Project Entity

Give me an Arm phone and a GGUF. ENTITY reads the silicon, tunes itself to it and proves which decision earned the speedup.

Comment

posted an update —

[2.3.0]

UI polish and quality-of-life. A visual pass over the whole app plus the small features a daily driver needs. No inference-path changes, so every published benchmark number carries over.

Added

  • Multi-select conversation delete: the Conversations dialog gains a Select mode with checkboxes and a single confirmed bulk delete.
  • Share chat: exports the current conversation as plain text through the system share sheet.
  • Graph style options (menu): Fill area and Smooth lines for the live metrics graph. Both are decorative, so they obey the Animations setting like every other effect; Animations off keeps the plain minimal graph.

Changed

  • Refined visual system: hairline borders instead of heavy strokes, neutral assistant bubbles with asymmetric corners, pill-shaped input bar, card-grouped Settings and Benchmark screens, soft status pills (green/amber) for the KleidiAI advisor on the model info card, consistent 16 dp spacing rhythm and a tightened type scale. Light and dark themes both reworked; the ENTITY teal accent and the metrics identity are unchanged.
  • The toolbar reset action now uses a reset glyph instead of a pencil, which read as "edit".
  • The sustained-benchmark duration selector spreads its 2/5/10 min options evenly across the row instead of overflowing on narrow screens.
  • Settings card dividers use consistent 12 dp spacing.

Fixed

  • Benchmark CSV meta: affinity_naive now reports mask_all_cores_effectively_unpinned instead of the misleading pinned_fast_cores (the naive arm's mask is the N fastest of N cores, i.e. all of them; the behavior was always correct, the label was not).

Log in or sign up for Devpost to join the conversation.