific JSON objectprovider add deepseek \ --type completions \ --base-url "https://api.deepseek.com/v1" \ --api-key "${DEEPSEEK_API_KEY:?set DEEPSEEK_API_KEY}"The type describes the API format the provider speaks: completions (OpenAI chat completions), responses (OpenAI responses), or messages (Anthropic messages). Local providers with a registered enricher (ollama, litellm, lmstudio, …) keep their own type. Legacy values (openai, openai-compat, anthropic, …) are migrated automatically at load.
Headers whose value resolves to the empty string (an unset $VAR, a $(...) that prints nothing, or a literal "") are dropped from the outgoing request. This makes env-gated headers safe:
provider add openai \ --extra-header OpenAI-Organization "$OPENAI_ORG_ID"If OPENAI_ORG_ID is unset, the header is simply not sent.
provider remove
Remove a provider and all custom models registered on it.
Usage: provider remove <id> provider rm <id>model
Manage custom models and the large/small model slots. Model references use the same <provider>/<id> form printed by crush models.
Usage: model [command]Available Commands: add Register a custom model on an existing provider remove Remove a custom model rm Alias for remove large Set or print the large model small Set or print the small modelmodel add
Register a custom model on an existing provider.
Usage: model add <provider>/<id> [flags]Flags: --name string display name --context-window int context window in tokens --default-max-tokens int default maximum output tokens --thinking string always, never, or toggleable --supports-images bool model accepts image input --price-input float input price per 1M tokens --price-output float output price per 1M tokens --price-cache-create float cache-creation price per 1M tokens --price-cache-hit float cache-hit price per 1M tokens --reasoning-effort string low, medium, or high --max-attachments int maximum attachments per requestmodel remove
Remove a custom model from its provider.
Usage: model remove <provider>/<id> model rm <provider>/<id>model large, model small
Set the large or small model slot. With no model argument, print the current selection.
Usage: model large [<provider>/<id>] [flags] model small [<provider>/<id>] [flags]Flags: --think enable thinking mode --reasoning-effort string low, medium, or high --max-tokens int maximum output tokens --temperature float sampling temperature --top-p float top-p sampling (0–1) --top-k int top-k sampling --frequency-penalty float frequency penalty --presence-penalty float presence penalty --provider-options JSON merge a provider-specific JSON objectmodel large openai/gpt-4o --thinkecho "coding with: $(model large)" # prints: openai/gpt-4oRead this post in your terminal with glow:glow -p https://crush-docs.fly.dev/docs/providers/reference.md