config.yaml

# LLM (Language Model) API Parameters
llm_config:
  # Main language model for generating detailed responses
  default_llm: "anthropic.claude-3-5-sonnet-20240620-v1:0"

  # Faster, lighter model used for conversation summarization
  fast_llm: "anthropic.claude-3-haiku-20240307-v1:0"

  # Model used for generating embeddings of text chunks
  embed_model: "cohere.embed-english-v3"

  # AWS Bedrock configuration
  # Note: Values starting with OS_ENV_ will be loaded from environment variables
  # AWS region where Bedrock service is located
  aws_region: "OS_ENV_AWS_REGION"

  # AWS credentials for accessing Bedrock API
  aws_access_key_id: "OS_ENV_AWS_ACCESS_KEY_ID"
  aws_secret_access_key: "OS_ENV_AWS_SECRET_ACCESS_KEY"

  # GitHub personal access token for repository access
  # Required for accessing private repos and avoiding rate limits
  github_token: "OS_ENV_GITHUB_TOKEN"

# Usage Limits
# These limits help prevent abuse and ensure system stability
limits:
  # Maximum number of questions a user can ask in a single browser session
  max_questions: 1000

  # Maximum number of words a user can submit across all questions
  max_words: 100000

  # Maximum tokens allowed in a single message (input)
  # 4096 tokens is approximately 3000-4000 words
  max_message_tokens: 4096

  # Maximum tokens maintained in conversation history
  # This affects context window and memory usage (summarization is used to handle longer histories)
  max_total_tokens: 8192

# Data Sources Configuration
# These sources form the knowledge base for the RAG system
sources:
  # List of URLs to scrape for building the knowledge base
  urls:
    # BossDB Documentation
    # - "https://metadata.bossdb.org/DataModel"     # Core data model documentation
    # - "https://metadata.bossdb.org/SampleCode"    # Example code and usage

    # Academic Papers and Publications
    - "https://www.frontiersin.org/articles/10.3389/fninf.2022.828787/full"  # BossDB platform paper
    - "https://www.biorxiv.org/content/10.1101/2020.05.15.098707v1.full"     # Technical implementation
    - "https://www.biorxiv.org/content/10.1101/2022.06.01.494307v1.full"     # Use cases
    
    # Tool Documentation
    - "https://github.com/jhuapl-boss/intern/wiki"                  # Intern client wiki
    - "https://github.com/jhuapl-boss/intern/tree/master/docs"      # Detailed documentation
    - "https://github.com/jhuapl-boss/intern/tree/master/examples"  # Usage examples
    - "https://github.com/aplbrain/bossdb_cookbook"                 # Cookbook recipes
    
    # API Documentation
    - "https://api.metadata.bossdb.org/openapi.json"         # Metadata API spec
    - "https://api.motifstudio.bossdb.org/openapi.json"     # MotifStudio API spec
    - "https://docs.theboss.io/reference/cutout-get"        # Core API documentation
    - "https://api.metadata.bossdb.org/api/v1/collections"  # Collections endpoint
  
  # GitHub Organizations to pull readmes from
  # These orgs contain repositories with relevant BossDB tools and documentation
  github_orgs:
    - "jhuapl-boss"
    - "aplbrain"

index_settings:
  force_reload: false  # Set to true to force rebuild (of only docs in config)
  # Set to false to process all documents, only gets new vector embedding if doc changed.
  # If true will not process existing documents, even if out of date as it will not even download them again.
  # Does not remove docs if no longer in config, use force_reload to get a clean start.
  incremental: false