.env.template

## Core API Keys
OPENAI_API_KEY=""                       # Used for LLM (GPT-4o) and text embeddings via OpenAI.
HUGGING_FACE_TOKEN=""                   # Used for Named Entity Recognition (NER) to extract location names.
OPENWEATHER_API_KEY=""                  # Used to fetch weather data from OpenWeatherMap.

## Optional Integrations
LITERAL_API_KEY=""                      # (Optional) For observability with Literal.dev. Along with Authentication Enabled, User can maintain chat persistence. Remove this key to disable chat persistence.

## Chainlit Authentication (JWT-based)
ENABLE_AUTH=False                       # Set to true to require login (username=admin/password=1234); false for public access. 
CHAINLIT_AUTH_SECRET=""                 # Used to sign JWT tokens in Chainlit for user sessions.

## OAuth2 via Auth0 (Optional - enable if using Auth0 login)
OAUTH_AUTH0_CLIENT_ID=""                # Auth0 client ID for OAuth login.
OAUTH_AUTH0_CLIENT_SECRET=""            # Auth0 client secret.
OAUTH_AUTH0_DOMAIN=""                   # Auth0 domain (e.g., your-tenant.auth0.com).

## ChromaDB Group Preview (Startup Logs)
GROUP_PREVIEW_ENABLED=True              # Whether to display grouped document summary when app starts.
GROUP_PREVIEW_LIMIT=5                   # Max number of document groups to preview (e.g., 5 manuals).

## Top-K Retrieval Results
TOP_K_RETRIEVAL_RESULTS=3               # Number of top similar chunks to log after retrieval (terminal log only)

## Image Base Path for Reference Previews
IMAGE_BASE_PATH=./data/images           # Base folder for reference page images. Format: ./base/{doc_id}/{page_id}.png

## Toggle whether to display reference previews after answering.
SHOW_REFERENCES=True

## If True, expects OPENAI_API_KEY via Chainlit `user_env`. 
ENABLE_USER_ENV=False                   # NOTE: This must be synced with your config.toml's [user_env] block.

## Decoupled Retrieval & Synthesis Feature Toggle
ENABLE_DECOUPLED_SYNTHESIS=True         # Enable/disable the feature entirely

## Chunk Grouping Parameters
CHUNK_GROUP_FIELD=parent_id             # Field to use for grouping chunks (default: parent_id)
CHUNK_SYNTHESIS_MIN=1                   # Minimum number of chunks required for synthesis
CHUNK_SYNTHESIS_MAX=6                   # Maximum number of chunks to group together
CHUNK_SYNTHESIS_ORDERED=True            # Whether to sort chunks by chunk_index before synthesis

## Short-Term Chat Context
ENABLE_CHAT_HISTORY_CONTEXT=True        # Enable short-term memory using cl.user_session
MAX_CHAT_TURNS=3                        # Number of recent user/assistant turns to include (each turn = 2 messages)

## Weather Bot 
WEATHER_KEYWORD_THRESHOLD=1             # Minimum keyword hits before triggering weather query detection

## Session-Based Evaluation
ENABLE_SESSION_EVAL=True                # Whether to enable session-based evaluation logging for RAG queries
EVAL_TRIGGER_KEYWORD="!freeze_eval"     # Keyword the user can send to freeze the evaluation and save logs/metrics
EVAL_OUTPUT_DIR=./evals                 # Directory where evaluation logs and metrics will be stored
EVAL_OUTPUT_FORMAT=json                 # File format for saved logs: json | csv | both
EVAL_MAX_LOGGED_QUERIES=0               # Max number of queries to retain in memory (0 = unlimited)
EVAL_ZIP_EXPORT=True                    # Optional zip export of evaluation sessions

## Evaluation Replay Settings
EVAL_REPLAY_ENABLED=True                # Enables !replay_eval <filename> command to replay saved sessions
EVAL_REPLAY_TOP_K=3                     # Number of top-k chunks to show per query during replay
EVAL_REPLAY_TRIGGER="!replay_eval"      # Trigger keyword to initiate a replay command

## Evaluation List Settings
EVAL_LIST_ENABLED=True                  # Enables !list_evals command to list available evaluation logs
EVAL_LIST_TRIGGER="!list_evals"         # Trigger to list all saved evaluation sessions in the ./evals folder
EVAL_LIST_LIMIT=10                      # Max number of evaluation logs to show when listing sessions  
EVAL_LIST_FORMAT=both                   # File format to display summaries when listing evaluations: json, csv, or both

## Evaluation Comparison Settings
EVAL_COMPARE_ENABLED=True               # Enables !compare_eval <file1> <file2> to compare two evaluation logs
EVAL_COMPARE_TRIGGER="!compare_eval"    # Trigger keyword for comparison

# Whether to offer download button after evaluation zip is created
EVAL_OFFER_DOWNLOAD=True    

## QA Dataset Generation Settings
ENABLE_QA_DATASET_GENERATION=False      # Enable or disable synthetic QA dataset generation (used for retrieval evaluation)
QA_NUM_QUESTIONS=1                      # Number of questions to generate per chunk for synthetic QA dataset
QA_SOURCE_FILE_FILTER="MF-194"          # Optional filter to restrict QA generation to a specific manual (e.g., MF-194)
QA_EMBEDDING_MODE_FILTER="full"         # Optional filter for embedding mode ("chunked" or "full")
                                        # WARNING: If left empty or set to "chunked" on large chunked+full collections, synthetic QA generation may trigger more OpenAIAPI calls and hit rate limits.
                                        # To reduce load, set to "full" or filter by a smaller manual.
QA_MATCH_THRESHOLD=0.15                 # Minimum semantic similarity (cosine) to match user query with synthetic QA (used in fuzzy matching for retrieval metrics)
                                        # Lower values allow looser matches (e.g., paraphrased questions); higher values enforce stricter alignment.
RETRIEVAL_DEBUG_MODE=False              # TODO