- Add ID column to stock table header and renderInventoryRow
- Add 'Entrada' button on each stock row to open purchase modal pre-filled
- Show inventory ID in product detail popup
- recordPurchase: close modal, clear form, reload stats and stock list on success
- Fix colspan 11 -> 12 for empty state rows
- Expose loadInventoryStats globally so inventory.js can call it after CRUD
- Fix token key: use pos_token (not access_token) to match auth scheme
- After successful POST /items: close modal, clear form inputs, reload stats
- Bump inventory.js cache-bust query param v3 -> v4
- Add Dockerfile.whatsapp-bridge with Baileys + env var support
- Modify whatsapp-bridge-server.js to accept PORT, TENANT_ID, WEBHOOK_BASE
- Add internal_bp.py with endpoints to provision/destroy bridges via Docker
- Register internal_bp in app.py
- Each tenant gets isolated container, port, and volume
- Refactor whatsapp_service.py to accept bridge_url parameter
- whatsapp_bp.py: remove hardcoded tenant_id=11, use g.tenant_id
- whatsapp_bp.py: webhook now accepts ?tenant_id param with fallback
- config_bp.py: add GET/PUT /config/whatsapp endpoints
- Each tenant can now have its own Baileys bridge URL and settings
Brand catalog now follows the same navigation as the regular catalog:
1. Brands -> 2. Models -> 3. Years -> 4. Engines -> 5. Categories -> 6. Parts
Backend:
- Add /mye-parts endpoint for MYE-specific parts with category filter
- Uses existing /models, /years, /engines, /categories endpoints
Frontend:
- Complete rewrite of brand-catalog.js with breadcrumb navigation
- State machine: brands -> models -> years -> engines -> categories -> parts
- Search and pagination preserved at parts level
- Breadcrumb allows jumping back to any previous step
Add updatefound listener in catalog.html that reloads the page
automatically when a new service worker is activated. This ensures
users get the latest HTML and JS without manual hard refresh.
- Bump cache to nexus-pos-v6 to force invalidation
- HTML pages now use network-first instead of cache-first
This ensures users always get the latest HTML with correct
JS/CSS references (?v=3) instead of stale cached HTML
- Remove HTML pages from APP_SHELL precache (only static assets)
- Keep cache-first for JS/CSS/images
- Add #brandCatalogSearch container in HTML for search inputs
- Move brand search input out of renderBrandList so it persists while typing
- Move parts search input out of renderPartsList so it persists while typing
- Reset now clears search container
- Bump JS cache bust to v=3
Update service worker cache name to nexus-pos-v5 to force cache
invalidation. Add brand-catalog.js to APP_SHELL precache list.
This should resolve stale cached JS causing parse errors.
/vehicle-brands now uses get_brands_for_mode('oem') to return the same
36 North American brands (Mexico/USA/Canada) as the regular catalog flow,
instead of all 619 brands in the database.
brand-catalog.js was missing Authorization header on fetch calls,
causing 401 Unauthorized errors. Now reads pos_token from localStorage
and includes Bearer token in every request. Also handles 401 responses
by redirecting to /pos/login. Bump JS cache bust to v=2.
Backend:
- Add 'search' param to /brand-parts endpoint (filters oem_part_number and name via ILIKE)
- Keep count query accurate with search filter
Frontend (brand-catalog.js):
- Brand search input: filters 619 brands locally while typing
- Parts pagination: Previous/Next buttons with page counter (50 per page)
- Parts search within category: search input + Enter key triggers backend search
- Visual polish: stock badges, empty-state messages, responsive layout
- Loading states and breadcrumbs improved
- Add QWEN (qwen3.6) as primary AI backend with short system prompt
- Hermes remains as fallback with 45s timeout
- Increase QWEN timeout to 35s, max_tokens to 4000
- Add conversation history loading from whatsapp_messages (last 4 msgs)
- Persist detected vehicle in whatsapp_sessions table
- Add 'limpiar chat' / 'nuevo chat' / 'reset' commands to clear history
- Fix CSS conflict: rename whatsapp chat-panel classes to wa-chat-panel
- Fix JS ID conflicts with chat.js widget (waChatPanel, waChatMessages, etc.)
- Improve no-stock response: conversational with alternatives
- Split search_query by | for multi-part lookups
- Add DEMO_PROMPTS.md and DEMO_PROMPTS_V2.md
- Added save_qwen_fitment() in inventory_vehicle_compat.py to centralize
inserting QWEN results into inventory_vehicle_compat
- Simplified inventory_bp.py create_item() and auto_match_item_vehicles()
to use the centralized function, removing duplicated INSERT logic
inventory_vehicle_compat.get_compatibility was trying to JOIN tenant
inventory_vehicle_compat with master tables (model_year_engine, brands,
models, years, engines) on a single tenant connection. Those tables only
exist in the master DB, causing the query to fail silently.
Fix: split into two queries:
1. Fetch MYE IDs from tenant's inventory_vehicle_compat
2. Resolve vehicle details from master DB via ANY(%s)
3. Merge results
Also fixes the argument mismatch: inventory_bp passed (tenant, master,
item_id) but the function only accepted 2 args.
Replaced simple exact-match with 8-layer fallback strategy:
1. Exact normalized part number (parts.oem_part_number)
2. Exact normalized aftermarket part number
3. Exact normalized cross-reference number
4. Partial ILIKE match on OEM numbers
5. Partial ILIKE match on aftermarket numbers
6. Partial ILIKE match on cross-reference numbers
7. Separator-stripped fallback (KYB-343412 → KYB343412)
8. Name-based search on parts.name_part / parts.name_es
and aftermarket_parts.name_aftermarket_parts when no part_number hit
Brand-aware filtering: when brand hint is provided and not 'GENERAL',
only returns MYEs for vehicles of that brand.
Limits: max 20 part IDs per layer, max 200 MYEs total.
Test: BPR5ES + TOYOTA → matched True, 2 parts, 200 MYEs inserted.
- tenant_db.py: add rollback() before returning conn to pool to prevent
'idle in transaction (aborted)' state that exhausts the pool
- tenant_db.py: increase pool maxconn from 10 to 20 for better concurrency
- inventory_vehicle_compat.py: fix column name cross_ref_number ->
cross_reference_number to match actual schema
Nginx auto-serves .min.js when .js is requested with try_files.
The browser had the old file cached with 6M expiry. Adding ?v=2
forces clients to fetch the new version with autoMatchCompat exposed.
- catalog_bp.py: /search and /part/<id> no longer blocked by CATALOG_OEM_ENABLED
These endpoints query the master parts DB and enrich with local stock;
they should work in both local and OEM modes.
- inventory.js: expose autoMatchCompat and removeCompat to window for
onclick handlers in dynamically generated HTML.
- Regenerated inventory.min.js
- Added all brands with vehicles >= 1980 relevant to Mexico-USA-Canada
- Covers: American, Japanese, Korean, German, UK, Italian, French,
Swedish, Spanish, Chinese (with MX presence), Indian, and commercial
- All 96 brands verified against master DB with year >= 1980
- Fixed column names: brands.name_brand, models.name_model, engines.name_engine
- Added fuzzy model matching with ILIKE %%pattern%% for TecDoc-style names
- Removed erroneous double cur.fetchone() that always returned None
- Added retry logic (3 attempts) for QWEN API empty responses
- Added fallback engine-less query when engine description doesn't match DB
- Protected _extract_json against None input
- bnpl_bp.py: APLAZO/Kueski/Clip application workflow (mock)
- erp_bp.py: Aspel/CONTPAQi/SAP/Odoo sync jobs (mock)
- whatsapp_cloud_bp.py: Meta Cloud API webhook, messages, templates
- supplier_portal_bp.py: demand by zone/branch and top-parts analytics
- app.py: register all new blueprints
The chat.js init() template did not include #chatTtsToggle, causing
a runtime TypeError when hasTTS was true. Added the toggle button
inside .chat-header-actions, matching chat-public.js structure.
Regenerated chat.min.js.
- Agrega TTS (speechSynthesis) a chat.js del POS para leer respuestas IA
- Copia lógica de voz completa (STT + TTS) a dashboard/chat-public.js
- Extiende estilos TTS en chat.css y chat-public.css
- Agrega chat widget a 13 templates POS que no lo tenían
- Corrige duplicado de chat.css en diagrams.html
- Minifica assets actualizados
- 73/73 tests pasan
Cambios implementados:
1. Lazy loading de imágenes:
- catalog.js: loading="lazy" decoding="async" en part cards y detail panel
- inventory.js: lazy loading en imagen de detalle de item
2. Minificación de assets:
- scripts/minify-assets.sh: minifica JS (terser) y CSS para POS y Dashboard
- 25 archivos .min.js + 5 .min.css generados en pos/static/
- 14 archivos .min.js + 8 .min.css generados en dashboard/
3. Nginx auto-serve minified:
- try_files $1.min.js antes de servir .js original
- try_files $1.min.css antes de servir .css original
- Transparente para los templates HTML (cero cambios en HTML)
4. Cache warming script:
- scripts/warm_vehicle_cache.py: pobla Redis con vehicle info por batches
- Mitiga DISTINCT ON + 4 JOINs sobre 2B filas
- Corre en background, procesa ~1.5M parts
Tests: 73/73 pasando
Cambios implementados:
1. Redis cache para _classify_cache (catalog_service.py):
- Reemplaza dict in-memory por Redis compartido entre workers
- TTL 5 minutos para clasificación Nexpart
- classify_cache_clear() y classify_cache_stats() actualizados
- Hit rate pasa de ~6% (15 cachés separados) a ~80%+ (cache unificado)
2. Redis cache para vehicle info en smart_search():
- Verifica Redis antes de ejecutar DISTINCT ON + 4 JOINs sobre 2B filas
- Cache miss: query solo para los part_ids faltantes
- TTL 1 hora por part_id
- Impacto: búsquedas repetidas pasan de 500ms–2s a < 50ms
3. Gunicorn gthread (gunicorn.conf.py):
- worker_class = 'gthread' con 4 threads por worker
- 4 workers × 4 threads = 16 requests concurrentes
- max_requests = 1000 para reciclar workers y prevenir memory leaks
Tests: 73/73 pasando
Cambios implementados:
1. Connection pooling (tenant_db.py):
- psycopg2.pool.ThreadedConnectionPool para master y tenants
- Wrapper _PooledConnection que devuelve al pool en .close()
- Cero cambios en blueprints (backward compatible)
2. Tabla inventory_stock_summary + triggers (v3.2):
- O(1) stock lookup en vez de SUM() sobre historial completo
- Trigger AFTER INSERT en inventory_operations recalcula stock
- Poblada inicialmente en ambos tenants
- Refactor en 6 archivos de servicios para usar la nueva tabla
3. Fix N+1 en process_sale (pos_engine.py):
- Precarga retail_price en bulk query FOR UPDATE
- Elimina SELECT individual por item en loop
4. Índices críticos:
- idx_parts_name_part + pattern_ops (master)
- idx_inv_ops_inventory_branch_created (tenants)
- idx_wi_part_stock_positive (master, ya existía desde Fase 1)
Tests: 73/73 pasando (compat + fase3 + fase5 + fase6)
Migración: v3.2_db_performance.sql
- Nueva tabla inventory_vehicle_compat (v3.1)
- Motor inventory_vehicle_compat.py: auto-match + gestión manual
- catalog_service.get_parts_local() ahora incluye piezas locales vinculadas
- inventory_bp: auto-match en create/update + endpoints REST /vehicles
- Frontend catalog.js: badge 'Stock Local' para piezas nativas del tenant
- Frontend inventory.js: panel de vehículos compatibles con auto-match
- Tests: test_compatibility.py (9/9 pasan)
Piezas locales aparecen en navegación por vehículo aunque no estén en TecDoc.
Auto-match busca part_number en parts/aftermarket_parts y copia MYEs compatibles.
Replaced hardcoded values (4,817 SKUs, $2.4M, 1,284 clients, $842,190
invoiced, $2,847,320 assets, etc.) with $0 / 0 placeholders with IDs
so the JS can populate them from real API data.
Affected: inventory.html, customers.html, invoicing.html, accounting.html
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Eliminado: diagrams_bp, ruta /pos/diagrams, link en sidebar,
boton "Ver diagramas" en catalogo. Los archivos SVG y blueprint
se mantienen en el repo por si se reactivan en el futuro.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
