NoctHost

One account for Vultr, DigitalOcean, Hetzner and Linode servers. Hourly billing, API, pay with crypto or card.

[NoctHost @nocthost]

Where to run a crypto trading bot? We timed exchange APIs from 21 cities today.

Binance, Bybit, OKX, Hyperliquid: Tokyo, under 25 ms
Coinbase: Chicago, ~30 ms
Kraken: London, ~15 ms

From Europe, Binance is ~230 ms away.

Table and CSV: https://t.co/4Y1Z3uTrEI

[NoctHost @nocthost]

@AM09_21 Phone + VPS works well if the session outlives the connection: run agents in tmux (or mosh on flaky mobile networks) so a dropped signal doesn't kill a task. And keep the UI off the public internet: bind to 127.0.0.1 and reach it over a VPN or tunnel.

[NoctHost @nocthost]

Vultr, DigitalOcean, Hetzner or Linode? For one ordinary server the choice barely matters, so pick on city. What actually differs: the map (we sell 30 Vultr cities, 18 Linode, 9 DigitalOcean, 6 Hetzner), browser console, on-demand backups, Windows. https://t.co/Abs5ySvUdr

[NoctHost @nocthost]

Running a MetaTrader EA on a VPS? Pick the city by your broker's trade server and read the ping from the VPS itself.

One EA: MQL5's own hosting ($10 to $15/mo) is enough. Several terminals, other software or DLLs: a full Windows VPS.

https://t.co/fGZs6nqAnf

[NoctHost @nocthost]

fail2ban is the last layer of VPS security, not the first. In order:

1. sudo ss -tlnp, bind anything private to 127.0.0.1
2. Docker -p skips ufw: publish 127.0.0.1:5432:5432
3. Non-root containers, cap_drop: [ALL]
4. unattended-upgrades

https://t.co/S11MneFFAB

[NoctHost @nocthost]

@MiaAI_lab That's the big win of one large card: with continuous batching (vLLM or SGLang) the weights load once and the batcher keeps the GPU busy, so per-session speed barely drops until the KV cache fills up.

[NoctHost @nocthost]

@apekshik Congrats on the launch. One thing to watch after a spike like this: per-GB egress. Serve the game assets from a CDN or a host with bundled traffic, and let the server only carry game state. That bill is what usually surprises solo devs a week later.

[NoctHost @nocthost]

@dhaiwat Worth checking what the bill actually is before killing it. For browser multiplayer it's usually relay or egress traffic. Per-GB hosts scale with every player; peer-to-peer first, relay only as fallback, on a box with a few TB included keeps it flat.

[NoctHost @nocthost]

Netdata on a 1 GB VPS: defaults keep lots of history in RAM. Three lines fix most of it:

update every = 2
dbengine tier 0 retention size = 256MiB
collectors off for what you don't run

Then bind it to 127.0.0.1. Port 19999 open to all maps your box.

https://t.co/euNlkPoJwy

[NoctHost @nocthost]

@texhnogatsby Right, a box doing real scheduled work is safe. Oracle reclaims only when CPU, network and memory (on A1) all stay under 20% for 7 days. The one at risk is the box that just waits for requests.

[NoctHost @nocthost]

@themishra4402 Postgres wins because every template and hosted backend ships it by default. Most of these apps would run fine on SQLite on one small server.

[NoctHost @nocthost]

@connorkapoor Best part is closing SSH to the public internet entirely. A fresh VPS gets login attempts within minutes of booting, over the tailnet there's nothing for them to hit.

[NoctHost @nocthost]

@andreamichi @depthfirstlabs Persistent per developer makes sense. For the agents themselves a fresh environment per task, destroyed after, is worth adding: long-lived agent boxes drift and quietly collect secrets.

[NoctHost @nocthost]

@MaranDefi Worth knowing they pause after 30 minutes idle, so it's not 24/7 for anything that waits on incoming traffic. Great for agents and builds, and keep anything you can't lose backed up off the box.

[NoctHost @nocthost]

@elliehuxtable Moving Postgres off first was the right order. Self-hosted compute is easy to rebuild from a manifest, the database is where the 3am pages come from.

[NoctHost @nocthost]

@thorstenball Postgres in Docker on a remote box works fine for agents. Two things bite: give each task a throwaway database it can reset, and bind the port to 127.0.0.1, because Docker publishes ports past ufw.

[NoctHost @nocthost]

@rasta_syntaX Flat monthly box is underrated. The surprise bills on usage-based clouds almost never come from CPU, it's egress, logs and forgotten snapshots.

[NoctHost @nocthost]

@NixFred Forgejo or Gitea on a small VPS is easy and runs fine on 1 GB RAM. Keep GitHub as a push mirror so a dead box never costs you the repo, and back up the data dir somewhere else.

[NoctHost @nocthost]

Jellyfin on a VPS: plan for direct play, not transcoding.

Direct play just streams bytes, so a small server handles several viewers. CPU transcoding is about one 1080p stream per 2 vCPU. 4K mostly won't work.

Budget 4-8 Mbit/s of bandwidth per viewer.

https://t.co/P3F5t3OVmp