2 GB RAM class
A 2 GB VPS covers the full stack for most sites; 2 vCPU / 4 GB is comfortable at roughly 60M events a month. ClickHouse does the heavy lifting, the Node processes barely notice.
Requirements
A €6 VPS class
runs the whole stack.
One small box, four Node 22 processes, one ClickHouse. No container registry to run, no orchestrator to babysit, no npm packages in the serving path. We size the exact machine with your DevOps during onboarding — this page is the envelope.
What runs, and what it demands
Four processes and a database.
| Component | Runtime | Port | What it demands from your box |
|---|---|---|---|
| Collector | Node 22, zero dependencies | :8090 | the only internet-facing listener — origin allowlist, rate limits, payload caps, disk-spool replay |
| API + dashboard | Node 22, zero dependencies | :8081 | PBKDF2 auth, RBAC, report cache, k-anonymity post-filter — internal to your network |
| ClickHouse | 24.8 | :8124 | localhost by default — TDigest percentiles, materialized rollups, TTL-enforced retention |
| MCP server | Node 22, zero dependencies | stdio | not a network service at all — 15 tools for AI agents, gated by scoped keys |
Default listeners: collector :8090, API/dashboard :8081, ClickHouse :8124 — ClickHouse bound to localhost unless you decide otherwise with your DevOps.
The envelope
Small on purpose, flat by design.
Zero npm packages
Collector, API, dashboard, SDK and MCP server run with no dependencies in the serving path. Nothing transitive to audit, nothing to patch at 3 a.m. because a left-pad broke.
ClickHouse 24.8
The one non-Node component, chosen for what it enforces: TDigest percentiles you can trust, materialized rollups for the 25-month window, and retention via TTLs in the schema.
Disk flat by design
Raw events live 24 hours, rollups 25 months, recordings 30 days — retention is enforced by TTLs in the schema, so disk use stays flat instead of growing with traffic.
Compose or systemd
Docker Compose for the turnkey path; bare Node 22 processes under systemd when your DevOps prefers no containers. No orchestrator, no service mesh, no Helm chart to maintain.
Sized together
You do not pick a box from a doc page. We look at your traffic, retention wishes and budget with your DevOps during onboarding — the €6 class is the floor, not a requirement.
The negative list
What you do not need.
No orchestrator
Kubernetes is not part of the deployment. Compose or systemd, decided with your DevOps — the stack was built to be operated by a small team.
No dependency audit
The serving path carries zero npm packages. Your supply-chain review shrinks to processes you can read.
No capacity planning ritual
TTL-enforced retention means disk stays flat; rollups carry the 25-month history. Growth shows up as CPU, not as a storage emergency.
One small box.
The whole stack.
Bring a €6-class VPS and your DevOps engineer — we size it together, deploy it together, and the marginal event stays free.