Configuration

At the end you will know every environment variable this project reads, grouped by what it controls, and which ones are required versus optional.

All configuration lives in one .env file at the repo root (copy from .env.sample; .env itself is gitignored — never commit it). Every variable is loaded once, in agents/config.py.

Never commit secrets

.env holds real AWS keys, Databricks tokens, and observability API keys. It is gitignored on purpose. Never paste its contents into a commit, a chat log, or a public issue.

Model (Amazon Bedrock)

Variable	Required	Meaning
AWS_REGION	yes	Region your Bedrock model access is in.
AWS_PROFILE	one of these two	A named AWS CLI profile — boto3's normal credential chain.
AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY	one of these two	Raw keys, if you prefer them over a profile. boto3 picks these up automatically once set; AWS_PROFILE is ignored if both keys are present.
BEDROCK_MODEL_ID	yes	The exact model or inference-profile ID to call. Never hardcoded anywhere in the code — some models need the inference-profile ID instead of the bare model ID (see Operations).

Data target

Variable	Required	Meaning
AGENT_TARGET	yes	duckdb (dev, zero infra) or databricks (prod). Every tool's SQL is portable across both.
DUCKDB_PATH	if AGENT_TARGET=duckdb	Path to the Claimwise repo's built rcm.duckdb.
CLAIMWISE_DBT_PATH	Data Steward only	The Claimwise dbt-pipeline repo root — run_dq_checks runs dbt test there; get_lineage reads its target/manifest.json.
DATABRICKS_HOST / DATABRICKS_HTTP_PATH / DATABRICKS_TOKEN	if AGENT_TARGET=databricks	Read-only warehouse credentials. Agents have no code path that can write — see Architecture.
DATABRICKS_CATALOG	if AGENT_TARGET=databricks	Defaults to workspace.

Observability (all optional)

Every one of these is off by default and independently switchable — see Operations for what each one shows you.

Variable	Meaning
LANGSMITH_TRACING	true to export traces to LangSmith.
LANGSMITH_API_KEY / LANGSMITH_PROJECT	LangSmith project credentials.
ARIZE_SPACE_ID / ARIZE_API_KEY / ARIZE_PROJECT_NAME	Arize AX credentials. Set either one and traces start exporting.
OTEL_EXPORTER_OTLP_ENDPOINT / OTEL_EXPORTER_OTLP_HEADERS	Any generic OTLP/HTTP collector — accepts either an org/stream base URL or a full .../v1/traces URL, both get normalized.

AgentCore Memory (optional)

Variable	Meaning
MEMORY_ID	Blank by default — Memory stays fully disabled, identical to not having this section at all. Points at a live AgentCore Memory resource; see Deployment & Integration.
MEMORY_ACTOR_ID	A placeholder actor identity (default demo-user) — a real deployment would derive this from the caller's authenticated identity, not a fixed string.

What next

• See these variables in action → Examples
• What each agent does with these settings → Architecture