A config file for using full self hosted stuff with [r2r](https://github.com/SciPhi-AI/R2R). Pain in the butt to figure out

ollama.yaml

[app]
# LLM used for internal operations, like deriving conversation names
fast_llm = "ollama/gemma3:12b"

# LLM used for user-facing output, like RAG replies
quality_llm = "ollama/gemma3:12b"

# LLM used for ingesting visual inputs
vlm = "ollama/gemma3:12b"

# Reasoning model, used for `research` agent
reasoning_llm = "ollama/gemma3:12b"
# Planning model, used for `research` agent
planning_llm = "ollama/gemma3:12b"

[ocr]
provider = "ollama"
model = "ollama/gemma3:12b"

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

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

[agent]
tools = ["search_file_knowledge"]

[graph_creation_settings]
enabled = true
graph_extraction_prompt = "graph_get"


[completion]
provider = "litellm"
concurrent_request_limit = 1

  [completion.generation_config]
  temperature = 0.1
  top_p = 1
  max_tokens_to_sample = 1_024
  stream = false


[orchestration]
provider = "simple"

SciPhi-AI/R2RR2R - SoTA production-ready AI retrieval system. Agentic Retrieval-Augmented Generation (RAG) with a RESTful API.