Vercel
Use it for: Next.js and React apps where you want preview deploys on every push.
Tradeoff: Free tier is for personal use; team and commercial use moves you to a paid plan.
Guide ยท Hosting
We earn commissions when you shop through the links below.
Most AI apps are a front end, a thin backend, a database and some model calls. Each layer has a free or low-cost option if you pick deliberately. Here is when to use what, and what you give up. For the full walkthrough, read the cheap AI infrastructure stack or estimate your model bill with the LLM cost calculator.
Your UI, docs or landing page. Free tiers usually cover a side project until it has real traffic.
Use it for: Next.js and React apps where you want preview deploys on every push.
Tradeoff: Free tier is for personal use; team and commercial use moves you to a paid plan.
Use it for: Static or framework sites that should sit on a global edge network.
Tradeoff: Generous free tier; some Node-specific libraries need adapting to the Workers runtime.
Skills for this: Deploy frameworks on Cloudflare ยท Let an agent test Vercel deploys
A thin backend that answers requests and calls a model. Pay per use, scales to zero.
Use it for: Short request/response handlers, webhooks and model-call proxies.
Tradeoff: CPU time limits per request; long jobs and websockets need other pieces.
Use it for: API routes that live next to a Vercel front end.
Tradeoff: Execution time limits; costs grow with invocations and duration.
Skills for this: Build and debug Workers
Agents with background loops, queues, websockets or a warm model server. Flat monthly pricing that does not jump with traffic.
Use it for: A simple Linux box for Docker Compose, a worker queue and a reverse proxy.
Tradeoff: You patch and back it up yourself; one box is one point of failure.
Use it for: Lots of raw compute if you are happy with a bare-bones experience.
Tradeoff: Fewer managed extras and data center regions than the bigger clouds.
Skills for this: Provider-neutral DevOps playbook
Postgres for app data and vectors. Managed means backups and upgrades are someone else's job.
Use it for: PostgreSQL, MySQL or Redis next to your Droplets, with backups built in.
Tradeoff: Flat monthly cost even when idle; no free tier.
Use it for: Postgres plus auth, storage and realtime in one dashboard.
Tradeoff: Free projects pause when inactive; you buy into its platform conventions.
Use it for: Serverless Postgres that scales to zero, with database branching for previews.
Tradeoff: Cold starts after idle; pay per use can be harder to predict.
Push a repo, get a running service. Less control than a server, much less to maintain.
Use it for: Web services and workers deployed from Git, alongside DO databases.
Tradeoff: Less control over the runtime than a Droplet.
Use it for: Web services, cron jobs and background workers from one repo.
Tradeoff: Free instances spin down when idle, so first requests are slow.
Use it for: Quick prototypes with a database attached in a couple of clicks.
Tradeoff: Usage-based billing; watch always-on services.
Use it for: Containers close to users in many regions.
Tradeoff: More moving parts (machines, volumes, regions) to understand.
Skip the per-token bill for dev, testing and private data by running open models yourself.
Use it for: Running open-weight models on your own machine or a server.
Tradeoff: Needs real RAM or a GPU; smaller models trail frontier APIs on hard tasks.
Skills for this: Ollama stack setup ยท LLM response caching ยท LLM cost optimization