Skip to content

Releases: extentos/swift-glasses

1.11.3 — lockstep with Android 1.11.3

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 26 Jul 17:07

No iOS behaviour change. Android fixed simulator auto-bind on physical devices (the transport probed only the emulator's host-loopback alias, so real phones always fell back to typed pairing). Tagged to keep the platforms in lockstep.

1.11.2 — lockstep with Android 1.11.2

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 26 Jul 16:01

No iOS behaviour change. Android moved its 1.7B rung to 4-bit (what iOS has always used) so mid-range Android phones get on-device models instead of cloud. Tagged to keep the platforms in lockstep and from: "1.11.2" resolvable.

1.11.1 — on-device voice (GlassesLocalVoice)

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 26 Jul 14:09

Distributes GlassesLocalVoice, the on-device neural voice (Kokoro via sherpa-onnx). It was the last piece of the local tier no customer could obtain: its sherpa and onnxruntime frameworks were local-path binary targets, which a published package cannot carry.

Both now ride the same url+checksum Release-asset pattern as extentos_coreFFI, so the package is self-contained. Eleven Kokoro speakers are selectable via the voice id (kokoro, kokoro-bella, kokoro-michael, …).

Pairs with Android com.extentos:glasses-local-voice:1.11.0, which solved the same problem by vendoring sherpa into the AAR.

1.11.0 — Automatic model selection + on-device model management

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 26 Jul 13:58

Lockstep release with Android com.extentos:*:1.11.0.

  • Automatic (local-auto): the SDK resolves the best on-device model this phone can sustain, and serves from the managed gateway when none fits. Reported via AssistantEvent.autoModelResolved.
  • Model management on ExtentosLocalTier: models(), autoChoice(), download(modelId:onProgress:), delete(modelId:) — so an app can offer the end user exactly the model Auto picked.
  • Kokoro multi-speaker map (11 voices) shared from the Rust core.
  • LibraryVersion corrected to 1.11.0; it had drifted to 1.4.0 and was mislabelling telemetry.

Platform floor is iOS 17 / macOS 14 (mlx-swift-lm's minimum, applied package-wide by SwiftPM), unchanged from 1.10.1.

GlassesLocalVoice (Kokoro) is not distributed yet — its sherpa binaries need republishing as URL+checksum assets.

1.10.1 — on-device local tier (GlassesLocal)

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 26 Jul 12:00

Distributes GlassesLocal, the on-device local tier. Previously an app could select a local-* model and it would silently run in the cloud, because ExtentosLocalTier.register() was not reachable from any published product.

Breaking: platform floor raised to iOS 17 / macOS 14. mlx-swift-lm declares those minimums and SPM applies platform floors package-wide. An unimported SPM product costs nothing to link, so the floor is the only cost of carrying it.

GlassesLocalVoice (Kokoro) is not distributed yet — its sherpa-onnx binaries need republishing as URL+checksum release assets first.

1.10.0 — local models (sim serving) + guided download events

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 26 Jul 10:11

Android-lockstep release with com.extentos:glasses:1.10.0.

  • Local realtime provider: local-* dashboard model ids serve in the browser simulator (hosted, same weights). On-device engine product follows in a later cut.
  • AssistantEvent.localModelRequired guided-recovery event + spoken refusals.
  • local-auto resolution seam + autoModelResolved event.
  • Prefix-KV persistence groundwork (cold-start elimination).

Binary target: extentos_coreFFI.xcframework.zip (checksum in Package.swift).

1.9.0

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 19 Jul 19:25

Lockstep with com.extentos:glasses:1.9.0 on Maven Central.

Adds OpenAI's two newest realtime models to the assistant catalog:

  • gpt-realtime-2.1 — drop-in successor to gpt-realtime-2: same voices, same reasoning.effort knob, same pricing.
  • gpt-realtime-2.1-mini — distilled reasoning at mini pricing. Despite the name it is not a newer gpt-realtime-mini: it does accept reasoning.effort (the older mini does not) and carries the full 128k context window (the older mini is 32k).

No Swift API change — the only diffs are Package.swift (binaryTarget url + checksum) and the regenerated bindings' doc comments. The default realtime model remains gpt-realtime-2, so upgrading moves no project onto a new model.

.package(url: "https://github.com/extentos/swift-glasses", from: "1.9.0")

1.8.0

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 17 Jul 19:27

Extentos iOS SDK 1.8.0 — version-locked with com.extentos:glasses 1.8.0 (Android). Gemini Live realtime adapter (binary WS frames), sendVideoFrame, modelSupportsVideoInput. Binary core: extentos_coreFFI.xcframework (device arm64 + universal simulator).

1.7.0

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 17 Jul 12:44

First stable release of the Extentos iOS SDK — Swift Package with GlassesCore, GlassesUI, GlassesDebug, GlassesLifecycle, GlassesTesting. Version-locked with com.extentos:glasses 1.7.0 (Android). Binary core: extentos_coreFFI.xcframework (device arm64 + universal simulator).

1.7.0-rc1 (pipeline dry run)

Pre-release

Choose a tag to compare

@Asgermolgaard Asgermolgaard released this 16 Jul 18:38

Pre-release dry run of the swift-glasses distribution pipeline, cut from Extentos monorepo 395d4b8b (current master core — NOT the final 1.7.0 core, which is gated on the Android release pass).

Binary core asset: extentos_coreFFI.xcframework.zip

  • Slices: ios-arm64, ios-arm64_x86_64-simulator (universal), macos-arm64
  • SPM checksum: 310cddc34290e7b7b2f83b8e218b8c8c2462a55f794c58f5fea42702034da456

Do not pin production apps to this tag.