env.example

# MONITOR repository environment variables
# Copy this file to .env and fill in your own values.

# =============================================================================
# UI Backend
# =============================================================================
UI_HOST=0.0.0.0
UI_PORT=8000
# Comma-separated list of allowed CORS origins for the FastAPI server
UI_CORS_ORIGINS=http://localhost:3000

# =============================================================================
# Core database credentials (used by infra/docker-compose.yml)
# =============================================================================
NEO4J_PASSWORD=changeme-neo4j
MONGODB_PASSWORD=changeme-mongodb
MINIO_PASSWORD=changeme-minio
OPENSEARCH_PASSWORD=changeme-opensearch
POSTGRES_PASSWORD=changeme-postgres
NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
# Pydantic Settings does NOT expand ${VAR}; hardcode the password here.
# When changing MONGODB_PASSWORD, update this URI in lockstep.
MONGODB_URI=mongodb://monitor:changeme-mongodb@localhost:27017/monitor?authSource=admin
QDRANT_URI=http://localhost:6333
# Storage backend selector:
#   minio  = local MinIO or any S3-compatible service via MINIO_* / STORAGE_ENDPOINT
#   s3     = AWS S3 (leave STORAGE_ENDPOINT blank for AWS)
#   folder = store uploaded files directly on disk
STORAGE_BACKEND=minio
STORAGE_ENDPOINT=
STORAGE_FALLBACK_TO_LOCAL=true
LOCAL_STORAGE_PATH=.local_storage

MINIO_ENDPOINT=http://localhost:9000
MINIO_ACCESS_KEY=monitor
MINIO_SECRET_KEY=${MINIO_PASSWORD}
OPENSEARCH_URL=https://localhost:9200
POSTGRES_HOST=localhost
POSTGRES_PORT=5432
POSTGRES_USER=monitor
POSTGRES_DB=monitor
REDIS_URL=redis://localhost:6379/0
REDIS_ENABLED=true
REDIS_KEY_PREFIX=monitor

# =============================================================================
# MCP server configuration
# =============================================================================
MCP_SERVER_PORT=8080
MCP_AUTH_SECRET=replace-with-random-secret

# =============================================================================
# Performance tuning
# =============================================================================
NEO4J_HEAP_INITIAL=512m
NEO4J_HEAP_MAX=2G
NEO4J_PAGECACHE=1G
MONGODB_CACHE_SIZE=1024
QDRANT_M=16
QDRANT_EF_CONSTRUCT=100

# =============================================================================
# LLM + embedding configuration
# =============================================================================
# LLM_PROVIDER is intentionally NOT set here — the runtime resolves provider
# from the `llm_providers` PostgreSQL table (seeded by `monitor init` or the
# scripts/seed_*_provider.py scripts). See docs/architecture/LLM_REGISTRY.md.
LLM_MODEL=claude-sonnet-4-20250514
LLM_TEMPERATURE=0.7
LLM_MAX_TOKENS=4096
ANTHROPIC_API_KEY=
OPENAI_API_KEY=

# GitHub Models (PAT with models:read scope — any of these three is accepted)
GITHUB_MODELS_TOKEN=
# GITHUB_TOKEN=
# GH_TOKEN=
GITHUB_MODELS_BASE_URL=https://models.github.ai/inference
GITHUB_MODELS_MODEL=gpt-4.1-mini

# Google AI Studio (Gemini)
GOOGLE_API_KEY=
GOOGLE_MODEL=gemini-2.5-flash

# Z.AI (ZhipuAI)
Z_AI_API_KEY=your_z_ai_api_key_here
Z_AI_MODEL=glm-5.1
Z_AI_BASE_URL=https://api.z.ai/api/coding/paas/v4

# MiniMax AI
MINIMAX_API_KEY=
MINIMAX_BASE_URL=https://api.minimax.io/anthropic

# Local Ollama
OLLAMA_ENDPOINT=http://localhost:11434
OLLAMA_MODEL=qwen2.5:latest

EMBEDDING_PROVIDER=ollama
EMBEDDING_MODEL=nomic-embed-text:latest
EMBEDDING_DIMENSION=768
# Vision model for image ingestion (must support image_url content)
# Options: gpt-4o-mini, gpt-4o, claude-3-5-sonnet-20241022, gemini/gemini-1.5-flash
VISION_MODEL=gpt-4o-mini

# =============================================================================
# Ingestion queue tuning
# =============================================================================
# Keep at 1 unless you have measured that your machine and LLM provider can
# safely handle parallel ingestion.
MONITOR_INGEST_MAX_WORKERS=1
MONITOR_INGEST_TIMEOUT=2700
MONITOR_MAX_INGEST_FILE_BYTES=209715200
REDIS_CACHE_TTL_SECONDS=30
REDIS_SOLO_PLAY_TTL_SECONDS=15
REDIS_SOCKET_TIMEOUT=0.15
REDIS_CONNECT_TIMEOUT=0.15

# =============================================================================
# Observability + flags
# =============================================================================
LOG_LEVEL=INFO
ENABLE_METRICS=true
METRICS_ENDPOINT=http://prometheus:9090
ENVIRONMENT=development
DEBUG=false
ENABLE_CORS=true
CORS_ORIGINS=http://localhost:3000,http://localhost:5173

# =============================================================================
# NLP / GLiNER (Named Entity Recognition)
# =============================================================================
NLP_ENABLED=true
# Default to spaCy (bundled, fast). GLiNER is opt-in via --profile gliner.
NLP_BACKEND=spacy
GLINER_PORT=8082
GLINER_MODEL=knowledgator/gliner-base-v0.1
GLINER_MAX_LENGTH=384
GLINER_BATCH_SIZE=8
GLINER_CPUS=2.0
GLINER_MEMORY=4G
ENTITY_EXTRACTION_ENABLED=true
ENTITY_TYPES=person,location,organization,product,event,work_of_art,law,language,date,time,percent,money,quantity,ordinal,cardinal

# ── Observability (T-044) ────────────────────────────────────
# Agent calls are traced with logfire (OpenTelemetry-compatible).
# Local-only by default. To export spans:
#   LOGFIRE_TOKEN=...                # send to logfire cloud, or
#   OTEL_EXPORTER_OTLP_ENDPOINT=...  # any OTLP collector (e.g. http://localhost:4318)
LOGFIRE_SEND_TO_LOGFIRE=false
#LOGFIRE_TOKEN=
#OTEL_EXPORTER_OTLP_ENDPOINT=

# --- Docker networking ---
# When the backend runs in Docker and a provider row points at a service on
# the HOST (ollama http://localhost:11434, LM Studio, ...), rewrite localhost
# to this hostname. docker-compose sets this automatically; leave empty for
# bare-metal runs.
MONITOR_LOCALHOST_REWRITE=