feat: SwiftBuddy Memory Palace V1 & Omni Audio Edge Stabilization#31

Open

solderzzc wants to merge 157 commits intomainfrom

feature/swiftbuddy-mempalace-v1

Commits on Apr 8, 2026

ci: secure MLX_METAL_PATH for E2E tests dynamically
solderzzc
committed
feat(chat): inject invisible RAG fetch into chat loops
solderzzc
committed
fix(gui): Resolve SwiftBuddy HuggingFace search blank list due to StateObject lifecycle bug
solderzzc
committed
fix(gui): Resolve ModelManagementView SwiftUI sheet presentation interference and make downloaded models directly tappable to load
solderzzc
committed
chore: Add diagnostic logging to HFModelSearch
solderzzc
committed
fix(chat): Improve extraction of raw <think> tags from model output that bypasses token.isThinking flag
solderzzc
committed
feat(chat): Persist generated reasoning in a collapsed ThinkingPanel for finalized chat messages
solderzzc
committed
fix(macOS): Re-enable native model downloading in ModelDownloadManager via HubApi for SwiftBuddy
solderzzc
committed
fix(models): Allow arbitrary Hugging Face models to appear in downloaded catalog
solderzzc
committed
feat(search): Dynamically fetch exact model storage sizes and display MLX/GGUF formatting for Hub queries
solderzzc
committed
fix(search): Prevent Hugging Face rate limits from abruptly hard-failing the Search UI model list
solderzzc
committed
fix(search): Refactor HFSearchTab to use ScrollView instead of List to prevent macOS layout recursion crashes resulting in blank models
solderzzc
committed
feat(search): Implement horizontal parameter size filtering UI and recursive background querying for HF Hub discovery
solderzzc
committed
fix(search): Support M-scale parameter size filtering and implement native Hub cursor pagination
solderzzc
committed
style(ui): Remove model picker auto-open on launch and enforce rigid skeleton constraints for HuggingFace Hub modal layout
solderzzc
committed
fix(search): Exclude GGUF models implicitly and fix CatalogTab layout recursion crash
solderzzc
committed
style(search): Move Hub search header to top of model list and prioritize absolute cached file size in row view
solderzzc
committed
feat(mempalace): Mock registry personas and implement strict InspectorView bounds
solderzzc
committed
chore(networking): Add User-Agent headers and explicit response logging to RegistryService to trace GitHub API access drops
solderzzc
committed
feat(mempalace): Refactor RegistryService fetching to rely on CDN-cached persona.json and statically request known room txt files
solderzzc
committed
fix(mempalace): Bypass aggressive internal URLSession CDN edge caching preventing successful 404 recovery
solderzzc
committed
perf(mempalace): Implement batch saveMemories API to prevent CoreData WAL transaction flooding during massive persona corpus ingestion
solderzzc
committed
feat(engine): Automatically load the last active model from UserDefaults on SwiftBuddyApp boot sequence
solderzzc
committed
fix(ui): Eliminate NSDetectedLayoutRecursion infinite AppKit redraw loops by converting TextEditor blocks to vertical TextFields inside iOS/macOS active ScrollViews
solderzzc
committed
feat(mempalace): Implement Wake-Up Persona integration into MLX pipeline and introduce Native graphical Map hierarchy for memory rooms
solderzzc
committed
fix(ui): Add explicit window dismiss button to PalaceVisualizerView for teardown on macOS modal sheets
solderzzc
committed
feat(ui): Add explicit swiftData driven Persona wing picker injected natively into ChatView toolbars for RAG identity mapping
solderzzc
committed
feat(ui): Bind chat navigation title and input placeholder to explicitly reflect the currently selected memory persona wing
solderzzc
committed
feat(ui): Decouple Text Ingestion / Memory Miner from Inspector Registry and pivot root Navigation to a primary Friends List model
solderzzc
committed
fix(ml): Resolve Jinja Template constraint failures with Gemma architectures by forcefully prepending RAG variables linearly against raw User instructions rather than allocating hostile System Role…
solderzzc
committed
feat(ui): Transform Persona Chat layout into immersive Visual-Novel RPG architecture with native Memory Palace TALK TONE adherence hooks.
solderzzc
committed
fix(ml): Squash sequential User/Model roles automatically within ChatViewModel routing logic to inherently satisfy strict Gemma Jinja alternation templates
solderzzc
committed
fix(ml): Wire missing Repetition Penalty properties uniformly across the Inference Engine and export a dedicated tuning slider to the Settings manifest to definitively halt Gemma token echoing dege…
solderzzc
committed
perf(ml): Physically separate static Persona Directives from dynamic volatile RAG vectors to immediately salvage MLX KV cache continuity across successive conversational turns
solderzzc
committed
fix(ml): Dynamically transform assistant roles into model invariants and universally trap <end_of_turn> string literals to definitively halt Gemma 3 execution streams
solderzzc
committed
feat(ui): Embed interactive asynchronous download resolution blocking natively into the HuggingFace Search Hub rows to ensure models are offline-verified before triggering picker teardown
solderzzc
committed
fix(ui): Enforce instantaneous visual feedback on remote download clicks and trap silent HF snapshot failures to guarantee observable developer console logs
solderzzc
committed

Commits on Apr 9, 2026

Commits on Apr 10, 2026

Commits on Apr 11, 2026

Commits on Apr 12, 2026

Commits on Apr 13, 2026

chore: sync mlx-swift-lm with MoE generic loader fix (scale param patch)
solderzzc
committed