Environment configuration

When to use: Adding or reading env vars, updating .env.example, or validating config at startup with parseEnv / parseEnvOptional.

LAT_ prefix convention

All application environment variables must be prefixed with LAT_ so they do not collide with third-party services, Docker, or common names.

Use LAT_ for:

• Database URLs and pool settings (LAT_DATABASE_URL, LAT_PG_POOL_MAX, …)
• Service endpoints the app reads (LAT_CLICKHOUSE_URL, LAT_REDIS_HOST, …)
• App ports (LAT_API_PORT, LAT_WEB_PORT, LAT_INGEST_PORT)
• Auth, email, OAuth, billing, CORS (LAT_BETTER_AUTH_SECRET, LAT_MAILPIT_HOST, …)
• Any new variable consumed by Latitude application code

Do not use LAT_ for:

• NODE_ENV
• Docker-only init variables (POSTGRES_USER, CLICKHOUSE_USER, …)
• Config read only by container images (Weaviate, Redis in compose, etc.)
• Browser-exposed Vite vars: use VITE_LAT_* (Vite requires the VITE_ prefix)

Reference: .env.example lists Docker “Services” vs “Latitude Application” (LAT_*) variables.

.env.example maintenance

Every new variable must appear in .env.example:

• Required: uncommented with a sensible local default (e.g. LAT_API_PORT=3001)
• Optional: commented with a placeholder (e.g. # LAT_STRIPE_SECRET_KEY=sk_test_xxx)

Keep the self-host surfaces in sync

Self-hosters configure Latitude through these env vars, so adding, renaming, or removing one means updating more than .env.example. Whenever you touch a LAT_* var, also reflect it in:

1. The configuration reference docs/deployment/configuration.mdx — so operators know what the var does — unless it is specific to Latitude's own cloud deployment, which stays out of the self-host docs: payments/Stripe, marketing/lifecycle email (Loops), support chat (Intercom), and internal analytics/observability vendors (PostHog, Datadog/OTEL export, GTM, Framer, ipinfo). General app/infra/auth/AI/email-transport vars all belong in the reference.
2. The Helm chart charts/latitude/ — decide where the var belongs:
   • non-secret → templates/configmap.yaml (and a values.yaml knob if it's a first-class setting);
   • secret → templates/secret.yaml and the chart README's existingSecret key list;
   • optional/rare → no template change needed; the documented config.extraEnv / secrets.extra pass-through already covers it.

docker-stack.yml needs no per-var change (it consumes the whole .env.production), but a new required var must land in .env.example's production guidance. The Railway template is dashboard-authored — no repo change. See dev-docs/self-hosting.md for the tier/contract overview.

Parsing in code

Always use parseEnv or parseEnvOptional from @platform/env — never process.env.FOO ad hoc or unprefixed names for app config.

// ❌ Bad - unprefixed or direct access
const port = Number(process.env.PORT)

// ✅ Good - pass the variable name string (parseEnv reads process.env internally)
import { parseEnv, parseEnvOptional } from "@platform/env"
import { Effect } from "effect"

const port = Effect.runSync(parseEnv("LAT_API_PORT", "number", 3001))
const dbUrl = Effect.runSync(parseEnv("LAT_DATABASE_URL", "string"))

For where configuration is wired in apps (clients, routes), see architecture-boundaries.