Skip to content

v0.7.0

Choose a tag to compare

@github-actions github-actions released this 26 May 10:19
· 23 commits to main since this release
  • Voice input in chat (#44). New mic button between the text field and the send button launches Android's built-in speech recognition; the transcript appends to whatever you already typed, so you can dictate the second half of a prompt. Works on any device with Google Speech Services; falls back gracefully with a toast on devices without it. No RECORD_AUDIO permission is requested by PokeClaw — the system dialog handles its own permission.
  • Persistent global instructions (#45). New Settings entry under "Models" → "Global instructions". Anything you put here is prepended to every conversation's system prompt — useful for "always reply in Cantonese" or "I'm in Vancouver, default to PST" style rules. Up to 2000 characters; clear the field to disable. Applies to both chat mode and task/agent mode.
  • Custom local model download URL (#36). Advanced Settings entry: point PokeClaw at a self-supplied .litertlm URL (HuggingFace mirror, your own server, etc.) and it appears as a new row in the Available Models list alongside the built-in Gemma options. File-size validation is relaxed for custom models because we can't know the expected size up front.
  • Pencil-icon rename shortcut in the sidebar. Each saved conversation now shows a small edit icon next to its title; tap to rename without having to long-press. Long-press menu still works for delete.
  • Floating button uses the PokeClaw icon instead of the "AI" text label.
  • Debug report includes GPU / OpenCL / RAM / ABI diagnostics (#41 / #14). Summary now lists supported ABIs, total RAM, and whether libOpenCL.so is present at any of the well-known driver paths — community OEM reporters can self-diagnose GPU-path issues without a back-and-forth.
  • i18n for new features. Voice input, global prompt, and custom model URL strings shipped in English, Simplified Chinese, and Japanese.