# Copy to .env (local) or set as HuggingFace Space secrets. # Every engine is optional — the app exposes whichever ones are configured. # Google Gemini (free tier, no credit card): https://aistudio.google.com/apikey GEMINI_API_KEY= # GEMINI_MODEL=gemini-2.5-flash-lite # Groq (free tier, no credit card): https://console.groq.com/keys GROQ_API_KEY= # GROQ_MODEL=llama-3.3-70b-versatile # Local NLLB / MADLAD (CTranslate2). Defaults are fine; override for GPU: # CT2_DEVICE=cuda # CT2_COMPUTE_TYPE=int8 # Local LLM via Ollama (unlocks Qwen / Gemma / TranslateGemma / Aya / Llama). # Install Ollama, then `ollama pull `. Engine appears when Ollama runs. # OLLAMA_MODEL=qwen2.5 # OLLAMA_HOST=http://localhost:11434 # MADLAD-400 (local, 400+ languages). Convert once (see README), then it appears: # MADLAD_MODEL_DIR=models/madlad400-3b-mt-int8 # Comma-separated allowed origins for the frontend (dev: http://localhost:3000) # CORS_ORIGINS=*