application.yaml

security:
  anonymous-user-enabled: false
  anonymous-user-session-timeout: 300 # Inactivity in seconds after which an anonymous session is removed from chat memory
  admin-write-only-enabled: false # With this enabled, only admins have write-access to knowledge, i.e. regular users cannot add or modify knowledge

frontend:
  # Provide the configuration for the frontend OIDC client
  oidc:
    client-id: llamara-frontend
    # Default paths for Keycloak
    authorization-path: /protocol/openid-connect/auth
    logout-path: /protocol/openid-connect/logout
    token-path: /protocol/openid-connect/token
  # Provide legals links for the frontend
  legals:
    # imprint: "https://example.com/imprint" # Link to imprint
    # privacy-policy: "https://example.com/privacy-policy" # Link to privacy policy

chat:
  memory:
    # Maximum number of messages to keep in memory, i.e. to present to the LLM when sending a prompt.
    # Keep in mind: bigger memory = more input tokens = more cost
    window: token # Supported values: message, token
    max-messages: 5
    max-tokens: 1000
    tokenizer:
      provider: openai # Supported providers: openai (as Azure is using OpenAI models, we can use OpenAI tokenizer)
      model: gpt-4o-mini
  history:
    max-messages: 50 # Maximum number of messages to keep in history, i.e. to present them to the user.
  models:
    # Supported model providers: azure, openai, ollama
    # Supported model parameters: temperature, top-p, frequence-penalty, presence-penalty, max-tokens
    # See https://platform.openai.com/docs/api-reference/chat/create for parameter description.
    # (Azure) OpenAI supports all model parameters, Ollama doesn't support frequence-penalty & max-tokens.

    # Azure OpenAI models need resource-name, model config and AZURE_API_KEY env variable
    #- uid: azure-gpt4o-mini
    #  provider: azure
    #  resource-name: my-azure-resource # Azure OpenAI resource name
    #  model: gpt-4o-mini-2024-07-18 # Azure OpenAI deployment name
    #  label: Azure GPT 4o Mini
    #  description: Azure OpenAI GPT 4 Omni Mini
    # OpenAI models need model config and OPENAI_API_KEY key env variable
    - uid: gpt4o-mini
      provider: openai
      model: gpt-4o-mini
      label: GPT 4o Mini
      description: OpenAI GPT 4 Omni Mini
    # OpenAI reasoning models require disabling the system prompt and setting the temperature to 1.0
    #- uid: o1-mini
    #  provider: openai
    #  model: o1-mini
    #  label: o1 Mini
    #  description: OpenAI o1 Mini
    #  system-prompt-enabled: false
    #  temperature: 1.0
    # Ollama models need base URL and model config
    #- uid: llama3-1
    #  provider: ollama
    #  base-url: http://localhost:11435
    #  model: llama3.1
    #  label: LLAMA 3.1
    #  description: Meta LLAMA 3.1 8b

embedding:
  store:
    type: qdrant # Supported types: qdrant
    host: localhost
    port: 6334 # Qdrant gRPC port
    tls: false
    collection-name: text-embedding-3-large
    vector-size: 3072
  model:
    provider: openai # Supported providers: azure, openai, ollama
    # Azure OpenAI models need resource-name, model config and AZURE_API_KEY env variable
    # OpenAI models need model config and OPENAI_API_KEY key env variable
    # Ollama models need base URL and model config
    model: text-embedding-3-large

ingestion:
  document-splitter:
    type: paragraph # Supported types: line, paragraph, recursive
    max-segment-size: 1000
    max-overlap-size: 200

retrieval:
  # The prompt template is used to inject the retrieved context into the user prompt and instruct the chat model to generate a response.
  # It must contain {{userMessage}} and {{contents}} variables.
  prompt-template: |-
    {{userMessage}}
    
    Answer preferably using the following information.
    When answering using the following information, ALWAYS provide the source of the information for EACH paragraph.
    The source is specified by the knowledge_id and MUST be in the following JSON format: { "knowledge_id": knowledge_id }.
    Again: ALWAYS provide the source of the information for EACH paragraph.
    
    If, I repeat ONLY if explicitly asked for citation, cite the relevant parts word-by-word and provide the source of the information.
    If the citation is in a different language than the question, provide a translation.
    If you are not asked for citation, answer in your own words.
    
    If the following information is not sufficient to answer, answer the question with your own knowledge.
    In that case, you MUST state that you are answering based on your own knowledge.
    
    Information:
    {{contents}}
  # The missing knowledge prompt template is used to prompt the chat model for its own knowledge if no context was retrieved.
  # It must contain the {{userMessage}} variable.
  missing-knowledge-prompt-template: |-
      {{userMessage}}
      
      Answer based on your own knowledge.
      You MUST state that you are answering based on your own knowledge.

quarkus:
  minio:
    # The name of the MinIO bucket to use.
    bucket-name: llamara