.env.example

# =============================================================================
# CanAID — environment template. Copy to .env and fill values.
# =============================================================================
#
# AWS credentials are read by boto3's default chain (do NOT manage them via
# CANAID_*). Common options:
#   1. AWS_PROFILE  — refers to a profile in ~/.aws/credentials
#   2. AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY (+ AWS_SESSION_TOKEN)
#   3. IAM role (in ECS / Lambda / EC2 — nothing to set)
#
# AWS_REGION is used by the Bedrock client. us-east-1 has the broadest model
# availability today; switch only if you have a regional reason.
# =============================================================================

# Override AWS_PROFILE if your local profile in ~/.aws/credentials is
# named differently. On Streamlit Cloud / ECS / etc., set AWS_ACCESS_KEY_ID
# instead and the profile path is bypassed.
AWS_PROFILE=default
AWS_REGION=us-east-1

# -----------------------------------------------------------------------------
# Per-agent Bedrock model IDs.
# Defaults use cross-region inference profile IDs (us. prefix). You can drop
# the prefix to pin a single region, or swap to claude-sonnet-4-6 / opus-4-7
# in your account if those inference profiles are enabled.
# -----------------------------------------------------------------------------
CANAID_SUPERVISOR_MODEL=us.anthropic.claude-sonnet-4-5-20250929-v1:0
CANAID_INTENT_MODEL=us.anthropic.claude-haiku-4-5-20251001-v1:0
CANAID_QUALIFIER_MODEL=us.anthropic.claude-sonnet-4-5-20250929-v1:0
CANAID_RAG_MODEL=us.anthropic.claude-sonnet-4-5-20250929-v1:0
# Tool-use agent uses a non-Anthropic model on purpose (Meta) to demonstrate
# multi-vendor portability. Override to mistral.* or amazon.nova-pro-* if
# Llama isn't enabled in your account.
CANAID_TOOL_MODEL=meta.llama3-3-70b-instruct-v1:0
CANAID_SUMMARIZER_MODEL=us.amazon.nova-lite-v1:0
# Titan v2 supports 256/512/1024 dim outputs. We pin 1024 in code; change
# the model id only if you've onboarded a different embedding model.
CANAID_EMBED_MODEL=amazon.titan-embed-text-v2:0

# -----------------------------------------------------------------------------
# Bedrock Guardrails (Phase 5). Create one via `python scripts/setup_guardrail.py`
# and paste the printed ID here. Leave empty to run without policy enforcement.
# -----------------------------------------------------------------------------
CANAID_GUARDRAIL_ID=
CANAID_GUARDRAIL_VERSION=DRAFT

# -----------------------------------------------------------------------------
# Caching + memory (Phase 6)
# -----------------------------------------------------------------------------
CANAID_CACHE_ENABLED=true
CANAID_CACHE_TTL_SECONDS=3600
# Set true to persist conversation state in Postgres (durable across restarts).
# Defaults to in-memory for dev. The Phase 9 CDK stack flips this on in prod.
CANAID_USE_POSTGRES_CHECKPOINTER=false

# -----------------------------------------------------------------------------
# API + UI
# -----------------------------------------------------------------------------
CANAID_API_HOST=0.0.0.0
CANAID_API_PORT=8000
CANAID_API_URL=http://localhost:8000

# -----------------------------------------------------------------------------
# Stores (Phase 3 / Phase 6) — defaults match docker-compose.yml
# -----------------------------------------------------------------------------
CANAID_PG_DSN=postgresql://canaid:canaid@localhost:5432/canaid
CANAID_REDIS_URL=redis://localhost:6379/0

# -----------------------------------------------------------------------------
# Observability (Phase 8) — get keys from https://cloud.langfuse.com
# -----------------------------------------------------------------------------
LANGFUSE_PUBLIC_KEY=
LANGFUSE_SECRET_KEY=
LANGFUSE_HOST=https://us.cloud.langfuse.com

CANAID_LOG_LEVEL=INFO