Skip to content

feat: map interchange's neutral effort to reasoning.effort - #33

Merged
TheGreatAxios merged 1 commit into
mainfrom
feat/map-inference-effort
Oct 29, 2026
Merged

TheGreatAxios merged 1 commit into
mainfrom
feat/map-inference-effort

Conversation

@TheGreatAxios

@TheGreatAxios TheGreatAxios commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

Closes #32

What changed

  • InferenceOptions.effort is sent as reasoning.effort, also on sources with no reasoning quirk: low, medium, high as given; max as the new reasoning.maxEffort quirk (xhigh | max), else high (no warning: the package has no logger).
  • Behaviour change: a caller that already set effort now gets reasoning.effort on the wire, where before nothing was sent. This matches Interchange's own OpenAI adapter, which also has no opt-out.
  • A non-empty providerOptions[reasoning.effortOption] still wins; includeReasoningEffort filters both.
  • README documents the quirk and the mapping.

Verification

  • bun run check (typecheck, lint, format, 53 unit tests) passes.
  • New tests pin the wire body with no reasoning quirk, the maxEffort fallback, option precedence and the hook filter.
  • Independent review findings addressed (comment accuracy, README, test trimming).

low, medium and high are sent as given; max sends the new reasoning.maxEffort quirk, else high, as Interchange's own OpenAI adapter does. An explicit providerOptions effort still wins.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Map Interchange's neutral effort to reasoning.effort

1 participant