r2rpy/py/core/configs/ollama.toml

[app]
# LLM used for internal operations, like deriving conversation names
fast_llm = "ollama/llama3.1" ### NOTE - RECOMMENDED TO USE `openai` with `api_base = "http://localhost:11434/v1"` for best results, otherwise `ollama` with `litellm` is acceptable

# LLM used for user-facing output, like RAG replies
quality_llm = "ollama/llama3.1"

# LLM used for ingesting visual inputs
vlm = "ollama/llama3.1" # TODO - Replace with viable candidate

# LLM used for transcription
audio_lm = "ollama/llama3.1" # TODO - Replace with viable candidate


# Reasoning model, used for `research` agent
reasoning_llm = "ollama/llama3.1"
# Planning model, used for `research` agent
planning_llm = "ollama/llama3.1"

[embedding]
provider = "ollama"
base_model = "mxbai-embed-large"
base_dimension = 1_024
batch_size = 128
concurrent_request_limit = 2

[completion_embedding]
provider = "ollama"
base_model = "mxbai-embed-large"
base_dimension = 1_024
batch_size = 128
concurrent_request_limit = 2

[agent]
tools = ["search_file_knowledge"]

[completion]
provider = "litellm"
concurrent_request_limit = 1

  [completion.generation_config]
  temperature = 0.1
  top_p = 1
  max_tokens_to_sample = 1_024
  stream = false
  api_base = "http://localhost:11434/v1"