Spaces:
Sleeping
Sleeping
Commit History
-
Fix subscription audit crash on subs missing 'occurrences'
build_money_input hard-indexed s['occurrences'] and s['last_charged'], which KeyError'd when audited subscriptions came from the Memory tab (name/cost/last_used) instead of a CSV upload, taking down the whole money_review stream. Use .get() with fallbacks, matching the style already used for income/budget in the same function. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>2bc3243
-
Add config, model status, and VLM support
Introduce centralized configuration (.env.example + config.py) and wire it into app and engine. Add logging and a /status endpoint, new streaming API payment_impact, and reset_account. Refactor engine to track model state, surface friendly load errors, support a lazily-loaded vision-language model (VLM) for meal photos (describe_food_image), and improve generation streaming by hiding model reasoning (ANSWER_DELIM/_clean_response). Update meal-photo upload to prefer vision with OCR fallback and adjust warmup/launch to use config host/port. Update docs/contracts and remove mock websearch stubs; tests/static assets adjusted accordingly.0c4cd3b
-
Add EasyOCR/CUDA support and UI improvements
Introduce GPU-capable OCR and CUDA bootstrap for Windows, plus several frontend UX features and docs updates. Key changes: - README: document EasyOCR (GPU/CPU) + Tesseract fallback and detailed CUDA/llama-cpp-python build instructions. - Add cuda_bootstrap.py and scripts/build_cuda.bat to preload NVIDIA/CUDA DLLs on Windows and assist building CUDA wheels. - engine.py / rag.py: attempt GPU offload using LIFEOS_GPU_LAYERS (default -1) with a safe fallback to CPU; ensure cuda_bootstrap runs before llama_cpp loads. - features/food.py: add ocr_image() that prefers EasyOCR (GPU when available) and falls back to Tesseract; remove direct tesseract coupling in app.py and use the new helper. - requirements.txt: add easyocr. - Static assets (JS/HTML/CSS): add workout log, calendar day modal view, "other" describe toggles, theme toggle with light-mode styles, small UI/text tweaks and polish. These changes enable optional GPU acceleration for model and OCR workloads, improve Windows CUDA runtime h…d1ce419 -
Configure Tesseract and fix stale embeddings
Add automatic Tesseract configuration and self-healing for embedding vectors. features/food.py: introduce configure_tesseract(pytesseract) which honors TESSERACT_CMD, probes common Windows install locations, and sets pytesseract.pytesseract.tesseract_cmd when needed. Call it from extract_deals to ensure OCR works when the tesseract binary isn't on PATH. app.py: call configure_tesseract before invoking pytesseract in upload_meal_photo. rag.py: detect notes whose stored vector length differs from the current embedder output, re-embed those notes, update in-memory entries and persist the corrected vectors under the store lock so recalls remain robust across different embedder dimensions.ab704ae
-
Major feature expansion: profile, calendar, voice chat, goal coach
- Profile tab: name/city/address/income/diet/budget -> memory; optional city flyer-deal web search (off-grid preserved when unused) - Chat: mic input + spoken replies (browser Web Speech API), opt-in web-search toggle (default off, clearly labeled) - Kitchen: meal/receipt photo -> local OCR -> buying-pattern analysis - Health: preferred training days/time, 7-day calendar with events, deterministic free-slot planning around the calendar, browser reminders - Money: monthly payments editor + Socratic goal coach (one probing question per turn, then a concrete savings plan) - Frontend rewritten in clean UTF-8 with inline SVG icons (fixes mojibake), finite-only animations, resilience banners - New tests: tests/test_extensions.py (7) + integration suite now 15 checks, all passing; no regressions in food/health/money suites Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>86f6b4e -
Add LifeOS: local-first personal assistant on Nemotron-3-Nano-4B
Gradio 6 Server-mode app with hand-built frontend, three feature modules (kitchen/health/money), short-term JSON memory plus a local RAG long-term memory, llama.cpp inference for chat and embeddings, seed demo data, docs/architecture.md, and test suites. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>40cf485