Prod capacity under Young's new stated load (2026-10-17)
Prod capacity under Young's new stated load (2026-10-17)
Young, 2026-10-17: prod hosts 2 live websites + e-commerce tenants (Mercury V2 + POS + EVE) + Hermes agents + EVE. That is a heavier prod than my earlier numbers assumed, so I re-ran the capacity maths. All figures measured or from Tam's live-measured tenant sizing — nothing estimated.
The arithmetic
| Line | RAM |
|---|---|
| prod available | 21 GB (of 23 GB total; 1 GB used) |
| − 2 live websites (docs + pulse) | ~1.0 GB |
| − Hermes agent gateway on prod | 0.69 GB (LAB-measured: 3 agents in 686 MB) |
| remaining for tenants | 19.3 GB |
| ÷ 4.8 GB per tenant (Tam's measured stack) | = 4.0 tenants |
Without agents on prod it would be 4.2 tenants. Co-locating the agent fleet costs 0.7 GB = 14% of one tenant slot. That is the whole trade.
What this changes
- My earlier "~4–6 tenants" was optimistic for the new load. It was right for the old plan (agents elsewhere). With agents + websites + tenants on one node, plan for 4 tenants, not 6. Four is the honest number and it still beats the pilot's needs (2–3 tenants).
- The consolidation question is now answered by arithmetic, not opinion. Moving agents off prod buys back 14% of one tenant slot — while costing the portability proof, adding a single point of failure, and widening the E5 blast radius. The trade is bad. Keep agents portable and present.
- EVE on prod is not installed. No
evebinary, no eve dir. Young lists "Hermes Agents + EVE" as prod's load; today onlyv3-heart-prodruns there. That is a build gap (Sol/Ace), not a capacity problem — and it fits in the 0.7 GB envelope.
Disk — the real constraint to watch
| prod disk | 14 GB used, 179 GB free |
| per-tenant stack on disk | ~5–10 GB (images + volumes) |
| tenants at 179 GB | 18–35 tenants |
RAM binds before disk (4 tenants vs 18+). Worth noting because the business plan prices per tenant; RAM is what caps the pilot node.
The cost line that still matters
Node capacity is not the money. Always Free covers 2 OCPU / 12 GB; we run 4 OCPU / 24 GB, and four tenants alone need ~19 GB. So:
- The node is firmly in billable territory at pilot scale (not a free-tier ride).
- The Oracle burn check remains the one open infra number (Young's console).
- My
business-plan-v2two-row infra table still collapses to one row when it lands.
Honest limits
- Website RAM (~1 GB) is an estimate; I measured the agent and tenant figures, not the Nuxt processes' RSS on prod.
- "Mercury V2" is a name I have not seen in a measured doc — I have priced the Medusa stack Tam measured. If Mercury V2 is a different stack, its footprint needs its own measurement.
- EVE's prod footprint is unknown because EVE is not installed on prod yet.
"Move all agents onto a single host to save infrastructure cost" — the numbers
Short answer: consolidating agents saves ~A$0/month, because the agents are not what costs money. Measured this morning, every figure below probed, not estimated.
Saturday morning answers — three questions, measured
Young asked three things. Two are mine; the third (A2A design) is Sol/Min's, but I measured it so nobody re-probes.

