Environment Variables

The script sets up various environment variables for AGiXT. Some key variables and their purposes include:

• AGIXT_API_KEY: API key for AGiXT (automatically generated if not provided)
• AGIXT_URI: URI for the AGiXT API
• AGIXT_BRANCH: AGiXT version to use (stable or dev)
• AGIXT_AUTO_UPDATE: Whether to automatically update AGiXT
• AGIXT_RLHF: Enable reinforcement learning from human feedback (thumbs up/down buttons in UI)
• AGIXT_SHOW_SELECTION: Controls which dropdowns are shown in the UI. Comma separated values. (default is agent,conversation)
• AUTH_PROVIDER: Authentication provider (none or magicalauth)
• INTERACTIVE_MODE: Should always be set to chat (form mode is experimental)
• THEME_NAME: UI color scheme (default, christmas, conspiracy, doom, easter, halloween, valentines)
• DATABASE_TYPE: Type of database to use (sqlite or postgres)
• UVICORN_WORKERS: Number of workers running on the application. Default is 10.

Environment variables specific to ezLocalai:

Note: If you do not have an NVIDIA GPU, the correct CUDA drivers, or enough VRAM, ezLocalai will still work running on CPU, but it will be slower.

• EZLOCALAI_URI: URI for ezLocalai. Default is http://{local_ip}:8091.
• DEFAULT_MODEL: Default language model for ezLocalai. Default is QuantFactory/dolphin-2.9.2-qwen2-7b-GGUF. This model takes ~9GB VRAM at 32k max tokens, lower the max tokens if you have less VRAM or use a different model.
• VISION_MODEL: Vision model for ezLocalai. Default is deepseek-ai/deepseek-vl-1.3b-chat. This model takes ~3GB VRAM in addition to the language model.
• LLM_MAX_TOKENS: Maximum number of tokens for language models. Default is 32768. Lower this value if you have less VRAM.
• WHISPER_MODEL: Whisper model for speech recognition. Default is base.en for a fast English model.
• GPU_LAYERS: Number of GPU layers to use. Default is -1 for all.

For a complete list of environment variables and their default values, please refer to the get_default_env_vars() function in the start.py script.