Reference

Point it at any brain.

tomte is provider-agnostic. Sign in with a subscription or an API key, switch models mid-session, and add any OpenAI-compatible endpoint. The trail, the map, and the proofs survive the switch. This catalogue reflects the latest release.

OpenAI

ChatGPT subscription (OAuth) or API key

The GPT-5 family over the Responses and Chat Completions APIs. A ChatGPT Plus, Pro, Team, or Enterprise subscription signs in over OAuth. An API key unlocks the full public catalogue.

Anthropic

Claude subscription (OAuth) or API key

Claude Fable 5 and the Claude 4 family over the Messages API, with adaptive thinking on the newest models. A Claude Pro or Max subscription signs in over OAuth after a terms acknowledgement.

OpenAI-compatible

Built-in presets or config.json

Groq, OpenRouter, DeepSeek, xAI, Together, Fireworks, Cerebras, Mistral, and local Ollama or LM Studio work out of the box as provider/model. Anything else: declare a base URL and key under providers in config.json.

The current catalogue.

Context windows are approximate. Retired ids auto-migrate to their current equivalent on startup, so an existing config keeps working across releases.

OpenAI

7 models

gpt-5.5

Default. Largest OpenAI context window.

1.05M

gpt-5.5-pro

Extended reasoning for hard agent tasks.

1.05M

gpt-5.4

Previous frontier, stable.

gpt-5.4-mini

Fast and cheaper, strong for routine code.

400K

gpt-5.4-nano

Latency-sensitive, cheapest.

200K

gpt-5.2

Earlier frontier, still selectable.

400K

gpt-5

Earlier frontier, still selectable.

400K

Anthropic

8 models

claude-fable-5

Top tier. Adaptive thinking, xhigh effort honoured.

claude-opus-4-8

Frontier Opus. Adaptive extended thinking.

claude-opus-4-7

Adaptive extended thinking.

claude-opus-4-6

Adaptive extended thinking.

claude-opus-4-5

Prior Opus generation.

200K

claude-sonnet-4-6

Balanced speed and capability.

claude-sonnet-4-5

Prior Sonnet generation.

200K

claude-haiku-4-5

Fastest, lowest cost.

200K

Reasoning levels.

The same effort scale works across providers. The newest Claude models think adaptively, and Fable honours the top of the range instead of clamping it.

noneminimallowmediumhighxhighmax

OpenAI-compatible endpoints.

Ten presets work out of the box as provider/model, local servers need no key, and anything else takes a base URL in config.json.

presets + config.json

# built-in presets work out of the box:
tomte config --set-model groq/llama-3.3-70b
# anything else: declare it in config.json
{
  "providers": {
    "myhost": {
      "base_url": "https://api.myhost.dev/v1",
      "api_key": "...",
      "context_limit": 131072
    }
  }
}

Model availability changes between releases. The catalogue in the binary is always the source of truth.

Read the source