- AiLogger.php: new service, inserts into ai_logs on every AI call
- ai_logs table: added to migrate.php (provider, model, call_type, tokens, duration_ms)
- AiBot: logs chat, media_response, nlu calls with timing and token counts
- MediaTranscriber: logs whisper, geminiAudio, openaiVision, geminiVision, claudeVision
- /admin/ai-logs: paginated table with provider/company filter
- Layout.php: added IA Logs nav link
- index.php: require AiLogger before AiBot; added GET /admin/ai-logs route
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- WpWebhook: inject media_id into context before calling BotRouter so
MediaTranscriber can download audio/image (was passing caption='')
- BotRouter: use context['media_id'] for transcription (non-breaking fallback)
- AiBot::processMedia: fix canUpload to include permType 1 (solo reporta)
- AiBot::buildFlowCatalog: include text flows that have nlu_description set
- DashboardController + index.php: add nlu_description field per flow card
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- NluRouter in AiBot: builds permission-filtered flow catalog, asks AI
to route user intent to exact flow key or return chat response
- MediaTranscriber: downloads WA media, transcribes audio (Whisper/Gemini),
extracts image intent (OpenAI/Gemini/Claude vision)
- BotRouter hybrid: text → NormalBot → NLU fallback if no match;
audio/image → transcribe → NLU route; sends transcript preview first
- Admin IA tab: NLU checkbox with description
- Per-company nlu flag saved in config_json
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Global settings: custom AI panel with OpenAI/Gemini/Claude each with
API key, model selector, and live test-connection button (no save needed)
- Per-company IA tab: same three providers + fallback to global when empty
- AiBot: reads per-company config_json first, falls back to env() global
- Added callClaude() for Anthropic Messages API
- More OpenAI models: gpt-4.1, gpt-4.1-mini
- test-ai endpoint: accepts provider/api_key/model from POST for live test
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Added gemini-2.5-flash to the admin panel selector (first/default option)
and updated the fallback default in AiBot and migrate seed.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Text/interactive/button messages → NormalBot only, never escalates to AI
- Image/audio/video/document → AiBot.processMedia() if ai_for_media enabled,
else falls back to category greeting menu
- AiBot.processMedia() uses permission_type context so AI acknowledges uploads
for perm 3 and redirects others to their menu
- WpWebhook.handleMedia now calls BotRouter so the bot responds to media
- Admin UI: checkbox toggle "Procesar imágenes y archivos con IA" in AI tab
- categoryMenuFallback() extracted in BotRouter as shared helper
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Returning an error string caused the bot to send "Lo siento, tengo
problemas para procesar tu mensaje" to the user whenever OpenAI/Gemini
was unavailable or misconfigured. Now it returns null so BotRouter can
fall back to the category greeting menu instead of sending the error.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Settings: agrega Gemini al selector de proveedor, campos gemini_api_key y gemini_model
- Modelos: gemini-2.0-flash, gemini-1.5-flash, gemini-1.5-pro
- AiBot::callGemini(): llama generateContent API con system_instruction y historial
- migrate.php: seed defaults gemini_api_key y gemini_model
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
