Skip to content
New Kodine v2 is now available

Models

Pick and tune LLM providers and models.

Built on the AI SDK and Models.dev, Kodine talks to 75+ LLM providers out of the box, and local models work too.


Providers

The well-known providers come preloaded. Credentials saved through the /connect command are picked up automatically the next time Kodine starts.

Full details: providers.


Select a model

With a provider configured, pick the model you want by typing:

/models

The model landscape shifts constantly, with new releases landing every week.

Only a handful of models truly excel at both writing code and calling tools.

The Kodine lineup below — available through Kodine Aura behind a single balance — is curated for exactly that, in no particular order (not exhaustive, and not necessarily current):

  • kodine-max
  • kodine-large
  • kodine-code
  • kodine-swift
  • kodine-small
  • kodine-mini

Set a default

Pin one of these as your default by setting the model key in the Kodine config.

kodine.json
{
"$schema": "https://kodine.net/config.json",
"model": "lmstudio/google/gemma-3n-e4b"
}

IDs follow the provider_id/model_id shape — with Kodine Aura, for instance, Kodine Large is kodine/kodine-large.

With a custom provider, the provider_id matches your key under the provider section, and the model_id matches the key inside that provider’s models.


Configure models

Per-model options can be set globally through the config.

kodine.jsonc
{
"$schema": "https://kodine.net/config.json",
"provider": {
"openai": {
"models": {
"gpt-5": {
"options": {
"reasoningEffort": "high",
"textVerbosity": "low",
"reasoningSummary": "auto",
"include": ["reasoning.encrypted_content"],
},
},
},
},
"anthropic": {
"models": {
"claude-sonnet-4-5-20250929": {
"options": {
"thinking": {
"type": "enabled",
"budgetTokens": 16000,
},
},
},
},
},
},
}

The example applies global settings to two built-in models — one reached through the openai provider, the other through anthropic. Models.dev lists the built-in provider and model identifiers.

The same options exist per agent, where they take precedence over the global values here. Learn more.

Custom variants can extend the built-in set, giving one model several named configurations without duplicating its entry:

kodine.jsonc
{
"$schema": "https://kodine.net/config.json",
"provider": {
"kodine": {
"models": {
"gpt-5": {
"variants": {
"high": {
"reasoningEffort": "high",
"textVerbosity": "low",
"reasoningSummary": "auto",
},
"low": {
"reasoningEffort": "low",
"textVerbosity": "low",
"reasoningSummary": "auto",
},
},
},
},
},
},
}

Variants

Plenty of models expose several variants with distinct configurations, and Kodine bundles sensible defaults for the major providers.

Built-in variants

These variant defaults ship out of the box:

Anthropic:

  • high — elevated thinking budget (default)
  • max — the largest thinking budget

OpenAI:

Exact names differ per model, but roughly:

  • none — reasoning off
  • minimal — the lightest reasoning effort
  • low — light reasoning effort
  • medium — moderate reasoning effort
  • high — heavy reasoning effort
  • xhigh — the heaviest reasoning effort

Google:

  • low — reduced effort and token budget
  • high — increased effort and token budget

Custom variants

Existing variants can be overridden, and brand-new ones added:

kodine.jsonc
{
"$schema": "https://kodine.net/config.json",
"provider": {
"openai": {
"models": {
"gpt-5": {
"variants": {
"thinking": {
"reasoningEffort": "high",
"textVerbosity": "low",
},
"fast": {
"disabled": true,
},
},
},
},
},
},
}

Cycle variants

The variant_cycle keybind jumps between variants from the keyboard. Learn more.


Loading models

At startup Kodine resolves the active model in this priority order:

  1. A --model or -m command line flag, in the same provider_id/model_id shape the config file uses.

  2. The model entry in the Kodine config.

    kodine.json
    {
    "$schema": "https://kodine.net/config.json",
    "model": "anthropic/claude-sonnet-4-20250514"
    }

    Again formatted as provider/model.

  3. Whichever model was used most recently.

  4. The first model according to an internal priority.