• v1.2.0 875aa41b10

    outage.sh released this 2026-11-03 09:47:01 +01:00 | 6 commits to main since this release

    Model attribution for every reply.

    • Every assistant reply now shows a model chip; click it for the provider, API, exact model ID and a copy button.
    • The session header summarizes the models used, with reply counts and a clear "not recorded" state.
    • Model metadata is now read from Goose, Hermes, GitHub Copilot, Codex and Mistral Vibe, in addition to the formats already supported.
    • Codex no longer invents a default "gpt-5" model: it uses the session's provider and each turn's recorded model.
    • Sources that do not store a model per reply stay honestly marked "Not recorded" instead of guessing.

    44 unit tests pass (7 skipped). Universal build (arm64 + x86_64), macOS 14+.

    Downloads