LettuceAI/app 1.2.0

on GitHub

Highlights

• Major investment in desktop UX, especially the Create Character flow.
• Significant expansion of the prompt system, including a new Prompt Structure Viewer and new runtime prompt-injection controls.
• Broad chat stability/performance hardening and large dynamic-memory improvements.
• Continued embedding/ONNX runtime reliability work across desktop, Android, and Windows.
• Expanded provider/model ecosystem (including NVIDIA NIM) and safer import behavior.

Desktop UI and Character Creation Redesign

• Added responsive desktop layouts across character creation steps.
• Reworked character creation step order to improve setup flow.
• Redesigned character “extras” inputs for clearer, faster editing.
• Improved character create/edit with fallback-model selector support.
• Fixed UX friction in character flows, including create-step ordering and navigation consistency.
• Added metadata handling improvements for imported cards and avatar URL behavior.
• Added lorebook import support in character creation workflows.

Prompt System Upgrades

• Added Prompt Structure Viewer in the system prompt editor to preview message composition.
• Added conditional prompt injection mode.
• Added interval prompt injection mode.
• Added runtime option to condense prompts into a single system message.
• Fixed prompt import behavior to correctly respect prompt_order.
• Fixed buggy drag-and-drop reorder behavior in prompt entry editor.
• Improved prompt-related import UX and editor predictability.

Chat, Group Chat, and UI

• Added shared ChatLayout for persistent background behavior across chat sub-routes.
• Added shared GroupChatLayout with lifted data loading.
• Added branch-to-group-chat action from message actions.
• Added lorebook-usage visibility per message.
• Added safe-area padding fixes for chat footer and bottom menu.
• Removed unwanted dark overlay above background images.
• Fixed chat search back-button sizing and related UI polish.
• Improved session back-stack handling across settings/history navigation.
• Fixed persona selection conflicts during scroll interactions.
• Removed duplicate dismiss controls in chat memories error state.

Chat Stability and Performance

• Fixed dynamic-memory listener leak during async chat setup.
• Bounded attachment cache growth in session hooks.
• Ignored stale attachment loads after chat state transitions.
• Fixed cleanup of jump-to-message RAF/timeout resources.
• Improved message memo checks with derived display props.
• Reduced attachment diff cost in chat memoization.
• Added fallback model retry logic with usage attribution.
• Disabled fallback attempts when no fallback model is configured.
• Added swap-places mode with role-aware generation.
• Reverted one streaming animation perf change after validation feedback.

Dynamic Memory System

• Added cursor-delta summarization of new messages.
• Added self-healing cursor behavior after deletes/rewinds.
• Added deduplication by cosine similarity at memory creation.
• Added adaptive decay rate based on access count.
• Added category tagging for memories.
• Added hybrid retrieval using similarity + recency + access frequency.
• Added configurable retrieval selection limit.
• Added smart and cosine retrieval strategies.
• Added memory panel category filter chips.
• Added memory activity log redesign with timeline/collapsible UX.
• Auto-refresh of memory views after dynamic-memory completion.
• Enforced gating behavior for dynamic-memory manual mode.

Embeddings, ONNX Runtime, and Android

• Fixed ONNX runtime bundling and dylib path handling.
• Fixed dev rebuild-loop behavior tied to ONNX runtime integration.
• Pinned/standardized dylib preloading and path behavior.
• Ensured Android ONNX resource directory and packaging consistency.
• Improved desktop guards around ONNX runtime initialization.
• Improved handling of ORT init result variants/booleans.
• Added pre-step for embedding download and runtime ORT fetch.
• Extracted Windows DLL dependencies for ONNX runtime packaging.
• Locked ORT version to 2.0.0-rc.10.
• Added embedding model v3 support and multi-version management.
• Added experimental keep-loaded embedding runtime with cache reset on version switch.
• Fixed Android post-regenerate WebView freeze and tracing consistency.
• Made Android ONNX runtime init deterministic.

Providers, Models, Endpoints, and Security

• Added NVIDIA NIM provider.
• Added custom-provider tool-choice mode configurability.
• Added OpenRouter free-model toggle in model selector.
• Improved model selector search and suggestions.
• Added custom endpoint config persistence and auth/model-fetch mapping controls.
• Hid llama.cpp provider on mobile onboarding/settings where unsupported.
• Added security toggle to disable remote avatar downloads on card import.
• Disabled Chutes API key validation where it blocked onboarding flows.

Lorebooks, Import/Export, and Content Controls

• Added world-info import/export and creation import action.
• Added character card metadata support and lorebook import path improvements.
• Added new “pure mode” content filtering system.

Usage, Sync, and Reliability

• Added app-time tracking backend support and analytics view.
• Enforced host-authoritative manifest diff in sync logic.
• Improved DB reset error surfacing and reset-in-place behavior.

Tooling, CI, and Maintenance

• Migrated workflows to Blacksmith.
• Switched workflows to Bun and refreshed README/tooling docs.
• Added libclang dependency for Windows CI builds.
• Removed duplicate Cargo libraries and cleaned project config.
• Added .gitignore updates and docs-folder ignore adjustments.
• Fixed Tailwind warning noise in UI build paths.

What's Changed

• .github/workflows: Migrate workflows to Blacksmith runners by @blacksmith-sh[bot] in #8
• Lock ort dependency to exact version 2.0.0-rc.10 by @Copilot in #9

New Contributors

• @blacksmith-sh[bot] made their first contribution in #8

Full Changelog: 1.1.0...1.2.0