.env.sample

# General

API_KEY=your_api_key_here
GPU_API_URL=http://gpu-api:8888
LOG_LEVEL=INFO
TZ=Europe/Skopje
WORKERS=4


# Database

POSTGRES_USER=bot
POSTGRES_PASSWORD=bot
POSTGRES_DB=bot
POSTGRES_PORT=5432
DATABASE_URL=postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@db:${POSTGRES_PORT}/${POSTGRES_DB}
DATABASE_POOL_MIN_SIZE=1
DATABASE_POOL_MAX_SIZE=10

# Resumable chat stream store (server-only; do not expose to browser)
RESUMABLE_STREAM_REDIS_URL=redis://stream-store:6379

# Web login (Auth.js OAuth; configure Google, Microsoft, or both)
AUTH_URL=http://localhost:3000
AUTH_SECRET=replace_with_auth_secret
AUTH_GOOGLE_ID=your_google_oauth_client_id
AUTH_GOOGLE_SECRET=your_google_oauth_client_secret
AUTH_MICROSOFT_ENTRA_ID_ID=your_microsoft_entra_application_client_id
AUTH_MICROSOFT_ENTRA_ID_SECRET=your_microsoft_entra_client_secret_value
# Use /common/ to allow personal, work, and school Microsoft accounts; use /<tenant-id>/ to restrict to one tenant.
AUTH_MICROSOFT_ENTRA_ID_ISSUER=https://login.microsoftonline.com/common/v2.0

# pgAdmin

PGADMIN_EMAIL=test@test.com
PGADMIN_PASSWORD=12345

# MCP

MCP_SERVERS='[{"name":"local","url":"http://local-mcp:8808/mcp","transport":"streamable_http","api_key":"SystemPass","allowed_tools":[],"blocked_tools":[]}]'

# BYOK credential encryption

CREDENTIAL_ENCRYPTION_KEY=your_byok_encryption_key_here
# Comma-separated custom provider endpoints users may select for BYOK credentials.
BYOK_ALLOWED_BASE_URLS=

# Chat

CHAT_HISTORY_MAX_TURNS=10
RERANKER_MIN_SCORE=0.1
SOURCE_RERANKER_MIN_SCORE=0.3

# Sponsored model (API service only; disabled unless explicitly enabled)
# Never reuse API_KEY here. This provider key is a separate server-only secret.

SPONSORED_MODEL_ENABLED=false
SPONSORED_MODEL_ID=gpt-5.6-luna
SPONSORED_MODEL_PROVIDER=openai
SPONSORED_MODEL_API_KEY=
SPONSORED_MODEL_BASE_URL=
# Model name sent upstream; this is not the public catalog ID.
SPONSORED_MODEL_UPSTREAM_MODEL=
SPONSORED_DAILY_USER_LIMIT=5
# Required positive global requests-per-UTC-day cap when sponsorship is enabled.
SPONSORED_DAILY_GLOBAL_LIMIT=
# Requested output tokens are hard-clamped to 8192 for sponsored inference.
SPONSORED_MAX_OUTPUT_TOKENS=1024
# Lease duration used to prevent concurrent sponsored requests for one user.
SPONSORED_REQUEST_LEASE_SECONDS=600