> ## Documentation Index
> Fetch the complete documentation index at: https://docs.fapost.in/llms.txt
> Use this file to discover all available pages before exploring further.

# Configuration

> The decisions behind the environment file, and what must change before production.

Configuration lives in `.env`. The full list of variables with their defaults is
in [Environment variables](/reference/environment-variables); this page is about
the choices behind them.

## Two database connections

FaPost uses one PostgreSQL server with two logical scopes.

**Landlord** holds platform-level records — the tenant list, the webhook routing
registry, queues and cache tables. **Tenant** holds everything belonging to an
organisation, in its own schema.

You configure one connection. The tenant schema is selected at runtime from it,
which is why there is no second set of `DB_*` variables and why the database user
needs the `CREATE` privilege: schemas are created as tenants are.

## Redis is not optional

Redis carries four things, and losing any of them has a distinct symptom:

| Carries                  | If it goes                                                    |
| ------------------------ | ------------------------------------------------------------- |
| Queues                   | Nothing is processed — messages arrive and are never answered |
| Cache                    | Slower, but working                                           |
| Locks                    | Concurrent messages from one contact can race                 |
| Webhook routing registry | Inbound messages resolve nowhere                              |

The registry is the one that surprises people. It exists so that resolving an
incoming webhook never touches the landlord database — the hot path stays a Redis
lookup. That makes Redis part of the delivery path, not just an accelerator.

<Warning>
  `REDIS_PREFIX` must match between the application and the
  [webhook gateway](/self-hosting/gateway) if you run one. A mismatch is silent:
  the gateway reads an empty registry, decides it knows nothing, and forwards
  every delivery to PHP. Everything works, slower, for no visible reason.
</Warning>

## Webhooks and the public address

Two variables decide where providers deliver.

`WEBHOOK_BASE_URL` is the address the application builds callback URLs from. It
must be publicly reachable over HTTPS — providers will not deliver to anything
else.

`WEBHOOK_INGRESS_DRIVER` chooses what handles them: `laravel` for the application
itself, `gateway` for the Go service.

<Note>
  The driver applies to **newly registered channels only**. Existing channels
  keep the URL their provider already stored, and the Laravel route stays live
  regardless — so switching is not a cutover. Move the rest with
  `php artisan ops:ingress-migrate --apply`.
</Note>

## What must change for production

The shipped `.env.example` is a development file. Four values are unsafe as they
stand:

|                     |                                                                      |
| ------------------- | -------------------------------------------------------------------- |
| `APP_DEBUG`         | `true` exposes stack traces with configuration in them. Set `false`. |
| `APP_KEY`           | Empty. Generate one, and never reuse it across installations.        |
| `TELESCOPE_ENABLED` | `true` records every request, including their contents.              |
| `HCAPTCHA_*`        | Ships with hCaptcha's public test keys, which always pass.           |

Set a `REDIS_PASSWORD` on anything but a local machine.

## Languages

`APP_LOCALE` and `APP_FALLBACK_LOCALE` are the **admin interface** language.

What an assistant says to a contact is decided elsewhere entirely — tenant
settings, assistant settings, and the flow itself. Changing these variables does
not change a single word a contact receives.

## Applying changes

Configuration is read at boot. After editing `.env`:

```bash theme={"theme":"one-dark-pro"}
php artisan config:clear
```

Then restart Horizon — long-running workers hold the configuration they started
with, and a worker that has not been restarted is still using the old values
however many times you clear the cache.