Skip to content

Yachiyo Claw v0.0.17

Choose a tag to compare

View all tags
@Wayne1145 Wayne1145 released this 05 Aug 09:02
· 4 commits to main since this release
v0.0.17
31ecea3

Yachiyo Claw v0.0.17

v0.0.17 focuses on making the Android local-agent workflow more dependable on real devices while improving the compact mobile interface.

Local models

  • Repairs GGUF startup on devices where Vulkan discovery is incomplete: CPU inference no longer attempts an unusable Vulkan device, and GPU initialization falls back to CPU when necessary.
  • Relaxes OEM-specific low-memory false positives and falls back from full preload to mmap-backed inference when a complete preload cannot fit.
  • Keeps local model deletion synchronized with the installed-model registry.

Android agent and interaction

  • Adds persistent Live2D move mode with drag, pinch zoom, save, and reset controls.
  • Lets agent sessions choose an app-private workspace, an SAF folder, or a project created in the Development page. Development projects now expose a direct "Use this project" action.
  • Improves narrow/tall-screen layouts, action density, glass fallback layering, and the Skills catalog install controls.

Mainland China connectivity

  • Adds optional mainland China mirrors for Alpine, PyPI, npm, and Android command-line tools, with official upstream fallback for every download.

Build and dependency reliability

  • Fixes GitHub Actions Android SDK validation when Google supplies a newer compatible platform-tools patch release.
  • Pins the workspace tool runtime to Node.js 22 and pnpm 10, and adds dependency overrides for current brace-expansion advisories.

Verification

  • Passed TypeScript checks, Android foundation tests, native-log checks, the full unit suite, web build, Android sync, Gradle unit tests, and debug/release APK builds.
  • Smoke-tested the signed build on an Android emulator at 1200x2608, including the download center, Skills catalog, Live2D move controls, agent workspace setup, and real GGUF generation.

Remaining device-dependent limits

Local inference speed and the largest usable GGUF model depend on the device ABI, available memory, thermal state, quantization, and the installed GPU driver. The app will prefer a working CPU fallback rather than claiming a broken GPU configuration is available.

Assets 4
Loading