Configuration reference

Settings are resolved in this priority order (earlier sources win):

1. Explicit initialization arguments
2. Environment variables with RIVERBANK_ prefix (nested with __)
3. TOML config file (~/.riverbank/config.toml or RIVERBANK_CONFIG_FILE)

Environment variables

Database

Variable	Type	Default	Description
RIVERBANK_DB__DSN	string	postgresql+psycopg://riverbank:riverbank@localhost:5432/riverbank	PostgreSQL connection string

LLM provider

Variable	Type	Default	Description
RIVERBANK_LLM__PROVIDER	string	ollama	Provider: ollama, openai, anthropic, vllm, azure-openai
RIVERBANK_LLM__API_BASE	string	http://localhost:11434/v1	Provider API base URL
RIVERBANK_LLM__API_KEY	string	ollama	API key
RIVERBANK_LLM__MODEL	string	llama3.2	Model identifier
RIVERBANK_LLM__EMBED_MODEL	string	nomic-embed-text	Embedding model
RIVERBANK_LLM__MAX_TOKENS	int	4096	Maximum tokens per LLM call

Langfuse (LLM observability)

Variable	Type	Default	Description
RIVERBANK_LANGFUSE__ENABLED	bool	false	Enable Langfuse tracing
RIVERBANK_LANGFUSE__PUBLIC_KEY	string	""	Langfuse public key
RIVERBANK_LANGFUSE__SECRET_KEY	string	""	Langfuse secret key
RIVERBANK_LANGFUSE__HOST	string	http://localhost:3000	Langfuse server URL

Tenant

Variable	Type	Default	Description
RIVERBANK_TENANT_ID	string	""	Current tenant context (sets app.current_tenant)

OpenTelemetry

Variable	Type	Default	Description
OTEL_EXPORTER_OTLP_ENDPOINT	string	""	OTLP endpoint for spans (Jaeger, Tempo, Honeycomb)
OTEL_SERVICE_NAME	string	riverbank	Service name in traces

Config file path

Variable	Type	Default	Description
RIVERBANK_CONFIG_FILE	string	~/.riverbank/config.toml	Path to TOML config file

TOML config file

[db]
dsn = "postgresql+psycopg://riverbank:riverbank@localhost:5432/riverbank"

[llm]
provider = "ollama"
api_base = "http://localhost:11434/v1"
api_key = "ollama"
model = "llama3.2"
embed_model = "nomic-embed-text"
max_tokens = 4096

[langfuse]
enabled = false
public_key = ""
secret_key = ""
host = "http://localhost:3000"

Settings resolution

The Settings class uses Pydantic Settings with these sources (highest priority first):

1. Init kwargs — programmatic overrides
2. Environment variables — RIVERBANK_ prefix, __ as nesting separator
3. TOML file — ~/.riverbank/config.toml (or RIVERBANK_CONFIG_FILE)

Warning

The RIVERBANK_DB__DSN variable contains credentials. Never log it or include it in error messages. Use Kubernetes secrets or Vault in production.

Type coercion

Pydantic handles type coercion automatically:

• "true" / "false" → bool
• "4096" → int
• Nested objects use __ separator: RIVERBANK_LLM__MODEL=gpt-4o