Epure
Self-host

What a 512 MB box can and cannot run

sentry self-hosted footprint

Documented RAM floor and container count decide what a 512 MB box can run. Self-hosted Sentry publishes 16 GB RAM plus 16 GB swap. Dated Epure idle receipts sit near 53 MiB on a VPS and 82 MiB on Desktop.

EpureUpdated 12 min read

On this pageShow
  1. Published floors, not screenshots
  2. Epure idle receipts you can re-run
  3. SQLite vs Postgres on a billed box
  4. License on a box you operate
  5. Official Sentry SDK, partial wire
  6. Same brief, five answers
  7. What breaks on a 512 MB brief
  8. Pick by the box you already pay for
  9. Limits of this ranking
  10. Questions

A small box fails on compose sprawl and published RAM floors before it fails on missing features. Read RAM floor and process count, then license and database. Replay and APM stay outside this note. If those are daily jobs, buy Sentry Cloud instead of running the official self-host on 512 MB.

Service-count narrative: Two-container error tracking. Method and named-box receipts: Sentry self-hosted RAM. Collected footprint: resource footprint. Decision cut: When error tracking is enough. Bias disclosed: first-party Epure docs and dated docker stats for Epure; peer floors from their public docs as of 2026-09.

Published floors, not screenshots

Self-hosted Sentry cites about 16 GB RAM, 16 GB swap, and 20+ containers for a typical full install (Sentry self-hosted). That number matches errors plus traces, replay, profiling, and brokers. Off this list on purpose.

Fair disclosure: from Sentry 24.8.0+, COMPOSE_PROFILES=errors-only keeps Issues and related error surfaces and drops traces, replays, and profiles (errors-only docs). It is still a large Compose fleet. It is not a 512 MB answer.

  • GlitchTip: 256 MB minimum, 512 MB recommended. About four containers (web, worker, Postgres, Valkey).
  • Bugsink: one container, SQLite default. No idle RAM published. Throughput numbers they show on 4 GB are not idle RAM.
  • Telebugs: 1 GB RAM and 1 CPU documented minimum. One Rails + SQLite container. $299 once.
  • Epure: two containers, Rust binary + PostgreSQL 16. No Redis. Dated idle receipts below.

Epure idle receipts you can re-run

Epure does not publish a vendor minimum. It publishes dated docker stats on classic Compose. Do not average rows.

  • 2026-09-23, 2 vCPU / 769 MiB Linux VPS (Alibaba Cloud): about 5 MiB RSS Epure + about 48 MiB RSS Postgres, about 53 MiB combined.
  • 2026-09-12, Docker Desktop idle: about 82 MiB combined for app + Postgres.
  • Small-VPS claims: cite the Alibaba row. Laptop demos: cite the Desktop row. Method and misreads: Sentry self-hosted RAM.

SQLite vs Postgres on a billed box

Bugsink defaults to a SQLite file and can attach Postgres. Telebugs is SQLite in the Rails image. GlitchTip and Epure default to Postgres. Pick the store you already back up. One container still beats two on a host that bills by RAM and by how many images you pull. Epure loses that row to Bugsink and Telebugs (2 vs 1) because Postgres is a separate process. That is the trade for a backup story many teams already run.

License on a box you operate

GlitchTip is MIT. Epure is Apache 2.0. Bugsink is Polyform Shield (source-available; self-host if you are not competing with it). Telebugs is a $299 paid binary. Sentry self-host is FSL. OSI counsel usually lands on GlitchTip or Epure. Shield is a product decision, not a footnote. Mesh: /vs/glitchtip, /vs/bugsink, /vs/telebugs, /vs/sentry.

Official Sentry SDK, partial wire

GlitchTip, Bugsink, Telebugs, and Epure take a DSN swap on official Sentry SDKs. That is envelope ingest for exceptions, not a clone of Sentry Cloud. Set traces, replay, and profile sample rates to 0 so you are not shipping volume nobody stores. More on the wire: Sentry SDK compatible.

Same brief, five answers

The dated idle receipts above stay in this essay. Collected footprint: resource footprint. Head-to-heads stay on /vs.

What breaks on a 512 MB brief

Stuffing Sentry onto the $4 box

You see: OOM kills during docker compose up, swap thrash, or finance rejecting a 16 GB instance. Cause: you treated Sentry's published floor as optional, or you assumed COMPOSE_PROFILES=errors-only equals two containers. Fix: stay on Sentry Cloud for the full suite, or pick a tracker whose published shape fits the invoice (Sentry self-hosted RAM).

Averaging Epure idle rows

You see: a slide that says “under 70 MiB” with no date or host. Cause: someone blended the 2026-09-23 VPS ~53 MiB idle with the 2026-09-12 Desktop ~82 MiB idle. Fix: cite one dated named box. Small-VPS claims use Alibaba. Laptop demos use Desktop.

Telebugs on a 512 MB brief

You see: the vendor floor is 1 GB RAM / 1 CPU while your brief is 512 MB. Cause: you ranked by one-container marketing without reading their minimum. Fix: give Telebugs a 1 GB box or pick GlitchTip / Bugsink / Epure for the tighter brief (Epure vs Telebugs).

Pick by the box you already pay for

Pick Epure if

  • Apache 2.0, clone and run, no Shield clause.
  • PostgreSQL 16 is already how you take backups.
  • You refuse Redis in the error stack.
  • You will re-measure idle RSS on your metal.

Pick another tool if

  • MIT plus Django (and a worker) is already home: GlitchTip.
  • One container and a SQLite file is the ops model: Bugsink.
  • The box has 1 GB and you want their installer: Telebugs ($299).
  • You still want Sentry's suite: buy Sentry Cloud, not a 16 GB self-host on a $4 VPS.

Limits of this ranking

  • No replay, APM, or profiling bake-off. Those belong on Sentry Cloud or a 16 GB self-host class.
  • Peer RAM rows are their published floors or absences, not Epure lab measurements of their stacks.
  • Epure idle is self-host Compose only. Not a PaaS SLA.
  • Partial SDK wire for exceptions. Not 100% Sentry protocol parity.

Compare stacks: Epure vs Sentry. Install Epure: self-host or one-click deploy. Cheap VPS commands: Cheap VPS install. Repo: GitHub.

FAQ

Questions

Can I self-host Sentry on 2GB?
No. Official Sentry self-host docs list about 16 GB RAM plus 16 GB swap as the minimum, with 20+ containers. A 2 GB VPS is below that floor. Use Sentry Cloud, or pick GlitchTip, Bugsink, or Epure. Row-by-row: Epure vs Sentry.
GlitchTip vs Bugsink RAM
GlitchTip publishes 256 MB minimum and 512 MB recommended on its install docs. Bugsink does not publish idle RAM; the vendor publishes throughput on a 4 GB box. Bugsink wins one container and SQLite-default. GlitchTip wins a written 512 MB-class floor and MIT. Details: Epure vs GlitchTip, Epure vs Bugsink.
Does Epure need Redis?
No. Epure compose is the Rust binary plus PostgreSQL 16. No Redis, no Kafka, no worker container. Install path: self-host.
Does Telebugs fit under 512 MB?
Not by their published floor. Telebugs documents 1 GB RAM and 1 CPU as the minimum, one Rails + SQLite container, $299 once. Compare: Epure vs Telebugs.
What is the best self-hosted error tracking on a small VPS?
Score RAM floor and containers, then license. GlitchTip if you want MIT Django at 256/512 MB. Bugsink if you want one SQLite container. Epure if you want Apache 2.0, Postgres default, and dated idle receipts (~53 MiB VPS / ~82 MiB Desktop). Self-hosted Sentry is the wrong class of machine.
Do these tools keep the official Sentry SDK?
GlitchTip, Bugsink, Telebugs, and Epure ingest via a DSN swap on official Sentry SDKs. Envelope coverage is a subset. Transactions, replay, and profiling are out of scope for this list.