Desktop Dev Build #194.1 (e2369cb)

Automated desktop build artifacts for commit e2369cb557895dcee5c903e4cd05c312dc0328b3.
Workflow run: https://github.com/LettuceAI/app/actions/runs/30229730432

Changes since previous Desktop build desktop-dev-193-1-f805d8d

Compare: desktop-dev-193-1-f805d8d...e2369cb

• e2369cb fix(ui): fix state mismatch on playground
• 6fc0f96 Desktop navigation and header customization (#80)
• b04abb2 feat(performance): paginate recent generations list
• 67eaac4 perf(ci): reduce native build and dependency overhead
• c3e4f23 feat: local image generation (#79)
• f4bd7d7 feat(companion): match dynamic-memory cycle status UI
• 96d19d5 fix(build): reduce Rust and CUDA cache pressure
• 7124e55 fix(memory): bound dynamic-memory rebuilds after rewind
• 95af6a0 fix(companion): preserve earned relationship progress
• f66a3ca fix(companion): separate turn effects from passive drift
• 803be8c fix(i18n): restore prompt preview mock labels
• d6ed0d5 feat(llama): add configurable penalty range
• 2c767bd fix(ui): rebuild settings nav when LLM metrics appear
• 89828a1 feat(i18n): translate feature generation settings
• 62440bb feat(ui): edit feature generation configs in the model editor
• 7ddbb61 feat(core): add per-model feature generation configs to request resolution
• 758390e fix(ui): constrain MTP placement select width
• d0b4549 feat(llama): add custom sampler chains and saveable presets
• cae4d76 feat(llama): add configurable repeat penalty
• 386a626 feat(i18n): translate MTP placement controls
• 7405637 fix(llama.cpp): make MTP-aware GPU offloading reliable
• 4f27894 fix(gemini): fix Express aspect ratio and image-block error surfacing (#73)
• 094b534 fix(group-chat): discard hidden card prompt fields
• af9aae4 fix(group-chat): honor character prompt fallback
• e8a9f7f fix(appearance): hide participants bar controls outside group chats
• 26401cf feat(chats): regenerate with a different model
• 48610e3 feat(i18n): translate performance and microbatch strings across locales
• c79f8ca fix(companion): surface real relationship movement
• 5cf4736 fix(group-chat): complete Director continuation turns
• 0f3726d fix(memory): refresh manual cycles in branched chats
• e10d53b fix(llama): honor Qwen thinking directives
• 8b92306 perf(llama): adapt speculative draft length
• 3dfdc51 perf(llama): reduce per-token streaming overhead
• 82504f4 feat(llama): cache prefixes across conversations
• d874be3 feat(llama): fit automatic GPU offload natively
• 27c7938 feat(llama): separate prompt microbatch sizing
• 799a0c6 feat(llama): report native inference timings
• 3af453c refactor(prompting): unify model context as prompt entries
• e2d4f0a perf(llama): reuse cached prompt prefixes
• c49abb3 fix(llama): fully offload bundled model layers
• fe8c040 fix(llama): measure generation throughput correctly
• 572b79f feat(memory): deterministic rewind with user-edit event log
• 40a435e feat(group-chat): branch to 1-on-1 from a message with memory cutoff
• 4898011 fix(smart-creator): preserve drafts when opening editors
• e61306e feat(group-chat): switchable classic and detailed list card layouts
• 0b827bc fix(group-chat): scope history to its group and align card design
• 1c6128e feat(group-chat): unify conversation backend and session UX
• 415f626 style(lorebook): align generator flow page with app design language
• aa39b4d fix(gemini): cache tool declarations with context
• 242450d fix(gemini): reject unsupported image edit inputs
• 4bfb98e fix(gemini): honor requested image aspect ratios
• 3c16b2a fix(gemini): preserve v1beta base URLs
• b5f85c5 fix(gemini): retain thought signatures in chat history
• 5fef562 fix(gemini): preserve raw tool call names
• cc70de9 fix(lorebook): match no-space language keywords and regex triggers
• c3c2ced fix(chat): keep cancelled swipe replies after their prompt
• f4ae97f fix(pricing): prevent negative OpenRouter Gemini completion costs
• 4d37d92 fix(tts): hide espeak console windows on windows and phonemize in one spawn instead of per-segment fallback