dsh-better-reasoning-effort

Usage

1. Configure a third-party provider (API key etc.) on the official Models page.
2. Expand a model row: the editor block sits under the official capacity fields.
   • Check levels (off / minimal / low / medium / high / xhigh / max) and fill the wire values (e.g. give high the spelling ultra, and the gateway receives ultra when you pick High in the composer);
   • Toggle Image input under Input modalities to declare what the model accepts (unchecked with no declaration = inherit the provider default, usually text-only);
   • Click Auto-adapt to fill recommended levels and modalities from the knowledge base / protocol / endpoint listing — reference capacities show up as read-only hints you can copy into the official fields yourself;
   • Click Apply to write the setting.
3. On a compatible protocol, the Endpoint compatibility section appears at the bottom — set the thinking budget field / vLLM priority on openai-completions, and max_output_tokens handling on openai-responses.
4. All levels off + Apply = unset the declaration; only off checked + Apply = disable reasoning (false); Clear declaration on the modality row + Apply = back to inheriting the provider default.

Declared models are immediately selectable for reasoning effort in the composer's model picker, and image-declared models accept attachments end to end.

Configuration

The host half accepts optional configuration on its profile row (the values below are the defaults):

- insert:
    - id: dsh-better-reasoning-effort
      name: dsh-better-reasoning-effort
      config:
        # Auto-fill undeclared models on boot and after settings updates.
        autofill: true
        # Whether the auto-fill above also fills input-modality declarations.
        modalityAutofill: true
        # Upstream /models probe fetch timeout, in milliseconds.
        probeTimeoutMs: 15000
        # Boot-fill retry backoff schedule; [] means "try exactly once".
        bootRetryDelaysMs: [1000, 2000, 4000, 8000, 16000, 30000]
        # Map effort-less calls on forced-thinking ladders to the vendor default.
        defaultGuard: true

Set autofill: false to disable the silent auto-fill entirely — the browser-side Auto-adapt button keeps working.