Configuration Overview

SQLatte is configured entirely through one file: config/config.yaml, loaded relative to the project root at startup — there's no --config flag or .env requirement, though every value supports ${ENV_VAR:default} interpolation if you want secrets sourced from the environment instead of the file.

Only two sections are required to boot: llm and database. Everything else is optional and off by default.

Sections at a Glance

Section	Default	Purpose
app	—	Host, port, app name
llm	required	LLM provider + model (Anthropic / Gemini / Vertex AI), optional per-task model_routing
database	required	Trino / PostgreSQL / MySQL / BigQuery connection
mcp.sse	off	Network MCP transport — see MCP Overview
plugins.auth	off	Multi-tenant per-user DB connections, catalog/schema restrictions, auto-session for widgets
admin	—	Admin panel username/password, session TTL
config_db	off	Persist runtime config + audit logs to PostgreSQL instead of memory
analytics	off	Query history, performance metrics — details
scheduler	off	Cron-based recurring queries — see Scheduled Queries
email	off	SMTP delivery for scheduled reports — details
export	—	CSV/Excel/HTML export formats and limits
insights	on	AI insights engine mode (llm_only / statistical_only / hybrid)
ops_agent	off	BigQuery Ops Console — see Ops Console
alarms / jira	off	Cost alarm scheduling + Jira ticket creation
rate_limiting	off	Per-endpoint request throttling — see Security Overview
ui.sections	—	Toggle visibility of Assistant / Demo / Analytics / Schedules / Dashboards / Ops / Admin tabs
prompts	—	The four editable AI prompts — see Runtime Prompts

See the Full Config Reference for every key, or jump to a specific area: Database Providers, LLM Providers, Analytics, Dashboard, Email.

Hot Reload

Most settings changed via the admin panel (/admin) apply immediately — no restart. Config edited directly in config.yaml requires a restart unless config_db is enabled, in which case runtime overrides persist and take precedence over the file on next load.