.env.example

# -----------------------------------------------------------------------------
# Reflexion RAG Engine - Environment Configuration Example
#
# Instructions:
# 1. Copy this file to a new file named `.env` in the project root.
# 2. Replace placeholder values (like `your_token_here`) with your actual credentials.
# 3. Adjust settings as needed for your specific deployment.
# -----------------------------------------------------------------------------

# --- Authentication ---
# Your GitHub Personal Access Token with 'repo' and 'read:org' scopes for accessing GitHub Models.
GITHUB_TOKEN=your_github_pat_token_here

# --- LLM Orchestration ---
# Models can be selected from the available list at https://github.com/features/models

# Primary model for generating initial and follow-up answers.
LLM_MODEL=openai/gpt-4.1-mini
LLM_TEMPERATURE=0.2
LLM_MAX_TOKENS=3000

# Model used for self-evaluation, confidence scoring, and gap analysis.
EVALUATION_MODEL=cohere/Cohere-command-r
EVALUATION_TEMPERATURE=0.7
EVALUATION_MAX_TOKENS=1000

# Model used to synthesize the final, comprehensive answer from multiple reflexion cycles.
SUMMARY_MODEL=mistral-ai/mistral-medium-2505
SUMMARY_TEMPERATURE=0.7
SUMMARY_MAX_TOKENS=4000

# --- Embedding Model (Azure AI Inference) ---
# The model used to create vector embeddings for documents and queries.
EMBEDDING_MODEL=text-embedding-3-large
# The endpoint for the Azure AI Inference service.
EMBEDDING_ENDPOINT=https://models.inference.ai.azure.com
# Number of documents to embed in a single batch request.
EMBEDDING_BATCH_SIZE=100

# --- SurrealDB Configuration ---
# The WebSocket URL for your SurrealDB instance (cloud or local).
# Example Cloud: wss://your-instance.surreal.cloud
# Example Local: ws://localhost:8000
SURREALDB_URL=wss://your-instance.surreal.cloud
# SurrealDB namespace and database name.
SURREALDB_NS=rag
SURREALDB_DB=rag
# Credentials for authenticating with SurrealDB.
SURREALDB_USER=your_username
SURREALDB_PASS=your_password

# --- Reflexion Engine ---
# The maximum number of self-correction cycles the engine will perform for a single query.
MAX_REFLEXION_CYCLES=3
# The confidence score (0.0 to 1.0) required to complete the reflexion loop.
CONFIDENCE_THRESHOLD=0.90
# The number of documents to retrieve in the initial RAG cycle.
INITIAL_RETRIEVAL_K=5
# The number of documents to retrieve in subsequent reflexion cycles.
REFLEXION_RETRIEVAL_K=3

# --- Web Search (Google Custom Search) ---
# Controls when to use web search. Options: off, initial_only, every_cycle.
WEB_SEARCH_MODE=every_cycle
# The number of search results to retrieve from the Google API.
WEB_SEARCH_RESULTS_COUNT=3
# Timeout in seconds for web search requests.
WEB_SEARCH_TIMEOUT=30
# The number of top web search results to use for context.
WEB_SEARCH_RETRIEVAL_K=3
# Whether to use Crawl4AI to extract full content from web pages.
WEB_SEARCH_ENABLE_CONTENT_EXTRACTION=true
# Minimum character length for extracted web content to be considered valid.
WEB_SEARCH_MIN_CONTENT_LENGTH=200
# Maximum length for a web page title.
WEB_SEARCH_MAX_TITLE_LENGTH=80

# Your Google Custom Search API Key and Search Engine ID.
GOOGLE_API_KEY=your_google_api_key_here
GOOGLE_CSE_ID=your_google_cse_id_here

# --- Document Processing ---
# The target size (in characters) for document chunks.
CHUNK_SIZE=1500
# The number of characters to overlap between adjacent chunks to maintain context.
CHUNK_OVERLAP=100

# --- Performance & Caching ---
# Enable or disable the LRU cache for reflexion loop results to speed up repeated queries.
ENABLE_MEMORY_CACHE=true
# The maximum number of query results to store in the cache.
MAX_CACHE_SIZE=100