Unified package deep-dive — Tam lane (2026-10-13)

Young: "Package 3 products into single package as sell services." The three lanes verified today chain into one deployable unit per client. This note is the packaging layer Mira/Sol/Min asked for — inputs + build spec, open questions flagged for the 4 Oct-14 lanes.

The single package: "SolOS Commerce-in-a-Box"

One per-tenant Docker compose project = the whole product. Nothing is sold à la carte; POS and AI staff are the same stack, switched on by config:

• What the customer buys: managed webshop + in-store POS + AI support/sales agent + admin dashboard, per-tenant domain + TLS, own data (own PG volume), staging env available (second compose project).
• What SolOS sells differently from Medusa Cloud: Cloud sells the runtime; we bundle POS + AI staff + our ops automation — verified neither Medusa Cloud nor the Medusa ecosystem ships an out-of-box POS-with-payments or an agent layer. This is the defensible wedge.
• Why one package is right technically: POS and EVE agent both consume the same per-tenant Admin API keys; splitting them would multiply key-management surfaces. One stack = one secrets register per tenant (already the fleet hygiene pattern).

Packaging mechanics (input to Sol's architecture + Ace's build)

1. Feature toggles, not forks: POS on/off = Agilo app connects or not (no server change, needs HTTPS); AI staff on/off = EVE service + proxy route present or absent. One image set, per-tenant env flags.
2. The single deployment unit: docker compose -p <tenant> up with .env generated by onboarding script (secrets 0600). Health contract: :9000/health (Medusa) + /eve/v1/health (agent) both green = tenant live.
3. Tier ladder (RAM accounting from measured study, Oracle ARM ~20G usable):
   • Starter — Medusa + storefront + POS, no agent: 3 G
   • Pro — Starter + EVE agent (chat + order-status + HIL refunds): 4.8 G
   • Multi/Market — Pro + staging env + custom skills modules: 7.8 G
   • Node capacity: ~5 Starter / ~3–4 Pro / ~2 Market per 23Gi ARM node.
4. Update + rollback: template git tag → predeploy auto-migrate → restart → health re-gate; rollback = previous tag. Staging copy for client acceptance before prod cutover.
5. Data/backup: per-tenant pg_dump cron → offsite (gdrive lane already proven); retention mirrors vault-archive dated-dir semantics.

Commercial inputs for Mira (verified facts only)

• Hosting cost base: existing Oracle ARM free tier = node 1 (4–6 light tenants); second node needed at ~5 paid tenants → first hard infra cost step.
• Licence cleanliness: MIT + MIT + Apache-2.0 + FSL-free (no Paper dependency) — we may compose/modify/redistribute all parts; only sold-as-a-service terms of our own apply. No GPL contamination found in the chosen repos.
• Competitive anchors: Medusa Cloud (managed runtime only), Shopify (closed, monthly), Square POS ecosystem (POS only). Our wedge = open stack + AI staff + POS bundle with per-tenant data ownership (a pitch Shopify cannot make).
• RAM/tenant numbers above are the unit-economics denominator: gross margin per tenant ≈ price − (infra share + backup storage + ops time). Numbers needed from Mira: price point per tier.

Open questions (flagged for lanes)

• Min + Tam deep dive (in progress): Which 3 packages exactly get bundled? Proposal A = classic/starter×3 per table above. Alternative reads: (i) three client segments (retail shop / services business / hybrid), (ii) three products bundled 1:1. Need Young's read; my default = tier ladder above.
• Mira: pricing per tier + target client count for break-even on node 2 (~5 tenants).
• Sol: tenant-side EVE agent key scoping; whether POS payments wait for Agilo terminal roadmap or ship manual-payment-only at launch.
• Services-selling module: custom Medusa module for bookable services — needs one spike (feasible, unscoped).

Sources

• All component facts: medusa-hosting-docker-study (sizing measured), medusa-eve-integration-study (integration verified), hermes-eve-cohosting-study (RAM budget), medusa-nextjs-commerce-study (POS MVP gaps, licences). No new claims introduced in this note beyond packaging arithmetic on those measured numbers.