config main 'main'
	option enabled '0'
	option api_port '8080'
	option api_host '0.0.0.0'
	option data_path '/srv/localai'
	option models_path '/srv/localai/models'
	option memory_limit '2G'
	option threads '4'
	option context_size '2048'
	option debug '0'
	option cors '1'

# GPU settings (experimental on ARM64)
config gpu 'gpu'
	option enabled '0'
	option backend 'vulkan'

# Default model to load on startup
config model 'default'
	option enabled '1'
	option name 'phi-2'
	option backend 'llama-cpp'

# Model presets
config preset 'phi2'
	option name 'phi-2'
	option url 'https://huggingface.co/TheBloke/phi-2-GGUF/resolve/main/phi-2.Q4_K_M.gguf'
	option size '1.6G'
	option type 'text-generation'
	option description 'Microsoft Phi-2 - Compact and efficient'

config preset 'mistral'
	option name 'mistral-7b'
	option url 'https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_K_M.gguf'
	option size '4.1G'
	option type 'text-generation'
	option description 'Mistral 7B Instruct - High quality assistant'

config preset 'tinyllama'
	option name 'tinyllama'
	option url 'https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf'
	option size '669M'
	option type 'text-generation'
	option description 'TinyLlama 1.1B - Ultra-lightweight'

config preset 'gte_small'
	option name 'gte-small'
	option url 'https://huggingface.co/Supabase/gte-small/resolve/main/model.onnx'
	option size '67M'
	option type 'embeddings'
	option description 'GTE Small - Fast embeddings'