Planned region → the United States
Not available yet. US East (New Jersey) is a region we are planning, not one you can deploy to. Hostim runs one live region today, in Falkenstein, Germany.
FastAPI on Hostim runs as a permanent container: the model loads once, stays in memory, and streams tokens for as long as the response takes. Weights live on a persistent volume so they are not redownloaded on every deploy. This runs in Germany today. A US East region in New Jersey is planned — join the waitlist if you would put an inference API there.
# docker-compose.yml
services:
api:
image: my-fastapi-app
command: uvicorn main:app --host 0.0.0.0 --port 8000
db:
image: postgres:16Serverless is the wrong shape for inference and most people find that out after they ship. A cold start reloads the model, the function timeout truncates long generations, and per-request billing turns a chatty client into a budget problem. A long-lived container fixes all three by not being clever: the process starts, loads the weights from a mounted volume, and stays up. Streaming works because the connection is held by a real server rather than a gateway with a timeout. Cost is the reserved box, not the request count. The honest limit is that Hostim is CPU-only right now — good for embeddings, classical ML, and smaller models, not for serving a 70B parameter model, and GPU support is roadmap rather than shipped. The other honest limit is geography: this is all in Germany, which adds about 100 ms for a US client on every call. That is what the planned New Jersey region would fix.
Expected latency. The expected site is Secaucus, New Jersey; the exact facility is not fixed and any northern New Jersey site gives similar figures. The figures below are typical round-trip times for that metro — they are estimates for the location, not measurements from Hostim hardware, because there is none there yet.
| City | Expected RTT to US East |
|---|---|
| New York | ~2 ms |
| Washington DC | ~6 ms |
| Boston | ~8 ms |
| Chicago | ~20 ms |
| Atlanta | ~20 ms |
| Dallas | ~35 ms |
| Denver | ~45 ms |
| Seattle | ~65 ms |
| San Francisco | ~70 ms |
Privacy law. CCPA and CPRA in California, plus state acts in Virginia, Colorado, Connecticut, Utah and a growing list of others. There is no single federal privacy law, so the obligations follow your users, not your servers.
Who enforces it. the California Privacy Protection Agency (CPPA) and state attorneys general — there is no federal data protection authority.
Data residency — read this before you plan a migration. A US East region means your app, your database and your volumes run on hardware in New Jersey. It does not make Hostim a US company: HOSTIM.DEV UG is German, and that is a real consideration if you have a customer contract that restricts foreign processors. Ask us before you assume either way.
Market context. US teams rarely pick a host for data residency. They pick it for price, for not being on-call for a Kubernetes cluster, and for latency to users on the East Coast. The pitch that works in Frankfurt does not work in Brooklyn — what works is a flat bill and a container that stays running.
What you are probably comparing us against: Heroku, Render, Railway, Fly.io, DigitalOcean App Platform, AWS App Runner.
FastAPI hosting means running a Uvicorn or Gunicorn-Uvicorn worker process and exposing it over HTTPS. The framework is async by default, so the host has to support long-lived connections — websockets, server-sent events, streaming responses.
Hostim runs your FastAPI Docker image as a normal container. Long-lived connections work. Managed PostgreSQL is attached at runtime. If you serve an ML model, mount a persistent volume for the weights so they do not redownload on every deploy.
Cold starts from serverless platforms are not a fit for ML inference workloads. Hostim runs a permanent container, so model weights stay in memory across requests.
DATABASE_URL, OPENAI_API_KEY, MODEL_PATH, LOG_LEVEL
No. It is in planning. Everything on this page works in our German region today; the US East site in New Jersey has not been built, and the waitlist is how we judge whether to build it.
Not today, in any region. CPU inference with generous RAM covers embeddings, classical ML and small transformer models. If you need GPU, say so on the waitlist form — it is useful signal even though we cannot promise it.
Mount a persistent volume at /models and load from there at startup. The volume is reattached to each new container, so a deploy does not trigger a redownload. This is region-independent behaviour.
Yes. Server-sent events and token-by-token streaming work because the container holds the connection for the full duration of the request. There is no gateway timeout cutting off a long generation.
For a single call it usually does not. It matters when a client makes several sequential calls per user action, or when you are streaming and the first token is already the slowest part of the experience. A US region removes that fixed cost for US clients.
We are deciding how much capacity to build. Add your name and what you would run — that is the whole input. You can also deploy to our live German region today if that works for you.
Platform figures from the published price list, docs and our own benchmark.
€2.50/month
Hostim entry app plan
Plan sa-1-1: 1 vCPU, 1 GB RAM. Flat price, billed hourly, no usage meter.
€0
Managed database free tier
PostgreSQL 256 MB, MySQL 256 MB, Redis 128 MB, volume 1 GB. No cap on how many free instances you create.
Included on every plan
Database high availability
PostgreSQL and MySQL run as a primary plus hot standby with automatic failover, shared plans included.
2,708 TPS
Managed PostgreSQL write throughput
pgbench, 4 clients, 300 s, plan drp-50 (2 vCPU / 4 GB). AWS RDS db.t4g.medium scored 1,080 TPS on the same test.
€0/GB
Traffic charges
No ingress fees and no per-GB egress line on app plans.
1 today (Falkenstein, Germany)
Regions
Bare metal, EU only, with more regions planned. No AWS, GCP or Azure underneath.
“Managed PostgreSQL runs as a replicated cluster with automatic failover on every plan, shared and dedicated alike.”
— Hostim.dev docs, managed PostgreSQL