config: add env schema and localization foundation

.env.example

@@ -31,25 +31,85 @@ ONLY_FOR_CREATOR_MODE=false
 # Use user names in AI prompts
 USE_NAMES_IN_PROMPT=true
 
+# Custom system prompt for AI (or put it into data/SYSTEM_PROMPT.md)
+SYSTEM_PROMPT=
+
 # Maximum photo size in pixels
 MAX_PHOTO_SIZE=1280
 
+# Directory with localization JSON files
+LOCALES_DIR=locales
+
 # ============================================
 # AI MODELS CONFIGURATION (Optional)
 # ============================================
 
 # Google Gemini
 GEMINI_API_KEY=
+# google: official Gemini API via @google/genai; openai: OpenAI-compatible Gemini endpoint; auto: infer from GEMINI_BASE_URL
+GEMINI_API_MODE=google
 GEMINI_MODEL=gemini-2.5-flash
+GEMINI_IMAGE_MODEL=gemini-2.5-flash-image
+GEMINI_TRANSCRIPTION_MODEL=gemini-2.5-flash
+GEMINI_TTS_MODEL=gemini-2.5-flash-preview-tts
+GEMINI_TTS_VOICE=Kore
+GEMINI_MAX_CONCURRENT_REQUESTS=3
 
 # Mistral AI
 MISTRAL_API_KEY=
 MISTRAL_MODEL=mistral-small-latest
+MISTRAL_TRANSCRIPTION_MODEL=voxtral-mini-latest
+MISTRAL_TTS_MODEL=
+MISTRAL_TTS_VOICE_ID=
+MISTRAL_MAX_CONCURRENT_REQUESTS=3
 
 # Ollama (Local AI Model)
 OLLAMA_ADDRESS=
-OLLAMA_MODEL=
+OLLAMA_CHAT_MODEL=
+OLLAMA_IMAGE_MODEL=
+OLLAMA_THINK_MODEL=
+OLLAMA_AUDIO_MODEL=gemma4:e2b
 OLLAMA_API_KEY=
+OLLAMA_EMBEDDING_MODEL=nomic-embed-text
+OLLAMA_RAG_CHUNK_SIZE=1400
+OLLAMA_RAG_CHUNK_OVERLAP=220
+OLLAMA_RAG_TOP_K=8
+OLLAMA_RAG_MAX_CONTEXT_CHARS=14000
+OLLAMA_RAG_MIN_SCORE=0.12
+OLLAMA_RAG_MAX_ARCHIVE_FILES=200
+OLLAMA_RAG_MAX_ARCHIVE_BYTES=52428800
+OLLAMA_RAG_MAX_ARCHIVE_DEPTH=2
+OLLAMA_MAX_CONCURRENT_REQUESTS=1
 
-# Custom system prompt for AI
-SYSTEM_PROMPT=
+# OpenAI
+OPENAI_API_KEY=
+OPENAI_BASE_URL=
+OPENAI_MODEL=gpt-4.1-nano
+OPENAI_IMAGE_MODEL=gpt-image-1-mini
+OPENAI_TRANSCRIPTION_MODEL=gpt-4o-mini-transcribe
+OPENAI_TTS_MODEL=gpt-4o-mini-tts
+OPENAI_TTS_VOICE=alloy
+OPENAI_TTS_INSTRUCTIONS=
+OPENAI_MAX_CONCURRENT_REQUESTS=3
+
+# Per-capability AI endpoint overrides
+# Pattern:
+# <PROVIDER>_<CAPABILITY>_MODEL=
+# <PROVIDER>_<CAPABILITY>_BASE_URL=
+# <PROVIDER>_<CAPABILITY>_API_KEY=
+#
+# Providers: OLLAMA, GEMINI, MISTRAL, OPENAI
+# Capabilities: CHAT, VISION, OCR, THINKING, EXTENDED_THINKING, TOOLS, AUDIO,
+# DOCUMENTS, OUTPUT_IMAGES, SPEECH_TO_TEXT, TEXT_TO_SPEECH
+# Endpoint aliases are also supported: *_ENDPOINT and *_ADDRESS.
+# Provider-wide fallback endpoints: OPENAI_BASE_URL, MISTRAL_BASE_URL,
+# GEMINI_BASE_URL, OLLAMA_ADDRESS or OLLAMA_BASE_URL.
+# Capability aliases are also supported: IMAGE, THINK, RAG, EMBEDDING,
+# TRANSCRIPTION, STT, TTS.
+#
+# Examples:
+# OPENAI_SPEECH_TO_TEXT_MODEL=gpt-4o-mini-transcribe
+# OPENAI_SPEECH_TO_TEXT_BASE_URL=https://api.openai.com/v1
+# OPENAI_SPEECH_TO_TEXT_API_KEY=
+# MISTRAL_TTS_BASE_URL=
+# OLLAMA_DOCUMENTS_ADDRESS=http://localhost:11434