github agent0ai/spynel v0.10.0
Spynel v0.10.0 — Reasoning effort selection

latest releases: v0.16.2, v0.16.1, v0.16.0...
one month ago

Spynel now lets you select reasoning effort after choosing a model, using the capabilities advertised by the harness. Supported service or speed choices follow before the complete selection is saved.

  • Codex discovers per-model reasoning efforts and service tiers and sends the selected values with subsequent turns, including newly advertised efforts such as ultra.
  • /effort, /speed, and their plain CLI equivalents inspect, select, or reset supported properties to inherit.
  • Model, effort, and service mode are persisted atomically. Active turns retain their admitted settings; later dispatches use the saved selection. Model or harness changes clear incompatible stored properties.
  • Completing the TUI selector refreshes the configuration row while preserving other edits; cancellation preserves the parent form.
  • Pi discovers model-specific thinking levels without changing Pi's global defaults. Claude uses adapter-known effort capabilities. ACP reasoning and speed selection remain unavailable because the necessary choices are not exposed before session creation.

Existing configurations that omit reasoning effort retain their historical behavior; explicit inherit uses provider defaults.

Native packages support Linux and macOS on amd64 and arm64. Windows remains temporarily unsupported. The release workflow attaches native archives and checksums, then publishes spynel@0.10.0 to npm latest.

Source verification: repository Go tests/vet/build, provider-free smoke and DOX, npm launcher tests, package dry-run, and native Linux arm64 archive smoke passed. The accepted feature also has isolated live Codex discovery, selection, persistence, and invocation evidence.

Full changes: v0.9.1...v0.10.0

Don't miss a new spynel release

NewReleases is sending notifications on new releases.