Skip to main content

​
Infrastructure

Moveat uses a pragmatic infrastructure model: managed hosting where it reduces work, and a DigitalOcean droplet where low-latency backend dependencies are useful. The goal is not to introduce heavy orchestration too early. The goal is to keep the system understandable, observable and cheap to operate while the product loops are being validated.

​
Hosting map

SurfaceHostingReason
Web clientVercelFast frontend deploys, previews and CDN delivery.
DocumentationMintlifyDocumentation UI, search, navigation and custom domain.
Platform APIDigitalOcean dropletBackend close to PostgreSQL and Redis.
Agent API/webhookDigitalOcean dropletWhatsApp webhook and backend orchestration.
PostgreSQLDigitalOcean dropletLow latency and direct ownership for current stage.
RedisDigitalOcean dropletLow-latency session and conversation state.
Grafana/Loki/AlloyDigitalOcean dropletLocal operational observability.
KomodoDigitalOcean dropletContainer management UI.

​
Domain map

DomainOwnerService
api.mov-eat.appDigitalOcean/CaddyPlatform.
agent.mov-eat.appDigitalOcean/CaddyAgent.
<grafana-domain>DigitalOcean/CaddyGrafana.
<ops-console-domain>DigitalOcean/CaddyKomodo.
docs.mov-eat.appMintlifyDocumentation portal.
App domain/subdomainVercelFrontend application.

​
DigitalOcean stack

The droplet runs the backend stack with Docker Compose. The core production services are:

Caddy

Terminates HTTPS and routes public domains to internal Docker services.

Platform

Business backend and source of truth for product APIs.

Agent

Conversational service that receives channel events and calls Platform.

PostgreSQL

Durable product database for users, meals, workouts and progress.

Redis

Short-lived sessions, cache and conversation state.

Grafana + Loki + Alloy

Logs, dashboards, correlation search and operational analysis.

​
Vercel role

Vercel should remain responsible for frontend delivery. The backend should not serve the web app directly unless there is a clear operational reason. This gives the frontend team:
  • Preview deployments.
  • Fast rollback.
  • CDN distribution.
  • Separation from backend container lifecycle.

​
Komodo role

Komodo is an operational tool, not the source of truth for the stack. The source of truth remains the infra repository and its Docker Compose files. Use Komodo to inspect containers, restart services, check resource usage and access operational controls. Do not make permanent infrastructure changes only through the UI unless those changes are also reflected in the repository.

​
Grafana role

Grafana is the first place to inspect production behavior:
  • Platform errors.
  • Agent errors.
  • Request correlation IDs.
  • Container logs.
  • Database and service-level operational signals.
Grafana should help answer what happened. It should not become a replacement for structured application logs, tests or runbooks.

​
Mintlify role

Mintlify hosts the documentation and reads the docs repository. The repository is the source of truth; Mintlify is the renderer and hosting layer. This matters operationally: if the live docs do not match the repository, check the Mintlify Git connection, branch, deployment activity and root directory.

​
Current trade-off

Running PostgreSQL and Redis inside the droplet is reasonable at this stage because it reduces latency and cost. The trade-off is that Moveat owns backups, monitoring and failure recovery. As traffic grows, moving PostgreSQL to a managed database may become attractive. That decision should be based on operational load, backup requirements and uptime targets, not on architecture preference alone.