← Back to all tipsOctober 14, 2026

Sizing your first index

Every Relatiq Search instance runs on a reserved share of compute and memory, set in whole slots (1 vCPU + 2 GB RAM each). Picking a starting size is less guesswork than it looks.

Start from RAM, not CPU

Vector search is RAM-bound before it's CPU-bound: the index needs to hold embeddings in memory to serve fast queries. A Starter slot (1 vCPU, 2 GB) comfortably covers a few thousand products or articles. If your catalogue is in the tens of thousands, start at Growth (2 vCPU, 4 GB) rather than waiting to hit a wall.

Query volume matters less than you'd think

CPU only becomes the bottleneck under sustained concurrent query load, which is rare for most catalogues and knowledge bases outside peak shopping events. Unless you know you'll have heavy concurrent traffic from day one, size for data volume first and upgrade CPU if latency climbs under load — it's a resize, not a migration.

Growing past one slot tier

Outgrowing your current tier isn't a wall either. Upgrades are prorated for the month you move, and the underlying index doesn't need to be rebuilt — the instance's reserved capacity simply increases. There's no tier ceiling: Scale and custom allocations beyond it follow the same per-slot math, just with more slots.

A rough starting point

  • A few thousand items, prototype or low traffic: Starter.
  • A production catalogue with steady traffic: Growth.
  • Tens of thousands of items or higher query concurrency: Scale, sized by your actual reserved vCPU/RAM.

If you're not sure, start one size down from what feels safe — resizing up later is the easy direction.