Free Tier Limits Explained: CPU, RAM, Storage and Bandwidth
What each free-plan limit means in practice — a CPU share, a hard RAM ceiling, NVMe storage and network speed — and how to tell when you're hitting one.
On this page
Every hosting plan is a set of limits, and knowing how each limit behaves makes the difference between a bot that runs happily for years and one that mysteriously slows down or restarts. This article explains the four resources on Kerit Cloud’s free plan — CPU, RAM, storage and network — what happens when you reach each one, and how to stay comfortably inside them.
| Resource | Free plan | What happens at the limit |
|---|---|---|
| CPU | 30% of an AMD EPYC core | The bot slows down |
| RAM | 256 MB | The process is stopped and restarted |
| Storage | 2 GB NVMe | Writes fail |
| Network | 10 Gbps port | Rarely a limit for bots |
CPU: a share of a core
CPU is allocated as a percentage of one processor core. 30% means your bot can use up to three-tenths of a core’s time. It’s a ceiling, not a reservation you waste when idle.
What 30% handles
Most bots spend the vast majority of their time waiting — for Discord events, Telegram updates, database replies. A command that parses input, queries a database and sends a message uses a tiny slice of CPU. A bot doing this for a few thousand commands a day barely registers.
What pushes CPU up
- Image generation — rank cards and welcome images with canvas or Pillow.
- Audio — decoding and encoding in-process. Music bots should use Lavalink instead.
- Startup of large bots — processing thousands of guilds at connect time.
- Tight loops — recalculating leaderboards over every user on every message.
- Blocking code in async bots, which also freezes everything else.
How you’ll notice
At the CPU limit, the process is throttled, not killed. Everything gets slower: commands take longer, Discord’s gateway latency number climbs because heartbeats are processed late, and in extreme cases the gateway reconnects. Your panel’s CPU graph sitting flat at the limit confirms it.
What to do
Optimise the hot spot (cache results, avoid per-message work, move heavy processing elsewhere), or move to a plan with a larger CPU share — Basic gives 75% and Pro a full core.
RAM: a hard ceiling
RAM is different. 256 MB is the maximum memory your bot’s process may use, and it’s enforced strictly.
How you’ll notice
When the process exceeds the limit, the system stops it. The watchdog restarts it within seconds, so you’ll see the bot drop offline briefly and a fresh startup in the console — often with no error message from your code, because the process didn’t crash; it was killed. On the memory graph, repeated climbs followed by sudden drops (a sawtooth) are the signature of out-of-memory restarts.
What to do
- Keep steady-state usage under about 70% (~180 MB), leaving headroom for spikes.
- Trim library caches and unneeded intents — see reducing memory usage in discord.js bots and in Python bots.
- Fix leaks; more RAM only postpones them.
- Upgrade when your features genuinely need more. What fits in 256 MB has real numbers.
Storage: 2 GB of NVMe
Your server’s disk holds your code, dependencies, data files and anything your bot writes. It’s persistent — files survive restarts and redeploys — and it’s NVMe, so reads and writes are fast.
What uses storage
- Dependencies —
node_modulescan easily reach 100–300 MB; Python packages less, but data-science libraries are large. - Databases — SQLite files grow with your data.
- Logs — the one people forget. A bot that writes a log file without rotation can fill a disk in weeks.
- Media — downloaded images, audio and transcripts.
How you’ll notice
A full disk makes writes fail: SQLite reports “database or disk is full”, file saves throw errors, and package installs fail on deploy. Bots often crash in confusing ways, so check disk usage whenever errors mention writing.
What to do
Log to standard output instead of files (the panel console captures it), delete temporary files after use, keep only the data you need, and prune dependencies you don’t use. The free plan’s included MySQL database also keeps structured data off your server’s disk.
Network: a 10 Gbps port
The free plan sits on a 10 Gbps network. For bots, the network is almost never the bottleneck: Discord and Telegram traffic is small JSON messages, and even busy bots move modest amounts of data.
What matters more than raw speed is latency to the services your bot talks to. Free servers run in Virginia, close to Discord’s US East infrastructure, for WebSocket latency under 18 ms. See why Discord bot latency matters.
Network usage only becomes significant for bots that move large files — media downloaders, file converters, audio streaming — and those usually need more CPU and RAM than the free plan offers anyway.
Other things to know
- Database: the free plan includes a shared MySQL database. It’s ideal for bot data; for heavy workloads or longer backup retention, paid plans and standalone database plans give dedicated resources. See using the shared MySQL database.
- Backups: the free plan doesn’t include automatic backups. Export anything important yourself; paid plans back up automatically.
- SLA: the free plan is best-effort and not covered by SLA credits, which apply to paid plans.
- Activity: active servers stay forever; long-stopped, untouched servers can be reclaimed after a warning — see the reclaim policy.
Reading your panel graphs
The panel shows live CPU, memory and network graphs. A quick health check:
- CPU mostly low, with short spikes during commands → healthy.
- CPU flat at the limit → throttled; optimise or upgrade.
- Memory flat after warm-up, well below the line → healthy.
- Memory sawtooth hitting the line → out-of-memory restarts.
- Memory climbing steadily for days → a leak.
Summary
CPU is a 30% share of an EPYC core: exceed it and the bot slows down. RAM is a hard 256 MB ceiling: exceed it and the process is restarted. Storage is 2 GB of persistent NVMe: fill it and writes fail — usually because of logs or dependencies. The 10 Gbps network is rarely a constraint for bots. Watch the graphs, keep headroom, fix leaks and log to the console, and upgrade in place when your bot’s real needs outgrow the free plan.